TWI815411B - Methods and non-transitory computer storage media of extracting linguistic patterns and summarizing pathology report - Google Patents

Abstract

Disclosed are methods and the non-transitory computer storage media of extracting linguistic patterns and summarizing a pathology report thereof. The present disclosure provides a method of extracting key linguistic patterns from a pathology report. The method comprises: determining a confidence degree and a support degree between a linguistic term and a next linguistic term based on co-occurrences of the linguistic term and the next linguistic term; generating a set of candidate linguistic terms; generating a first set of linguistic patterns through performing random walks on the set of candidate linguistic terms; and determining the key linguistic patterns through removing redundant linguistic patterns from the first set of linguistic patterns.

Description

Methods for extracting semantic patterns and summarizing pathology reports and non-transitory computer storage media

The present invention relates to a method of processing a pathology report. Specifically, the present invention relates to a method of extracting key semantic patterns from a pathology report, a method of summarizing a pathology report, and non-transitory computer storage media thereof. The invention further relates to a method of determining similarity between pathology reports.

A patient's pathology report contains a large amount of information, especially for cancer patients, and this pathology report contains a large amount of complicated and boring information. Attending surgeons and physicians can spend a lot of time understanding a patient's condition, but computers can help reduce the amount of time wasted and therefore increase overall efficiency.

The present invention can analyze a pathology report. A pathology report may contain a diagnosis determined by examining cells and tissue under a microscope. This report may be for a lung cancer patient. Important information can be summarized from a complex and boring pathology report. This information can include features in six categories: basic pathological description, tumor characteristics, histological description, immunohistochemistry (IHC) information, a genetic test result, and a pathological TNM (tumor, lymph node and metastasis) stage. The present invention can further summarize multiple pathology reports of a patient. The present invention can further provide a function of searching among a large number of patient data, and the search results can be used for reference by surgeons and physicians.

One embodiment of the present invention provides a method for extracting key semantic patterns from a pathology report. The method includes: determining a degree of trust and a degree of support between a semantic term and the next semantic term based on the co-occurrence of a semantic term and a next semantic term; generating a group of candidate semantic terms; The semantic terms perform a random walk to generate a first set of semantic patterns; and the key semantic patterns are determined by removing redundant semantic patterns from the first set of semantic patterns. In the pathology report, the semantic term appears before the next semantic term. The reliability between a candidate semantic term and a corresponding next candidate semantic term is equal to or greater than a trust threshold value. The support between the candidate semantic term and the corresponding next candidate semantic term is equal to or greater than a support threshold value.

Another embodiment of the present invention provides a method of summarizing a pathology report. The method includes obtaining a plurality of pathological features from the pathology report based on key semantic patterns. The key semantic patterns are generated according to any one of the methods or operations of the present invention.

A further embodiment of the present invention provides a non-transitory computer storage medium. The non-transitory computer storage medium has program instructions stored thereon. The program instructions, when executed by a processor, cause the performance of a set of operations according to any of the methods of the invention.

The following disclosure provides many different embodiments or examples for implementing different features of the provided subject matter. Specific examples of operations, components, and configurations are described below to simplify the present invention. Of course, these are examples only and are not intended to be limiting. For example, the description of a first operation being performed before or after a second operation may include embodiments in which the first operation and the second operation are performed together, and may also include embodiments in which the first operation and the second operation may be performed together. Examples of performing additional operations in between. For example, in the following description, forming a first feature over, on, or in a second feature may include embodiments in which the first feature and the second feature are formed in direct contact, and may also include embodiments in which the first feature may be in direct contact with the second feature. Embodiments in which additional features are formed between the features and the second features so that the first features and the second features may not be in direct contact. Additionally, the present invention may repeat reference numbers and/or letters in various instances. This repetition is for simplicity and clarity and does not by itself indicate a relationship between the various embodiments and/or configurations discussed.

For ease of description, time-relative terms such as “before,” “before,” “after,” “after,” and the like may be used herein to describe an operation or feature. The relationship with another operation(s) or characteristics, as shown in the diagram. Time-relative terms are intended to cover the different sequences of operations depicted in the figures. In addition, for ease of description, spatially relative terms such as “below,” “below,” “lower,” “above,” “upper,” and the like may be used herein to describe a The relationship of an element or feature to another element or feature(s), as illustrated in the drawings. Spatially relative terms are intended to cover different orientations of the device in use or operation in addition to the orientation depicted in the figures. The device may be otherwise oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly. For ease of description, connection-relative terms such as "connected," "connected," "connected," "coupled," "coupled," "in communication," and the like may be used herein to describe two elements or features. An operational connection, coupling or connection between. Connection relative terms are intended to cover various connections, couplings or connections of devices or components. Devices or components may be connected, coupled or connected to each other directly or indirectly, such as through another set of components. Devices or components may be wired and/or wirelessly connected, coupled, or connected to each other.

As used herein, the singular terms "a," "an," and "the" may include plural references unless the context clearly dictates otherwise. For example, a reference to a device may include multiple devices, unless the context clearly dictates otherwise. The terms "include" and "include" may indicate the presence of described features, integers, steps, operations, elements, and/or components, but may not exclude one or more of the features, integers, steps, operations, elements, and/or components. The existence of the combination of those. The term "and/or" may include any and all combinations of one or more of the listed items.

Additionally, quantities, ratios, and other numerical values are sometimes presented herein in a range format. It should be understood that this range format is used for convenience and brevity, and should be flexibly understood to include values expressly designated as the limits of a range, but also to include all individual values or subranges encompassed within that range, as if each Values and subranges are explicitly specified.

The nature and uses of the embodiments are discussed in detail below. It should be appreciated, however, that the present invention provides many applicable inventive concepts that can be embodied in a variety of specific semantic contexts. The specific embodiments discussed are merely illustrative of specific ways of making and using the invention and do not limit its scope.

In order to summarize important pathological features from a pathology report (for example, a pathology report of lung cancer), the present invention provides a method named PR2Sum (Pathology Report Summary). In some embodiments of the present invention, 50 important pathological features among six categories are selected, filtered or defined from a pathology report. Fifty exemplary pathological features are listed in Table 1. The six categories may include: basic description, findings (of (several) tumors), histology (information of (several) tumors), IHC information, genetic testing (results), and TNM stage. In a further embodiment, the reported data of a patient(s) may be represented by the 50 pathological features shown in Table 1. Category Pathological features Basic description Organs, Bx Sites, Sampling Methods, Diagnosis discover Maximum size, tumor size, proximal margin, lymphovascular invasion, VPI, tumor focus Histology Histology type, histology class IHC CK7, TTF-1, aspartic proteinase A, CK20, P40, CDX2, P63, P16, cytokeratin (AE1/AE3), vimentin, PAX-8, CD56, chromogranin A, synaptophysin, GATA3, P53, S100, Ki67, EBER genetic testing EGFR, ALK, ROS1, BRAF, MET, KRAS, ERBB2, PIK3CA, NRAS, MEK1, NTRK, RET, PDL1 TNM period Version, pT, pN, pM, pStage, N information Table 1

In the case of using machine learning algorithms (or deep learning algorithms), various feature combinations must be constantly tried during model training to increase performance. The process of generating various combinations of features may be called feature engineering. Feature engineering for machine learning algorithms is expensive. For example, feature engineering for machine learning algorithms will take a lot of time to generate various feature combinations and detect these combinations. However, the combination of features obtained from this high-cost model may only be applicable to the current research area. For example, a completely new set of feature combinations should be generated for a pathology report associated with another disease. Once the research domain or disease changes, feature engineering should be restarted for a new set of feature combinations, and the model should be trained again using this new and different set of feature combinations. It is difficult for the constructed or trained machine learning model to achieve the effect of knowledge sharing. In addition, known machine learning algorithms for similar problems are not interpretable because they only generate many uninterpretable parameters and probabilities.

Human knowledge can be accumulated. In human thinking, the situation/situation of an article or a problem stated in a language can be narrowed down through some important semantic terms. For example, when a clinician interprets a pathology report, if the terms "immunohistochemistry," "immune response," and "aspartic acid protease A" appear in the report at the same time, he will automatically assume that the corresponding sentences are related to pathology. The immunohistochemical staining reactions of aspartate proteinase A are related because the three terms have a strong correlation. That is, humans can read an article by skimming. If the terminology in a pathology report satisfies a clinician's knowledge structure, the clinician can understand the pathological characteristics described in a sentence or paragraph.

