CN109522974B - Lesion level selection method to improve the positive rate of needle biopsy - Google Patents

Abstract

The invention discloses a lesion level selection method for improving the positive rate of puncture biopsy. The system includes a data acquisition module, an image scanning module, an image processing module, a data analysis module and a storage module. The selection method is as follows: S1: collect training data; S2: according to training Data delineate the ROI at the sample lesion level; S3: Establish a positive rate prediction model; S4: Obtain prediction parameters according to the positive rate prediction model; S5: Fit the obtained prediction parameters to obtain the final prediction data; S6: Delineate the patient's lesions according to the final prediction data Level ROI and obtain patient characteristics; S7: Obtain the predictive value of the positive rate according to the patient lesion level ROI and patient characteristics; S8: Select the lesion level ROI, the advantage is that the quantitative imaging characteristics of the lesions and the doctor’s empirical characteristics are used to reduce The subjectivity of the skin puncture level selection improves the stability of the puncture level selection technique and the positive rate of the puncture biopsy technique.

Description

Lesion level selection method to improve the positive rate of needle biopsy

technical field

The invention relates to the technical field of medical detection, in particular to a lesion level selection method for improving the positive rate of puncture biopsy.

Background technique

Before determining the clinical diagnosis and treatment plan, it is necessary to biopsy the lesions and conduct relevant tests such as pathological cells to determine the nature of the lesions, and then formulate and implement relevant treatment plans.

Due to the heterogeneity of tumors, the choice of puncture level is directly related to the number of punctures and the positive rate of the test results. At present, the selection of percutaneous puncture levels is mainly based on the medical images of patients such as CT, MRI, PET, US, etc., and the radiologists observe the selected puncture levels with the naked eye based on personal experience, and then perform biopsy to obtain materials. Due to the heavy subjectivity of the existing methods, there are no quantitative indicators to guide the selection of puncture levels and sites, and the positive rate is not high. Sometimes repeated inspections are required, which increases the economic burden of patients and the probability of complications. Wang Jie et al. compared the number of puncture points, biopsy times, sampling satisfaction rate, and pathological positive results of lung CT plain scan and enhanced image-guided puncture. The study showed that doctors’ subjective selection of puncture layers based on images requires a higher proportion of repeated examinations, and positive results. The rate is not high, which increases the incidence of complications, and at the same time brings a greater probability of radiation damage and economic burden to patients; Wang Song compared the artificial intelligence ultrasound CT-guided and ultrasound system combined with mpMRI-assisted system for prostate targeted puncture. The positive rate of targeted prostate puncture under the guidance of artificial intelligence ultrasound CT is higher than that of the ultrasound system combined with the mpMRI auxiliary system. By quantitatively extracting DWI images of rectal cancer, Liu et al. realized the use of extracted texture features to distinguish between pT3-4 and pT1-2 of tumors, and at the same time, it has significance in identifying whether there is lymph node involvement. This study further verifies the medical The application value of image quantitative features in clinical prediction and discrimination problems. In addition, the uneven distribution of senior radiologists resources limits the implementation of percutaneous puncture in areas with insufficient doctor resources.

At present, the subjective selection of the target level of puncture biopsy technology leads to low accuracy and poor stability of the positive rate judgment result, and repeated puncture increases the economic burden of patients and the incidence of complications. To a certain extent, the clinical application of needle biopsy technology is limited.

SUMMARY OF THE INVENTION

The purpose of the embodiments of the present invention is to provide a lesion level selection method for improving the positive rate of needle biopsy, so as to solve the problem of low accuracy of the positive rate judgment result caused by the subjective selection of the target level of the existing needle biopsy technology.

To achieve the above purpose, the technical solutions of the embodiments of the present invention are:

A lesion level selection system for improving the positive rate of needle biopsy, including a data acquisition module, an image scanning module, an image processing module, a data analysis module and a storage module;

The storage module is respectively electrically connected with the data analysis module, the data acquisition module, the image scanning module and the image processing module, and is used for storing data information;

The image scanning module is used to scan the image of the lesion level, and store the image data of the patient's lesion level into the storage module;

The data acquisition module is used to collect sample data and store the sample data in the storage module;

The image processing module is used to calculate the sample data in the storage module and the image data at the lesion level, and transmit the calculation result to the data analysis module;

The data analysis module is used to perform statistical analysis on the calculation results of the image processing module.