Obtain human perception of a topic by identifying important entities or related content to filter out possible candidates. For example, when highly related words such as "immunohistochemistry" and "aspartic acid protease A" appear together in one sentence, it is natural to conclude that this is more likely to describe the patient's response to aspartic protease A. A statement of the reaction of immunohistochemical staining. The invention may be similar to what humans do when skimming a pathology report to capture its main ideas. Furthermore, knowledge gained from different topics can be accumulated and used to identify other new topics.

In view of this, the method of the present invention imitates the perceptual behavior of human understanding, which is a highly automated method of learning semantic patterns that characterize the lung cancer field from the original text in the pathology report. One of the main advantages of the present invention is its high accuracy and knowledge accumulation capabilities. When facing a new field, knowledge can be further expanded to accommodate unknown information by adding new rules.

Different from machine learning algorithms, the present invention provides a novel method for natural language understanding. Regarding the acquisition of pathological features, the present invention simulates the behavior of a clinician when reading a pathology report (for example, a pathology report of a lung cancer patient). The present invention can quickly narrow down the situation/situation of an article or a problem by understanding the key points or semantic terms. Reasons include strong correlations between the gist of the article (or question) and adjacent terms. Therefore, the gist or question can be identified naturally. Therefore, the method or algorithm provided in the present invention may be interpretable.

In addition to capturing important features of a pathology report, the present invention also focuses on flexible comparisons that cannot be performed by rigid formal expressions. Therefore, the present invention has a higher degree of freedom during pathological semantic pattern matching.

The present invention provides a pathological semantic pattern generation algorithm. Pathological semantic pattern generation algorithm can be used for lung cancer patients. In some embodiments, the pathological semantic pattern generation algorithm can be used for patients with various cancers, including but limited to prostate cancer, colorectal cancer, gastric cancer, breast cancer, colorectal cancer, and cervical cancer. In some embodiments of the present invention, a semantic pattern for identifying pathological features of lung cancer can be generated based on pathology reports of 500 lung cancer patients.

In the present invention, the process of generating pathological semantic patterns of lung cancer can be regarded as a problem of frequent pattern exploration. A pathological semantic association map of lung cancer can be constructed based on the co-occurrence of terms in pathology reports. The pathological semantic association map can describe the strong semantic correlation between different terms.

Figure 1 illustrates a pathological semantic association diagram 100 according to some embodiments of the present invention. The pathology semantic association graph 100 may be constructed based on the co-occurrence of terms in one or more pathology reports of one or more lung cancer patients. In some embodiments, the pathology semantic association graph 100 is constructed based on terms appearing in multiple pathology reports for multiple patients. After calculating the frequency of occurrence of each term appearing in the pathology report and the number of co-occurrences between different terms appearing in the pathology report, the pathology semantic association graph 100 can be constructed.

Since the pathological semantic pattern to be generated (for example, for lung cancer) can be a sequential directed graph, the present invention constructs a semantic association graph with association rules.

Each vertex in Figure 1 may indicate a different term. For example, the terms (or phrases) S ₁ and S ₂ are expressed as different vertices. Each edge in Figure 1 is constructed based on co-occurring terms (or phrases). For example, the co-occurrence C ₁₂ of terms S ₁ and S ₂ is expressed as an edge between corresponding vertices, where term S ₁ occurs before term S ₂ . In other words, co-occurrence C ₁₂ indicates that the co-occurrence of term S ₁ occurs before the next term S ₂ . The value of co-occurrence C _ij indicates the degree _{of trust in the terms Si and S j .} The value of co-occurrence C _ij indicates a conditional probability, which is a measure of the probability of occurrence of term S _j (assuming that S _i has occurred). In some embodiments, the term S ₁ that co-occurs with the term S ₂ may be a generic term of the term S ₂ . The term S ₁ that co-occurs with the term S ₂ may be a dependent term of the term S ₂ . The degree of confidence in terms S _i and S _j is defined as equation (1). Equation (1)

Support can be defined as "the number of samples of true responses belonging to that category." In the present invention, the support of the term _Si can indicate the frequency of occurrence of the term _Si . In some embodiments, the support of term _Si may indicate the number of occurrences of term _Si . The support of term _Si and the next term S _j may indicate the co-occurrence frequency of term _Si and the next term S _j . The support of the term _Si and the next term S _J may indicate the number of co-occurrences of the term _Si and the next term S _j . The relationship between a trust degree and the corresponding support degree can be defined as equation (2). Equation (2)

In order to make the generated semantic patterns distinctive in terms of pathological characteristics, terms with higher frequencies are retained in some embodiments of the present invention. Based on the co-occurrence times of retained frequent terms, a pathological semantic association graph of lung cancer can be constructed. In some embodiments of the present invention, the minimum support is set to 10, and the minimum reliability is set to 0.3, in order to avoid generating too short semantic patterns.

After constructing the pathological semantic association map of lung cancer, terms with higher frequency and better distinction are strung together based on random walk. A semantic pattern of pathological characteristics can be generated based on terms strung together.

One pathological semantic association map of lung cancer can be defined as ,in . V refers to the vertex set (for example, the vertex set of the terms S ₁ to S ₆ shown in Figure 1), E refers to the edge set (for example, the edge set of the co-occurring C _ij shown in Figure 1), and p refers to the set V in The number of elements, and k indicates the number of elements in the set E. A random walk procedure can be performed on graph G. A random walk program can consist of a sequence of random paths. The value of C _ij indicates _{the probability of moving from term Si to term S j .} For a term S _n , the set of all adjacent terms can be expressed as N(S _n ). For a term _Sn , the sum of the probabilities of each term moving from the term _Sn to the set N(S _n ) satisfies equation (3). Equation (3)

By applying a random walk procedure to the pathological semantic association graph G, the probability matrix Pr obtained conforms to equation (4). _Xk indicates the kth step of the random walk procedure. Therefore, a series of randomly generated vertices is a Markov chain. Equation (4)

According to the research results of Dr. L Lovász, the coverage time (CT) on a graph can be expressed as equation (5). Equation (5)

Therefore, by applying the random walk theory to the pathological semantic association graph of lung cancer, the present invention can search for possible semantic frequent patterns. In addition to avoiding the loss of combinations with lower probability, the present invention can also generate semantic patterns of different pathological characteristics.

The present invention describes the correlation between frequent terms through a pathological semantic correlation diagram of lung cancer. Then, by applying the random walk procedure to the pathological semantic association graph, terms frequently appearing in pathology reports are strung together and semantic patterns of pathological characteristics are generated.

However, some redundant semantic patterns may be generated through random walk theory, and some further integration may be necessary. In some embodiments, the present invention removes a semantic pattern that is completely included in another semantic pattern, and thereby attempts to retain a semantic pattern (or semantic framework) with longer length and better coverage. For example, the semantic patterns (or semantic frames) of "S ₁ àS ₂ àS ₃ " and "S ₁ àS ₄ àS ₂ àS ₃ àS ₅ àS ₆ " are generated, and the previous pattern will be retained because the previous pattern contains The latter mode. In other words, if a second semantic pattern is a subset of a first semantic pattern, the second semantic pattern will be removed and the first semantic pattern will be retained. Furthermore, if a first semantic pattern dominates a second semantic pattern, the second semantic pattern will be removed and the first semantic pattern will be retained. After removing a specific semantic pattern, each of the remaining semantic patterns may be a set of unordered terms/phrases (S _i ) or a set of ordered terms/phrases (S _i ).

Figure 2 is a flowchart of a method 200 according to some embodiments of the present invention. Method 200 may extract key semantic patterns from one or more pathology reports of one or more patients. Method 200 includes operation 201 . In operation 201, a degree of trust and a degree of support between a semantic term and a next semantic term may be determined based on the co-occurrence of a semantic term and a next semantic term. This semantic term may appear before the next semantic term in one or more pathology reports.

Method 200 includes operation 203. In operation 203, a set of candidate semantic terms (or phrases) may be generated or selected from one or more pathology reports. In the group of candidate semantic terms, the reliability value between one candidate semantic term and the corresponding next candidate semantic term is equal to or greater than a trust threshold value. In the group of candidate semantic terms, the support value between one candidate semantic term and the corresponding next candidate semantic term is equal to or greater than a support threshold value.

Method 200 includes operation 205. In operation 205, a first set of semantic patterns may be generated or selected from the set of candidate semantic terms. The first set of semantic patterns may be generated or selected by performing a random walk on the set of candidate semantic terms.

Method 200 includes operation 207. In operation 207, key semantic patterns may be generated or selected from the first set of semantic patterns. Key semantic patterns may be generated or selected by removing redundant semantic patterns from the first set of semantic patterns. Each of the key semantic patterns may be an unordered set of terms/phrases. Each of the key semantic patterns may be a set of ordered terms/phrases.

Method 200 includes operation 209. In operation 209, a plurality of pathology features may be obtained from one or more pathology reports based on key semantic patterns. The plurality of pathological characteristics may include the exemplary 50 pathological characteristics listed in Table 1. The acquired, extracted or summarized pathological features forming a pathology can allow clinicians to quickly understand the situation or condition of the corresponding patient.

In some embodiments of the present invention, the support between a semantic term and the next semantic term can be generated or calculated based on the number of co-occurrences of a semantic term and the next semantic term. When a semantic term appears, a reliability value between the semantic term and the next semantic term may be generated or calculated based on the occurrence probability of one of the next semantic terms.

In some embodiments of the present invention, the trust threshold may be set to 0.3, and the support threshold may be set to 10. The trust threshold can be selected in the range of 0.2 to 0.5. Support thresholds can be selected in the range 7 to 12.

In some embodiments of the present invention, when a second semantic pattern is a subset of a first semantic pattern, the first semantic pattern is removed. That is, if a first semantic pattern includes a second semantic pattern, the first semantic pattern is removed.

In some embodiments of the present invention, the reliability between a semantic term and the next semantic term is independent of a second reliability value between a previous semantic term and the semantic term. In pathology reports, the previous semantic term appears before this semantic term, and this semantic term appears before the next semantic term.

In some embodiments of the present invention, 50 semantic patterns of important pathological features of lung cancer are generated through a pathological semantic pattern generation algorithm. After verification by clinicians, the method for obtaining six categories of pathological characteristics is described below, including: basic description, discovery of (several) tumors, histological description of (several) tumors, IHC information, genetic test results, and TNM stage.

A pathology report contains information classified in the basic description. This basic description can be provided in the SOAP (Subjective, Objective, Assessment, and Plan) section. "Subjective" can refer to subjective data, including the main complaints, symptoms, medical history, drug allergy history, adverse drug reaction history, and medication history expressed by the patient. "Objective" can refer to objective data, including vital signs, physical examination results, laboratory test results, and patient medical imaging results. "Assessment" may include impression/diagnosis, patient's condition, disease status, and analysis and evaluation of treatment. A "plan" may include diagnostic methods (laboratory tests), treatment methods (medications, procedures, surgeries, etc.), and health care education methods.

The content of a SOAP section can be separated or divided into multiple parts by commas (i.e. "," (in Chinese, use commas ",")). In some embodiments, when the number of comma-separated parts in the SOAP section is less than 4, the content of the SOAP section may be the detection result. For lung cancer patients, when the number of parts separated by commas in the SOAP section is greater than or equal to 4, the content of the SOAP section can be a description of the lungs.

Table 2 shows four exemplary SOAP sections. For each of cases 1, 2, and 4 in Table 2, the number of parts is judged to be less than 4. For each of cases 1, 2, and 4 in Table 2, the content of the SOAP section is determined to be relevant to the detection results. For case 3 in Table 2, the number of parts is judged to be greater than 4. Case SOAP section 1 The ALK expression is negative. 2 No ROS1 gene rearrangements were detected. 3 Lung, lower lobe, right side, bronchoscopy biopsy, adenocarcinoma 4 EGFR exon 18 mutations: no mutations detected EGFR exon 19 mutations: no mutations detected EGFR exon 20 mutations: exon 20 insertion (EX20Ins) EGFR exon 21 mutations: no mutations detected mutation Table 2

In some embodiments, when the number of comma-separated parts in the SOAP section is equal to 4, the four parts will be related to "organ", "location", "sampling method" and "diagnosis" in order. That is, the first part will be related to one or more organs, the second part will be related to one or more locations, the third part will be related to one or more sampling methods, and the fourth part will be related to one or more diagnoses.

In some embodiments, when the number of parts separated by commas in the SOAP section is greater than 4, some parts may be merged. For example, the semantic patterns generated for the basic description indicate some key semantic patterns for the part related to "organ" and some key semantic patterns for the part related to "sampling method". One or more parts related to the "organ" can be located or identified through some key semantic patterns. One or more parts related to the "sampling method" can be located or identified through some key semantic patterns. For lung cancer patients, exemplary key semantic patterns for locating parts related to "organ" and "sampling method" are listed in Table 3. After locating one or more parts related to "organ" and "sampling method", one or more parts related to "location" can be positioned between the parts related to "organ" and "sampling method". For example, if the first and fifth parts are related to "organ" and "sampling method" respectively, then the second to fourth parts will be judged to be related to "location". One or more sections following the section related to "Sampling Methods" will be judged to be related to "Diagnosis". For example, if the SOAP section is divided into seven parts, and if the fifth part is related to "Sampling Method", then the sixth and seventh parts will be judged to be related to "Diagnosis". In some preferred embodiments, only a portion following the portion related to "Sampling Method" will be determined to be related to "Diagnosis". part Key semantic patterns organ Lungs, lymph nodes, brain, skin Sampling method Biopsy, VATS, EBUS Table 3

For Case 3 in Table 2, the content of the SOAP section is "Lung, lower lobe, right side, bronchoscopy biopsy, adenocarcinoma." For Case 3, the number of sections is judged to be greater than 4. Based on the key semantic patterns in Table 3, the "lung" part in Case 3 will be judged to be related to "organ", and the "Bronchoscopic Biopsy" part in Case 3 will be judged to be related to "Sampling Method". In case 3, parts of "lower lobe" and "right side" will be judged to be related to "location" because they are located between parts related to "organ" and "sampling method". In case 3, the part "adenocarcinoma" will be judged to be related to "diagnosis" because it follows the part related to "sampling method". For Case 3 in Table 2, the parts related to "organ", "location", "sampling method" and "diagnosis" are marked in gray.

Figure 3 illustrates a section of a pathology report according to some embodiments of the present invention. Figure 3 shows the SOAP section of a pathology report. In the case of Figure 3, the SOAP section contains 6 statements. The number of parts separated by commas of each statement in Figure 3 will be judged to be greater than 4. According to the present invention, these six statements with more than 4 parts will be selected as candidates for further processing. For each statement in Figure 3, which part(s) are related to "organ" will be determined based on one or more key semantic patterns of the parts related to "organ" listed in Table 3. For each statement in Figure 3, which part(s) are related to the "sampling method" will be determined based on one or more key semantic patterns of the parts related to "organs" listed in Table 3. After determining the part related to "organ" and "sampling method", the part between the parts related to "organ" and "sampling method" will be judged to be related to "location". A section following the section related to "Sampling Methods" will be judged to be related to "Diagnosis". For the case in Figure 3, the basic description of the output will be "Organ: Lung/Location: Upper lobe, left side/Sampling method: VATS lobectomy/Diagnosis: adenocarcinoma." In the case of Figure 3, with "Organ", Sections related to "Location", "Sampling Method" and "Diagnosis" are marked in gray.

One or more key semantic patterns of the "organ"-related parts listed in Table 3 may be selected or generated for a lung cancer patient. One or more of the key semantic patterns listed in Table 3 for parts related to "organ" may include terms other than "lungs." The reason is that metastasis or invasion of lung cancer can occur in other organs. Therefore, one or more of the key semantic patterns of parts related to "organ" listed in Table 3 may include terms such as "lymph node", "brain" and "skin".