The lesion-level selection method to improve the positive rate of needle biopsy includes the following steps:

S1: Collect training data;

S2: Outline the ROI at the sample lesion level according to the training data;

S3: Establish a positive rate prediction model;

S4: Obtain prediction parameters according to the positive rate prediction model;

S5: Fitting the obtained prediction parameters to obtain the final prediction data;

S6: According to the final prediction data, delineate the ROI at the lesion level of the patient and obtain the patient characteristics;

S7: Obtain the predictive value of the positive rate based on the patient's lesion-level ROI and patient characteristics;

S8: Select the ROI at the lesion level.

The embodiment of the present invention is further configured as follows: after step S2 is completed, the following three parallel steps are performed,

C: Subjective scoring of ROI at the sample lesion level;

D1: Extract the sample features of the sample lesion ROI;

E: Deep learning is performed on the ROI image of the sample lesion.

The embodiment of the present invention is further configured as follows: step D1 further includes step D2: performing dimension reduction screening on the sample feature data of the sample lesion ROI.

The embodiment of the present invention is further set as: step S3 includes the following two parallel steps,

F: Establish a positive rate prediction model A according to the subjective score in step C and the sample feature data after dimensionality reduction and screening in step D2;

G: Establish positive rate prediction model B according to the deep learning data in step E.

The embodiment of the present invention is further set as: step S4 includes the following two parallel steps,

H: Obtain prediction parameter 1 according to model A in step F;

I: Obtain prediction parameter 2 according to model B in step G;

The fitting data source in step S5 is the prediction parameter 1 in step H and the prediction parameter 2 in step I.

The embodiment of the present invention is further set as follows: the samples collected in S1 are the latest image data before the puncture of at least 3000 patients with positive puncture results.

The embodiment of the present invention is further set as: the step of dimensionality reduction and screening is to extract more than 3000 features from the images of each patient in the training set, and screen out and delete useless information or features with small contribution value in prediction.

The advantages of the embodiments of the present invention are:

Compared with the traditional puncture level selection technology, the embodiment of the present invention combines radiomics and interventional medicine, and uses quantitative imaging features of lesions and empirical features of doctors, which greatly reduces the subjectivity of skin puncture level selection, and improves the efficiency of skin puncture level selection. The stability of the puncture layer selection technique is improved, and the positive rate of the puncture biopsy technique is improved. Increase the reproducibility of the technology by standardizing the process of the technology.

Description of drawings

Fig. 1 is the structural representation of embodiment 1;

2 is a schematic flowchart of a method for selecting a lesion level for improving the positive rate of needle biopsy in Embodiment 2;

FIG. 3 is a schematic flowchart of the method for selecting a lesion level for improving the positive rate of needle biopsy in Example 2. FIG.

in,

1. Data acquisition module;

2. Image scanning module;

3. Image processing module;

4. Data analysis module;

5. Storage module.

Detailed ways

The following examples are intended to illustrate the present invention, but not to limit the scope of the present invention.

Example 1

A lesion level selection system for improving the positive rate of needle biopsy, including a data acquisition module, an image scanning module, an image processing module, a data analysis module and a storage module; the storage module is respectively connected with the data analysis module, data acquisition module, image scanning module and image processing module The electrical connection is used to store data information; the image scanning module is used to scan the image of the lesion layer and store the image data of the patient's lesion layer in the storage module; the data acquisition module is used to collect the sample data and store the sample data in the storage module ; The image processing module is used to calculate the sample data in the storage module and the image data of the lesion level, and transmit the calculation results to the data analysis module; the data analysis module is used to perform statistical analysis on the calculation results of the image processing module.

The image scanning module is a CT scanner, the image processing module and the data analysis module both use a computer, and the storage module uses a mechanical hard disk or a solid-state hard disk. In practical applications, the instruments, devices, or hardware exemplified in this embodiment are not limited to be used, and any instruments, devices, or hardware that can satisfy the same or similar functions may be used.

Example 2

The lesion-level selection method to improve the positive rate of needle biopsy, as shown in Figure 2, includes the following steps:

S1: Collect training data;

S2: Outline the ROI at the sample lesion level according to the training data;

S3: Establish a positive rate prediction model;

S4: Obtain prediction parameters according to the positive rate prediction model;

S5: Fitting the obtained prediction parameters to obtain the final prediction data;

S6: According to the final prediction data, delineate the ROI at the lesion level of the patient and obtain the patient characteristics;

S7: Obtain the predictive value of the positive rate based on the patient's lesion-level ROI and patient characteristics;

S8: Select the ROI at the lesion level.

Among them, in conjunction with Figure 3,

After the completion of step S2, the following three parallel steps are performed,

C: Subjective scoring of ROI at the sample lesion level;

D1: Extract the sample features of the sample lesion ROI;

E: Deep learning is performed on the ROI image of the sample lesion.