In some embodiments of the present invention, the content in the sections related to "location" and "diagnosis" can be further standardized for output. For example, content in the section(s) related to "location" may include various types of descriptions. The content in the part(s) related to "location" may be further standardized as at least one of "upper left lobe", "lower left lobe", "upper right lobe", "middle right lobe" or "lower right lobe". If the tumor invades other organs, the content in the section(s) related to "location" can be further standardized to at least one of "pleura", "bone", "brain" or "skin". The content in the section(s) related to "Diagnosis" may also include various types of descriptions. The content in the section(s) related to "Diagnosis" may be further standardized as at least one of "adenocarcinoma", "non-small cell carcinoma", "squamous cell carcinoma" or "large cell carcinoma".

A pathology report contains information classified among the findings of tumor(s). According to some embodiments of the invention, the semantic pattern generated for the discovery of tumor(s) is associated with volume. The semantic pattern of the discovery of (several) tumors may be a semantic pattern of volume. The semantic pattern of (several) tumor findings may include multiple numbers, multiple multiplication symbols, and one unit. For example, the discovered semantic pattern may include three numbers, two multiplication symbols, and one unit. The first multiplication symbol may be between the first number and the second number. The second multiplication symbol may be between the second number and the third number. The unit can be placed after the third digit. The unit can be a unit of length (for example, "cm" or "mm") or a unit of volume.

According to some embodiments of the invention, one or more candidate fragments may be selected based on semantic patterns generated from findings for tumor(s). If the semantic context of one or more candidate fragments contains one or more terms of the key semantic patterns listed in Table 4, then the associated one or more volume values will be determined as information about tumor size. In addition, the maximum value of the values indicating a volume will be judged as the value of the maximum size. For example, "50 mm × 35 mm × 20 mm" contains three values indicating the volume, and the maximum value "50" is determined as the value of the largest size. Key semantic patterns of tumor size Cutting, hard tumors, tumor measurement Table 4

In addition to the findings of the tumor(s), a pathology report also contains information classified as histological information of the tumor(s). Information on the discovery of (several) tumors and histological information on (several) tumors can be observed through a microscope in a laboratory. Information observed through a microscope can be described in different items of a pathology report. Items associated with microscopy information may be described in the "Microscopic Assessment" section of a pathology report.

In some embodiments of the invention, a microscope assessment section may be located or identified based on the term "microscope assessment." Information about one of the items in the microscope evaluation section can be described after a colon (i.e. ":"). In some embodiments of the present invention, the information described in each item can be obtained by locating the corresponding colon (ie, ":"). For example, after locating the microscope evaluation section, a given item can be located by searching for the corresponding item name, then the corresponding colon of the given item is located, and the information after the colon located in the given item is obtained as the item. information. The microscopic evaluation section may include at least one item of "tumor focus", "histological type", "histological classification", "lymphovascular invasion", "visceral pleural invasion" and "nearest edge". The items of "tumor focus", "lymphovascular invasion", "visceral pleural invasion", and "nearest edge" can be classified into (several) tumor findings. The items of "histological type" and "histological category" can be classified into the histological information of (several) tumors.

A pathology report contains information classified as IHC information. According to some embodiments of the present invention, the terms listed in the semantic pattern indication table 5 for IHC information generation (which may come from one or more key semantic patterns) may be regarded as one of the initial terms of the IHC section. The key semantic pattern of one of the initial terms in the IHC section Immunohistochemistry, immunohistochemistry, immunohistochemistry, immunohistochemistry, immunoresearch, IHC, immunostaining, immunohistochemistry, immunoresearch, immune response, demonstrating adenocarcinoma composition Table 5

After locating or identifying the IHC segment from the initial term, further information on the IHC segment can be obtained. The target item in the IHC section may be located or identified based on one or more key semantic patterns listed in Table 6.

After locating or identifying the target items in the IHC section, for each target item, a first modifier may be located or identified before the target item, and a second modifier may be located or identified after the target item. For each target item, a first distance between the first modifier and the target item and a second distance between the second modifier and the target item can be calculated. For each target item, if the first distance is smaller than the second distance, the first modifier will be judged as the modifier of the target item; if the second distance is smaller than the first distance, the second modifier will be judged as the modifier of the target item. Modifier. The first modifier and the second modifier can be "masculine" or "feminine". When the first modifier and the second modifier are the same, there is no need to select or determine which modifier to use to modify the target item. Therefore, calculation of the first distance and the second distance may be unnecessary. Furthermore, the comparison between the first distance and the second distance may not be necessary. Key semantic patterns of target items in the IHC section CK7, TTF-1, aspartic proteinase A, CK20, P40, CDX2, P63, P16, cytokeratin (AE1/AE3), vimentin, PAX-8, CD56, chromogranin A, synaptophysin, GATA3, P53, S100, Ki67, EBER Table 6

A pathology report contains information classified in genetic testing. Genetic testing information may include testing for immune checkpoint inhibitors (eg, PDL1 inhibitors), genetic testing for epidermal growth factor receptor (EGFR), and other genetic molecular testing.

In some embodiments of the present invention, the semantic patterns generated for the detection of immune checkpoint inhibitors indicate some terms related to the PDL1 detection kit. One or more exemplary key semantic patterns related to the PDL1 detection kit are listed in Table 7. Key semantic patterns in PDL1 detection 22C3, 28-8, SP142, SP263 Table 7

Search the entire pathology report based on one or more terms of one or more key semantic patterns listed in Table 7. Based on these searches, it can be determined whether PDL1 detection has been performed. For example, if one or more terms of the key semantic patterns listed in Table 7 are searched on the entire pathology report, it can be determined that PDL1 detection has been performed. If a PDL1 test is executed, information on items associated with the PDL1 test will be further obtained. Items related to PDL1 detection include: tumor proportion score (TPS), comprehensive positive score (CPS), tumor cells (TC), and immune cells (IC).

The key semantic patterns listed in Table 7 can be used to locate or identify the PDL1 detection part. Items associated with PDL1 testing may be provided in the PDL1 testing section. Information about an item associated with the PDL1 test can be described after a colon (i.e. ":"). In some embodiments of the present invention, the information described in each item can be obtained by locating the corresponding colon (ie, ":"). For example, after locating the PDL1 detection part, a given item can be located by searching the item name; then the corresponding colon of the given item can be located, and the information after the colon located in the given item will be obtained as the information of the item .

In some embodiments of the invention, the semantic pattern generated for the detection of EGFR indicates one or more terms, for example, the term "EGFR". The entire pathology report can be searched based on the term "EGFR". It can be determined based on these searches whether EGFR testing has been performed. For example, if the term "EGFR" or other similar terms is searched on the entire pathology report, it can be determined that an EGFR test has been performed. If EGFR testing is performed, further information on mutations in exon 18, exon 19, exon 20, and exon 21 of EGFR will be obtained.

The term "EGFR" can be used to locate or identify the detection portion of EGFR. Information about EGFR exon 18, exon 19, exon 20, and exon 21 mutations can be provided in the EGFR testing section. In some embodiments, whether a mutation is in exon 18, exon 19, exon 20, or exon 21 can be determined based on a search for the terms "18", "19", "20", and "21" middle. For example, if the term "18" is searched in the EGFR detection section, a mutation will be determined in exon 18. In some other embodiments, a determination of whether a mutation is located at position 790 of exon 20 can be based on a search for the term "T790M." For example, if the term "T790M" is searched in the EGFR detection section, a mutation will be determined to be located at position 790 of exon 20.

In some embodiments of the present invention, the semantic patterns generated for other genetic molecular detection indicate some of the key semantic patterns listed in Table 8. The entire pathology report can be searched based on the terms of the key semantic patterns listed in Table 8. These searches can be used to determine whether certain genetic molecular tests have been performed. For example, if one or more terms of one or more key semantic patterns listed in Table 8 are searched on the entire pathology report, it can be determined that the corresponding genetic molecular test has been performed. Key semantic patterns in genetic molecular testing ALK, ROS1, BRAF, MET, KRAS, ERBB2, PIK3CA, NRAS, MEK1, NTRK, RET Table 8

The key semantic patterns listed in Table 8 can be used to locate or identify specific genetic molecular detection portions. After locating or identifying a portion of a genetic molecule tested, further information related to the results can be obtained. For example, if the semantic context of a genetic molecular test portion includes the term "positive," it indicates that a mutation may occur in the corresponding gene. If the semantic context of a genetic molecular test section includes the term "negative," it indicates that a mutation may not have occurred in the corresponding gene.