Step D1 further includes step D2: performing dimension reduction screening on the sample feature data of the sample lesion ROI.

Step S3 includes the following two parallel steps,

F: Establish a positive rate prediction model A according to the subjective score in step C and the sample feature data after dimensionality reduction and screening in step D2;

G: Establish positive rate prediction model B according to the deep learning data in step E.

Step S4 includes the following two parallel steps,

H: Obtain prediction parameter 1 according to model A in step F;

I: Obtain prediction parameter 2 according to model B in step G.

The fitting data source in step S5 is the prediction parameter 1 in step H and the prediction parameter 2 in step I.

The samples collected in S1 are the most recent imaging data before the puncture of at least 3000 patients with positive puncture results;

In S2, before delineating the ROI of the sample lesion layer, first select the sample lesion layer that is positively matched with the puncture result;

The subjective score in step C is the score of different lesion levels by radiologists who have been practicing for at least three years, and the score range is 1-5 points.

The method of establishing the positive rate prediction model in S6 is to perform statistical analysis or deep learning on the sample features after dimensionality reduction and screening and combine the doctor's score, and then use the image of the ROI at the sample lesion level to perform deep learning training, and finally obtain the positive rate prediction. Model.

The image type of the sample lesion-level ROI must be consistent with the image type of the patient's lesion-level ROI (CT image, MRI image, fused CT and MRI image, PET image, US image, etc.).

The image imaging parameters selected for the prediction data are consistent with the image imaging parameters selected for the training data, which improves the prediction accuracy of the prediction model.

When selecting the lesion level, the patient lesion level should be matched with the sample lesion level as much as possible, and the matching degree should not be lower than 90%.

When performing puncture biopsy, the actual puncture level should match the lesion level of the patient selected for prediction as much as possible, and the matching degree should not be less than 90%.

The specific steps of dimensionality reduction screening are to extract more than 3000 features from the images of each patient in the training set, and there are more than 3000 patients in total, so the training data set matrix is more than 3000×3000. Among all the extracted features, some of the features contain information that is highly correlated in predicting the positive rate, and some are useless information or contribute too little to prediction. These useless information or features that contribute too little to prediction Filter them out and delete them to reduce the data dimension of the training set to facilitate model building.

Although the present invention has been described in detail above with general description and specific embodiments, some modifications or improvements can be made on the basis of the present invention, which will be obvious to those skilled in the art. Therefore, these modifications or improvements made without departing from the spirit of the present invention fall within the scope of the claimed protection of the present invention.

Claims (4)

1. A method for selecting a lesion level for improving the positive rate of needle biopsy, characterized in that it comprises the following steps: S1: Collect training data; S2: Outline the ROI at the sample lesion level according to the training data; S3: Establish a positive rate prediction model; S4: Obtain prediction parameters according to the positive rate prediction model; S5: Fitting the obtained prediction parameters to obtain the final prediction data; S6: According to the final prediction data, delineate the ROI at the lesion level of the patient and obtain the patient characteristics; S7: Obtain the predictive value of the positive rate based on the patient's lesion-level ROI and patient characteristics; S8: Selected lesion level ROI; After the completion of step S2, the following three parallel steps are performed, C: Subjective scoring of ROI at the sample lesion level; D1: Extract the sample features of the sample lesion ROI; E: Deep learning is performed on the ROI image of the sample lesion; Step D1 further includes step D2: performing dimension reduction screening on the sample feature data of the sample lesion ROI; Step S3 includes the following two parallel steps, F: Establish a positive rate prediction model A according to the subjective score in step C and the sample feature data after dimensionality reduction and screening in step D2; G: Establish positive rate prediction model B according to the deep learning data in step E. 2. The method for selecting a lesion level for improving the positive rate of needle biopsy according to claim 1, wherein step S4 comprises the following two parallel steps, H: Obtain prediction parameter 1 according to model A in step F; I: Obtain prediction parameter 2 according to model B in step G; The fitting data source in step S5 is the prediction parameter 1 in step H and the prediction parameter 2 in step I. 3 . The method for selecting a lesion level for improving the positive rate of puncture biopsy according to claim 1 , wherein the samples collected in S1 are the most recent imaging data before puncture of at least 3000 patients with positive puncture results. 4 . 4. The method for selecting a lesion level for improving the positive rate of needle biopsy according to claim 1, wherein the step of dimensionality reduction screening is to extract more than 3000 features from the images of each patient in the training set, and use the method for Information or features with small contribution to prediction are screened out and removed.

Images