In some embodiments, after locating or identifying a genetic molecule detection portion, a first modifier may be located or identified before the corresponding key semantic pattern, and a second modifier may be located or identified after the corresponding key semantic pattern. For the located genetic molecule detection part, a first distance between the first modifier and the corresponding key semantic pattern and a second distance between the second modifier and the corresponding key semantic pattern can be calculated. If the first distance is less than the second distance, the first modifier will be determined as the modifier of the located genetic molecule detection portion; if the second distance is less than the first distance, the second modifier will be determined as the located portion. Modifier for the genetic molecular testing part. The first modifier and the second modifier can be "masculine" or "feminine".

A pathology report contains information classified into pathology TNM stages. In some embodiments of the present invention, the semantic patterns generated for pathological TNM stages indicate some terms. One or more exemplary key semantic patterns of pathological TNM stages are listed in Table 9. Key semantic patterns of pathological TNM stages Pathological staging, pTNM, AJCC Table 9

Search the entire pathology report based on one or more terms of one or more key semantic patterns listed in Table 9. One or more of the key semantic pattern terms listed in Table 9 can be used to locate or identify the segment of the pathological TNM stage. For example, the pathological TNM stage section may be the "pathological stage (pTNM)" section.

Figure 4 illustrates a pathological TNM stage section of a pathology report according to some embodiments of the present invention. The entire pathology report can be searched based on terms of one or more key semantic patterns listed in Table 9. For example, if one or more terms of one or more key semantic patterns listed in Table 9 are searched on the entire pathology report, the pathological TNM stage section can be located or identified. Figure 4 shows an example in which the terms "pathological stage", "pTNM" and "AJCC" are positioned at the beginning of the pathological TNM stage section. The staging information of the items in the pathological TNM stage section will be further obtained. Items in the pathological TNM stage section include: pT (i.e. primary tumor), pN (i.e. regional lymph nodes), and pM (i.e. distant metastasis).

In some embodiments, after locating the pathological TNM stage segment, the version number provided after the term "AJCC" (ie, American Comprehensive Committee on Cancer) will be obtained. Installment information can be provided after or below the AJCC edition number. For example, Figure 4 shows that the items "primary (pT), regional lymph node (pN), distant metastasis (pM)" and corresponding information are provided below the AJCC version number.

The associated semantic pattern indicates that the staging results may follow the name of a given project. For example, the staging results for the item "primary tumor (pT)" can be found after the item's name (eg, "primary tumor" or "pT"). Figure 4 shows the staging result "pT1B" for the project "original (pT)" provided after the project name.

The staging information of one item in the pathological TNM stage section can be described after a colon (i.e. ":"). In some embodiments of the present invention, the information described in each item can be obtained by locating the corresponding colon (ie, ":"). For example, after locating the pathological TNM stage section, a given item can be located by searching the item name; then the corresponding colon of the given item can be located, and the information after the colon located in the given item will be obtained as the Project information. Figure 4 shows the staging result "pN0" for the item "regional lymph node (pN)" provided after the corresponding colon.

In some embodiments, the pathological TNM stage section may further include the item "TNM stage grouping". Figure 4 shows one of the pathological TNM stage sections including the item "TNM stage grouping". The information of the item "TNM stage grouping" indicates a TNM stage based on the information of the items "primary (pT), regional lymph node (pN)" and "distal metastasis (pM)".

The associated semantic pattern indicates that the TNM period results can be followed by the name of the project "TNM period grouping". Figure 4 shows the TNM phase result "Phase IA2" of the project "TNM Phase Group" provided after the project name.

The TNM installment information of the item "TNM installment group" can be described after a colon (i.e. ":"). In some embodiments of the present invention, the TNM staging information described in each item can be obtained by locating the corresponding colon (ie ":"). For example, after locating the project by searching for the project name "TNM installment group", the corresponding colon of the project can then be located, and the TNM installment information located after the colon of the project will be obtained as the TNM installment information. Figure 4 shows the TNM installment result "Period IA2" provided with the item "TNM Period Group" after the corresponding colon.

In some embodiments, the pathological TNM stage section may not include the item "TNM stage grouping". The present invention can obtain TNM staging information based on the staging information of pT items, pN items, and pM items. In some embodiments, the TNM staging information can be obtained through the lookup table 10 . Table 10 is based on the 8th edition of the AJCC/UICC (Union for International Cancer Control) TNM staging system. In Table 10, T1, T2, T3, T4 and the corresponding subcategories are the stages of “primary tumor (pT)”; N0, N1, N2, and N3 are the stages of “regional lymph node (pN)”; and M1 and corresponding subcategories are the phases of "remote transfer (pM)". In Table 10, IA1, IA2, IA3, IB, IIA, IIB, IIIA, IIIB, IIIC, IVA, and IVB are the phases of the "TNM phase grouping". T/M subcategory N0 N1 N2 N3 T1 T1a IA1 IIB IIIA IIIB T1b IA2 IIB IIIA IIIB T1c IA3 IIB IIIA IIIB T2 t2a IB IIB IIIA IIIB T2b IIA IIB IIIA IIIB T3 T3 IIB IIIA IIIB IIIC T4 T4 IIIA IIIA IIIB IIIC M1 M1a IVA IVA IVA IVA M1b IVA IVA IVA IVA M1c IVB IVB IVB IVB Table 10

In some embodiments of the present invention, a field of "N information" can be generated to verify or supplement the information of the item "regional lymph node (pN)". In some embodiments, a pathology report may provide the lymph nodes examined and associated information. The present invention can further verify the staging information of "regional lymph nodes (pN)" based on the examined lymph nodes and related information.

Figure 5 illustrates a portion of a pathology report according to some embodiments of the present invention. In Figure 5, the lymph nodes examined and associated information are provided. In Figure 5, the third row indicates: 7 lymph nodes were tested at the "hilar" position, 2 of which were involved; 5 lymph nodes were tested at the "5" position, and 1 lymph node was involved; 5 lymph nodes were tested at the "6" position 2 lymph nodes, none of which are involved; 1 lymph node at location "9", none of which is involved; 2 lymph nodes at location "10", one of which is involved; and 8 lymph nodes at location "11" lymph nodes, 5 of which were involved.

The present invention will determine the pN stage based on the involved or invaded lymph nodes. In the pathology report shown in Figure 5, 9 lymph nodes were involved or invaded. Based on the clinician's staging method, the staging results of different locations in the pathology report in Figure 5 can be obtained. The stage of "hilar" position is N2 because 2 of the 7 lymph nodes are involved. The stage at "No. 5" is N2 because one of the two lymph nodes is involved. The stage at the "No. 10" position is N1 because one of the two lymph nodes is involved. The stage at "No. 11" is N1 because 5 of the 8 lymph nodes are involved. Among the staging results at different positions, the pN period will be determined as the largest period. For the pathology report shown in Figure 5, the pN stage will be judged as "N2". This is because "N2" is among the stages at the positions of "hilar", "No. 5", "No. 10" and "No. 11" The biggest difference. In the pathology report in Figure 5, the item "regional lymph node (pN)" also provides the stage "N2".

In order to verify the performance of the method provided in the present invention, according to some embodiments of the present invention, 849 pathology reports from 203 lung cancer patients were provided to a program. The accuracy of illustrative 50 pathological features of lung cancer is provided in Table 11. For an entire pathology report, the method of the present invention can provide an overall accuracy of 86.69%. Basic description organ Bx site Sampling method Diagnosis 98.8% 89.5% 97.9% 93.6% discover biggest size tumor size nearest edge 98.5% 98.2% 99.5% lymphovascular invasion VPI tumor focus 99.4% 99.8% 99.8% Histology Histological type, Organization semester 97.9% 100% IHC CK7 TTF-1 aspartic proteinase A CK20 P40 98.3% 95.4% 97.1% 99.8% 98.6% CDX2 P63 P16 Cytokeratin (AE1/AE3) Vimentin 100% 100% 99.8% 99.8% 100% PAX-8 CD56 chromogranin A synaptophysin GATA3 100% 100% 99.8% 100% 99.7% P53 S100 Ki67 EBER 100% 99.8% 99.2% 100% genetic testing EGFR ALK ROS1 BRAF MET 100% 100% 99.2% 100% 100% KRAS ERBB2 PIK3CA NRAS MEK1 100% 100% 100% 100% 100% NTRK RET PDL1 100% 100% 98.8% TNM period Version pT pN pM pStage 100% 98.3% 99.7% 99.8% 94.8% N information 100% Table 11

In some embodiments of the present invention, if a patient has multiple pathology reports, these pathology reports will be summarized, combined and divided into "an initial diagnosis report" and "a latest diagnosis report". The time stamps of the initial diagnosis and the latest diagnosis should be determined. For example, after arranging multiple pathology reports in time series, the time stamps of the initial diagnosis and the latest diagnosis may be the dates of the first non-puncture pathology report and the latest non-puncture pathology report respectively.

Based on the experience of clinicians, information on genetic testing and IHC staining may not appear in the first non-puncture pathology report and the latest non-puncture pathology report. Genetic testing and IHC staining testing can be performed one month after the first non-puncture pathology report and the latest non-puncture pathology report. Therefore, the contents of pathology reports within one month from the time stamp of the first non-puncture pathology report or the latest non-puncture pathology report should be monitored or retained. Information on basic description, tumor findings and histology from the first non-puncture pathology report or the latest non-puncture pathology report can be retained. Information about genetic testing and IHC staining testing can be summarized and combined with information about basic description, tumor findings, and histology. In some embodiments, the information in the latest non-puncture pathology report and the information about genetic testing and IHC staining testing in (several) subsequent pathology reports may be for patients who have been treated by surgery or therapy.

According to some embodiments of the present invention, each of multiple pathology reports for a patient will be summarized based on illustrative 50 pathology features. In summary, each of multiple pathology reports for a patient can be expressed in terms of illustrative 50 pathological features. Data expression for each of multiple pathology reports of a patient may be expressed based on 50 exemplary pathological features. Multiple pathology reports that have been expressed in terms of the illustrative 50 pathology features can be combined based on the sequence of time stamps.

The data expression of a patient's pathology report can be expressed based on 50 exemplary pathological characteristics. In some embodiments, if the data for a given pathological characteristic is descriptive data, the descriptive data may be expressed using a sentence embedding method. In some embodiments of the present invention, the sentence embedding method can be trained based on pathology reports and can be 300-dimensional. Sentence embedding is a general term for a set of natural language processing (NLP) techniques in which sentences are mapped to real-number vectors. Sentence embedding is a technique used to express a descriptive feature, in which words or paragraphs are mapped or projected into a high-dimensional space, and the meaning of the descriptive feature is expressed in a high-dimensional space or a high-dimensional vector.

The data expression of a patient's pathology report can be expressed based on 50 exemplary pathological characteristics. Therefore, pathology reports can be compared with each other. However, the importance of each pathological feature may differ from one another. Therefore, in some embodiments of the invention, different pathological features may be weighted with different weight values. Exemplary weight values for 50 pathological features are provided in Table 12. According to some embodiments of the present invention, data expression in a pathology report may be expressed based on 50 exemplary pathological characteristics. In addition, the data representation of a multiple pathology report can be normalized based on the weighted 50 pathological features, where the weighted 50 pathological features can be weighted with the weight values provided in Table 12. Basic description organ Bx site Sampling method Diagnosis 100 5 100 100 discover biggest size tumor size nearest edge 1 1 20 lymphovascular invasion VPI tumor focus 20 20 5 Histology Histological type, Organization semester 100 5 IHC CK7 TTF-1 aspartic proteinase A CK20 P40 5 20 5 5 5 CDX2 P63 P16 Cytokeratin (AE1/AE3) Vimentin 5 20 20 5 5 PAX-8 CD56 chromogranin A synaptophysin GATA3 1 5 5 5 5 P53 S100 Ki67 EBER 40 5 5 5 genetic testing EGFR ALK ROS1 BRAF MET 100 100 100 100 100 KRAS ERBB2 PIK3CA NRAS MEK1 100 100 100 100 100 NTRK RET PDL1 100 100 50 TNM period Version pT pN pM pStage 5 5 5 5 100 N information 5 Table 12

Figure 6 illustrates a diagram of data representation of pathology reports 610 and 620 according to some embodiments of the present invention. Pathology reports 610 and 620 may belong to the same patient or different patients. The pathology report 610 can be expressed as a pathology report feature vector V _j . The pathology report feature vector V _j may include several features summarized or extracted from the pathology report 610 . The characteristics of the pathology report feature vector V _j can be summarized or extracted from the pathology report 610 through one or more methods or algorithms of the present invention. In Figure 6, the pathology report feature vector V _j contains features f _j1 to f _j50 . Features f _j1 to f _j50 may be numerical data or descriptive data corresponding to the 50 features listed in Table 1, Table 11 or Table 12. In the embodiment of FIG. 6 , features f _j1 and f _j50 are coefficient value data, and features f _j2 and f _j3 are descriptive data.

The pathology report 620 can be expressed as a pathology report feature vector V _k . The pathology report feature vector V _k may include several features summarized or extracted from the pathology report 620 . The characteristics of the pathology report feature vector V _k can be summarized or extracted from the pathology report 620 through one or more methods or algorithms of the present invention. In Figure 6, the pathology report feature vector _Vk contains features _fk1 to _fk50 . The features f _k1 to f _k50 may be numerical data or descriptive data corresponding to the 50 features listed in Table 1, Table 11 or Table 12. In the embodiment of FIG. 6 , features f _k1 and f _k50 are coefficient value data, and features f _k2 and f _k3 are descriptive data.

For each of the characteristics of numerical data (for example, each of the characteristics f _j1 , f _j50 , f _k1 , and f _k50 in FIG. 6 ), the numerical data can be further transformed into a category type data. For example, the features f _j1 , f _j50 , f _k1 , and f _k50 in Figure 6 are transformed into categories C _j1 , C _j50 , C _k1 , and C _k50 respectively. In some embodiments, when the numerical data of feature f _j1 is lower than a given threshold value, the value of category C _j1 may be 0; when the numerical data of feature f _j1 is greater than the given threshold value, the value of category C _j1 The value can be 1.

For each person with the characteristics of descriptive data (for example, each of the characteristics f _j2 , f _j3 , f _k2 , and f _k3 in Figure 6 ), the descriptive data can be further transformed into a multi-dimensional space through sentence embedding ( For example, 300-dimensional space). For example, the features f _j2 , f _j3 , f _k2 , and f _k3 in Figure 6 are transformed into sentence vectors Em _j2 , Em _j3 , Em _k2 , and Em _k3 respectively. Therefore, the pathology report feature vectors V _j and V _k can be transformed into vectors V' _j and V' _k .

Each element of vector V' _j may be equivalent to the corresponding element of vector V' _k . Therefore, vectors V' _j and V' _k may be equivalent based on a comparison between corresponding elements of vectors V' _j and V' _k . That is, it can be determined how similar the pathology reports 610 and 620 are based on the comparison between the corresponding elements of the vectors V′ _j and V′ _k .

For category type data in vectors V' _j and V' _k (for example, C _j1 , C _j50 , C _k1 , and C _k50 ), if one category of vector V' _j matches (or is the same as) one of vector V' _k corresponds to one, then the comparison result between the two corresponding categories of vectors V' _j and V' _k will be 1 (ie, indicating that the match is true). If a category of vector V' _j does not match (or is not identical to) a corresponding category of vector V' _k , then the comparison between the two corresponding categories of vectors V' _j and V' _k will be 0 (e.g., indicates that the match was false). For example, if category C _j1 of vector V' _j matches (or is the same as) category C _k1 of vector V' _k , then the comparison between categories C _j1 and C _k1 is 1 (eg, indicating that the match is true). If category C _j50 of vector V' _j does not match (or is not the same as) category C _k50 of vector V' _k , then the comparison between categories C _j50 and C _k50 is 0 (eg, indicating a false match). For example, the comparison between categories C _j1 and C _k1 or between categories C _j50 and C _k50 can be expressed as equation (6). Equation (6)

For the sentence vectors transformed by sentence embedding in vectors V' _j and V' _k (for example, Em _j2 , Em _j3 , Em _k2 , and Em _k3 ), their equality is determined based on the cosine similarity of the two corresponding bi-sentence vectors. similarity between them. For example, the similarity between the sentence vectors Em _j2 and Em _k2 is determined based on the cosine similarity between the sentence vectors Em _j2 and Em _k2 . The similarity between the sentence vectors Em _j3 and Em _k3 is determined based on the cosine similarity between the sentence vectors Em _j3 and Em _k3 . For example, the cosine similarity between sentence vectors Em _j2 and Em _k2 or between sentence vectors Em _j3 and Em _k3 can be expressed as equation (8). In equation (8), the sentence vector is located in a 300-dimensional space. Equation (7)

The 50 weight values listed in Table 12 can be further applied to the similarity between two pathology reports. The normalization of the 50 weight values in Table 12 can be expressed as equation (9). The variables w ₁ to w ₅₀ indicate the 50 weight values listed in Table 12, and the variables w' ₁ to w' ₅₀ indicate the corresponding 50 normalized weight values. Equation (8)

The similarity score between the n-th feature of the pathology report 610 and the n-th feature of the pathology report 620 may be the similarity score between the n-th feature of the vector _V'j and the n-th feature of the vector _V'k . The similarity score of the nth feature can be expressed as equation (9). When the nth feature coefficient value data or category type data, match _n will be normalized by the corresponding weight value. When the nth feature is descriptive information or a sentence vector, similarity _n will be multiplied by the corresponding normalized weight value. Equation (9)

The similarity score between the pathology reports 610 and 620 may be the sum of the similarity scores from the first feature to the 50th feature. The similarity score between vectors V' _j and V' _k may be the sum of the similarity scores from the first feature to the 50th feature. The sum of the similarity scores from the first feature to the 50th feature can be expressed as equation (10). Equation (10)

Therefore, the present invention can provide clinicians with an efficient method to score the similarity between two cases. Because the similarity of two pathology reports for two patients can be scored, clinicians can more easily search for similar cases in the past. In addition, clinicians can know which parts of two cases are similar and why the two cases are similar based on the scores of 50 important pathological features as shown in Table 1.

Therefore, the present invention provides an interpretable method of scoring the similarity between two cases. In contrast, many parameters and probabilities generated by a machine learning algorithm (or a deep learning algorithm) are not interpretable. The interpretability of the invention enables clinicians to explain how two cases are similar and why. The interpretability and similarity provided by the present invention will assist clinicians in evaluating a patient's subsequent diagnosis and treatment.

Referring to Figure 7, shown is an example of a computer system 700 capable of performing one or more operations of the method of the present invention. In at least some embodiments of the present invention, computer system 700 includes a computing device 710 and a database 720 . Computing device 710 may be a server computer, a client computer, a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, or a smart phone. The computing device 710 includes a processor 711, an input/output interface 712, a communication interface 713, and a memory 714. Database 720 may store pathology reports from which key semantic patterns are to be extracted. The database 720 may store pathology reports to be analyzed or summarized. Input/output interface 712 is coupled to processor 711. The input/output interface 712 allows the user to manipulate the computing device 710 to perform operations or methods of the present invention (eg, the method disclosed in FIG. 2 ). Communication interface 713 is coupled to processor 711. The communication interface 713 allows the computing device 710 to communicate with the database 720 . Communication interface 713 may support one or more of the following protocols: Universal Serial Bus (USB), Ethernet, Bluetooth, IEEE 802.11, 3GPP Long Term Evolution (LTE) (4G), and 3GPP New Radio (5G). A memory 714 may be a non-transitory computer-readable storage medium. Memory 714 is coupled to processor 711. Memory 714 has stored program instructions executable by one or more processors (eg, processor 711). When executed, program instructions stored on memory 714 cause the performance of one or more operations of the methods disclosed in this disclosure. For example, the program instructions may cause the computing device 710 to perform a set of actions including at least the following steps: (i) determining a degree of trust between a semantic term and a next semantic term based on the co-occurrence of the semantic term and the next semantic term. and a support; (ii) generate a set of candidate semantic terms; (iii) generate a first set of semantic patterns by performing a random walk on the set of candidate semantic terms; (iv) generate a first set of semantic patterns by removing redundant semantic patterns from the first set and (v) generating a second group of semantic patterns based on the remaining semantic patterns; and (v) outputting key semantic patterns based on the second group of semantic patterns.

As another illustrative example, program instructions may cause the computing device 710 to perform a method of summarizing a pathology report. The method may include obtaining a plurality of pathology features from the pathology report based on key semantic patterns. Key semantic patterns are generated according to any of the methods of the present invention.

The scope of the invention is not intended to be limited to the particular embodiments of the procedures, machines, manufacture and composition of matter, means, methods, steps and operations described in the specification. Those skilled in the art will readily appreciate from the disclosure of the present invention that currently existing or hereafter embodiments that perform substantially the same functions or achieve substantially the same results as corresponding embodiments described herein may be utilized in accordance with the present invention. A process, machine, manufacture, composition of matter, component, method, step or operation developed. Therefore, the patent scope of the appended invention application is intended to include within the scope of the process, machine, manufacture and material composition, component, method, step or operation thereof. Additionally, each claim constitutes a separate embodiment and combinations of various claims and embodiments are within the scope of the invention.

Methods, procedures or operations according to embodiments of the present invention may also be implemented on a programmed processor. However, the controller, flowcharts, and modules may also be represented by a general or special purpose computer, a programmed microprocessor or microcontroller and peripheral integrated circuit components, an integrated circuit, a hardware electronic or logic circuit ( such as a discrete component circuit), a programmable logic device, or the like. In general, any device having resident thereon a finite state machine capable of implementing the flowcharts shown in the figures may be used to implement the processor functions of the present invention.

An alternative embodiment preferably implements methods, procedures or operations according to embodiments of the invention in a non-transitory computer-readable storage medium storing computer programmable instructions. The instructions are preferably executed by a computer executable component preferably integrated with a network security system. The non-transitory computer-readable storage medium can be stored on any suitable computer-readable medium, such as RAM, ROM, flash memory, EEPROM, optical storage device (CD or DVD), hard drive, floppy disk drive, or any other suitable device. The computer-executable component is preferably a processor, but the instructions may alternatively or additionally be executed by any suitable special purpose hardware device. For example, one embodiment of the invention provides a non-transitory computer-readable storage medium having computer programmable instructions stored therein.

Although the invention has been described using specific embodiments thereof, it will be apparent to those skilled in the art that many alternatives, modifications, and variations will be apparent to those skilled in the art. For example, various components of the embodiments may be interchanged, added, or replaced in other embodiments. Furthermore, not all elements of the figures are necessary to the operation of the disclosed embodiments. For example, a person of ordinary skill will be able to make and use the teachings of the present invention, by simply employing the elements of the independent claims. Accordingly, the embodiments of the present invention as set forth herein are intended to be illustrative and not restrictive. Various changes may be made without departing from the spirit and scope of the invention.

Even though the several features and advantages of the invention have been set forth in the foregoing description, as well as details of the structure and function of the invention, the invention is intended to be illustrative only. Within the principles of the invention, modifications may be made in details (especially relating particularly to the shape, size, and arrangement of the parts) to give the full scope indicated by the broad and common meaning of the terms indicating the patentable scope of the invention to which the invention is claimed.

100: Pathological semantic association map

200:Method

201:Operation

203:Operation

205:Operation

205:Operation

207:Operation

209:Operation

610: Pathology report

620: Pathology report

700:Computer system

710:Computing device

711: Processor

712:Input/output interface

713: Communication interface

714:Memory

720:Database

For purposes of describing the manner in which the advantages and features of the invention may be obtained, a description of the invention is presented by reference to specific embodiments of the invention illustrated in the accompanying drawings. The drawings depict only example embodiments of the invention and, therefore, should not be construed as limiting its scope.

Figure 1 illustrates a pathological semantic association diagram according to some embodiments of the present invention.

FIG. 2 illustrates a flow chart of a method for extracting key semantic patterns according to some embodiments of the present invention.

Figure 3 is a section of a pathology report according to some embodiments of the present invention.

Figure 4 is a section of a pathology report according to some embodiments of the present invention.

Figure 5 is a portion of a pathology report according to some embodiments of the present invention.

Figure 6 is a schematic diagram illustrating data expression of a pathology report according to some embodiments of the present invention.

FIG. 7 is a schematic diagram showing a computer system according to some embodiments of the present invention.

700:Computer system

710:Computing device

711: Processor

712:Input/output interface

713: Communication interface

714:Memory

720:Database

Claims (20)

A method for extracting key semantic patterns from a pathology report through a computing device, which includes: based on the co-occurrence of a semantic term and a following semantic term, using a processor of the computing device to determine the semantic term and the following semantic term. A degree of reliability and a degree of support between a semantic term, wherein in the pathology report, the semantic term appears before the next semantic term; a set of candidate semantic terms is generated by the processor of the computing device, wherein The degree of trust between a candidate semantic term and a corresponding next candidate semantic term is equal to or greater than a trust threshold value, and the degree of support between the candidate semantic term and the corresponding next candidate semantic term is equal to or greater than a supporting a threshold; by the processor of the computing device, generating a first set of semantic patterns by performing a random walk on the set of candidate semantic terms; and by the processor of the computing device, by performing a random walk on the set of candidate semantic terms; The group semantic pattern removes redundant semantic patterns and determines the key semantic patterns. The method of claim 1, wherein: based on the number of co-occurrences of the semantic term and the next semantic term, the processor of the computing device generates the support between the semantic term and the next semantic term; And when the semantic term appears, the reliability between the semantic term and the next semantic term is generated by the processor of the computing device based on an occurrence probability of the next semantic term. The method of claim 1, wherein removing the redundant semantic patterns from the first set of semantic patterns includes when a first semantic pattern indicates a subset of a second semantic pattern, by the processing of the computing device The processor removes the first semantic pattern. The method of claim 1, wherein the degree of reliability between the semantic term and the next semantic term is independent of a second degree of reliability between a previous semantic term and the semantic term, wherein in the pathology report, the The previous semantic term appears before this semantic term. A method for summarizing a pathology report through a computing device, which includes: based on key semantic patterns, using a processor of the computing device to obtain a plurality of pathological features from the pathology report, wherein the method generated according to claim 1 These key semantic patterns. The method of claim 5, further comprising: locating, by the processor of the computing device, a description in a SOAP (Subjective, Objective, Assessment, and Plan) section, wherein the description has four or more parts, and a "," separates any two adjacent parts, where when the description has four parts, a first part of the description relates to an organ, a second part of the description relates to a location, the The third part of the description relates to a sampling method, and the fourth part of the description relates to a diagnosis. Such as the method of claim 6, wherein when the description has four parts, the method further includes: locating a first part related to an organ based on the key semantic patterns and a sample a second part associated with a method; and a third part associated with a location positioned between the first part and the second part; and a fourth part associated with a diagnosis positioned after the second part. Such as requesting the method of item 7, wherein: the first part related to the organ is located based on a first subset of the key semantic patterns, and the first subset of the key semantic patterns includes "lungs", "lymph nodes" ", "brain" and "skin"; and the second part related to the sampling method is located based on a second subset of the key semantic patterns, the second subset of the key semantic patterns includes "biopsy" ”, “VATS” and “EBUS”. The method of claim 5, further comprising: locating an immunohistochemistry (IHC) section based on a third subset of the key semantic patterns, wherein the third subset of the key semantic patterns includes "immunohistochemistry" of", "immunohistochemistry", "immunohistochemistry", "immunohistochemistry", "immunoresearch", "IHC", "immunostaining", "immunohistochemistry", "immunoresearch", " Immune response", "showing adenocarcinoma composition". The method of claim 9, further comprising: locating, by the processor of the computing device, a characteristic term of a fourth subset of the key semantic patterns in the IHC part; Locating by the processor of the computing device a first modifying term that appears before the characteristic term; locating a second modifying term that appears after the characteristic term by the processor of the computing device; and based on A first distance between the first modified term and the characteristic term and a second distance between the second modified term and the characteristic term, the processor of the computing device selects the first modified term and One of the second modifying terms. For example, the method of claim 10, wherein: the fourth subset of the key semantic patterns includes "CK7", "TTF-1", "Aspartate Protease A", "CK20", "P40", "CDX2" ", "P63", "P16", "cytokeratin (AE1/AE3)", "vimentin", "PAX-8", "CD56", "chromogranin A", "synaptophysin", " GATA3", "P53", "S100", "Ki67" and "EBER"; and the first modified term and the second modified term include "masculine" and "feminine". The method of claim 6, further comprising: locating at least one candidate segment by the processor of the computing device based on a semantic pattern of the volume; when the semantic context of one of the at least one candidate segment includes the key When a key semantic pattern of one of the fifth subsets of semantic patterns is obtained, by the processor of the computing device, one of the volumes of the at least one candidate fragment is obtained as a tumor size; and by the computing device The processor determines a maximum value of the tumor size as a maximum Large size, wherein the fifth subset of the key semantic patterns includes "cutting", "hard tumor" and "tumor measurement". The method of claim 12, further wherein the semantic pattern of the volume includes: a first number, a second number, and a third number, a multiplication symbol between the first number and the second number, the third number Another multiplication symbol between the second number and the third number, and a unit of length after the third number. The method of claim 5, further comprising: locating a microscope evaluation section by the processor of the computing device based on the term "microscope evaluation"; and locating a microscope evaluation section by the processor of the computing device in the locating a colon of one of the first group items in the microscope evaluation section; and obtaining, by the processor of the computing device, information following the colon as the item of the first group of items, wherein the first group Items include: tumor focus, histological type, histological grade, lymphovascular invasion, visceral pleural invasion, and nearest margin. The method of claim 5, further comprising: locating a PD-L1 detection portion by the processor of the computing device based on a sixth subset of the key semantic patterns; and Locating a colon of one of the items in a second group of items in the PD-L1 detection part by the processor of the computing device; and obtaining information after the colon as the The item of the second group of items, wherein the sixth subset of the key semantic patterns includes: "22C3", "28-8", "SP142" and "SP263", and the second group of items includes: tumor proportion score (TPS), combined positive score (CPS), tumor cells (TC), and immune cells (IC). The method of claim 5, further comprising: locating an EGFR portion by the processor of the computing device based on a term of epidermal growth factor receptor (EGFR); and based on "18", "19", " 20", and "21", the processor of the computing device determines whether the mutation is in exon 18, exon 19, exon 20 or exon 21; and based on "T790M" In one term, the processor of the computing device determines that a mutation is located at position 790 in exon 20. The method of claim 5, further comprising: locating, by the processor of the computing device, a molecule detection portion based on a seventh subset of key semantic patterns; and locating, by the processor of the computing device, Identify a modifying term in the semantic context of a key semantic pattern in the seventh subset of key semantic patterns; and When the modified term is identified as "positive", it is determined by the processor of the computing device that a mutation is in a gene associated with the one key semantic pattern, wherein the seventh subset of the key semantic patterns Includes: "ALK", "ROS1", "BRAF", "MET", "KRAS", "ERBB2", "PIK3CA", "NRAS", "MEK1", "NTRK" and "RET". The method of claim 5, further comprising: locating a pathological staging (pTNM) portion by the processor of the computing device based on a term of "pathological staging" and a term of "pTNM"; and based on "pT ” a term based on retrieving a first phase indicator by the processor of the computing device; a term based on “pN” retrieving a second phase indicator by the processor of the computing device indicator; and retrieving a third phase indicator by the processor of the computing device based on the term "pM". The method of claim 5, further comprising: converting the pathology report into a first vector by the processor of the computing device according to the plurality of pathology characteristics, wherein the first vector includes multiple elements of a category and multiple elements of the sentence vector; according to the plurality of pathological characteristics, a second pathology report is transformed into a second vector by the processor of the computing device, wherein the second vector includes multiple elements of the category and a plurality of elements of the statement vector; the processor of the computing device calculates a difference between the pathology report and the second pathology report by summing the scores of the corresponding elements of the first vector and the second vector. Similarity score, where when an n-th element is one of the elements of the category, one of the n-th element's scores is

, C _1n indicates the n-th element of the first vector, C _2n indicates the n-th element of the second vector, w _n indicates a weight value of the n-th element, and where when the n-th element When the series is one element of the sentence vector, the fractional system of the n-th element

, Em _1n indicates the n-th element of the first vector, and Em _2n indicates the n-th element of the second vector. A non-transitory computer storage medium having program instructions stored thereon which, when executed by a processor, result in the execution of a set of operations such as the method of claim 1.