CN111814901A - Simulation method of doctor's operation technique based on data mining and state learning - Google Patents

Abstract

The invention provides a physician operation method simulation method based on data mining and state learning. Learning the rule of the operation method of a doctor by taking parameters such as pressure of an ultrasonic probe and a tissue skin contact end, ultrasonic image quality, an ultrasonic probe normal vector and the like as learning objects in the operation process; a fuzzy C-means clustering method is adopted to perform clustering analysis on the operation methods of the region of interest and the common region, so that the complexity of the operation methods is simplified; the Markov chain model is adopted to perform state transition modeling on the dynamic change process of the parameters corresponding to the operation technique in the ultrasonic scanning process, describe the scanning rule of a doctor, effectively learn the operation technique of a professional doctor in the ultrasonic department, conveniently realize anthropomorphic automatic ultrasonic scanning of a robot and acquire high-quality medical ultrasonic images.

Description

Simulation method of doctor's operation technique based on data mining and state learning

technical field

The invention belongs to the technical field of data analysis, and in particular relates to a method for simulating a doctor's operation technique based on data mining and state learning.

Background technique

At present, although there are many methods for simulating the operation of doctors in the ultrasonic scanning robot system, these methods can only be used for certain specific clinical applications, and the operation process of the robot system lacks flexibility and intelligence. Methods and fixed operation procedures, or the scanning can be completed only after the operator teaches, and it is impossible to complete the scanning similar to the doctor changing the scanning method according to the specific situation of the patient.

Carotid artery scanning is proposed in the document "Abolmaesumi P, Salcudean S, Zhu W H, et al. Image-guided control of a robot for medical ultrasound[J]. IEEE Transactions on Robotics and Automation, 2002, 18(1): 11-23" The system can change the scanning method, but the system can only simply repeat the current operation method of the operator. On the other hand, the current ultrasound robot has no feedback process from imaging quality to control technique, and it is completely impossible to adjust the scanning technique in real time according to the ultrasound imaging effect.

SUMMARY OF THE INVENTION

In order to solve the problem that the learning and simulation effects of doctor's manipulations are poor due to differences in doctors' personalities and different situations of scanned persons, the present invention provides a method for simulating doctor's manipulations based on data mining and state learning. The parameters such as the pressure at the contact end of the ultrasound probe and the tissue skin during the doctor's operation, the perception of the ultrasound image, and the normal vector of the ultrasound probe are used as the learning objects to learn the rules of the doctor's operation; then, the fuzzy C-means clustering (FCM) data is used. The mining method conducts cluster analysis on the operation techniques of the region of interest and the common area to simplify the complexity of the operation techniques; then, the Markov chain model is used to analyze the dynamic changes of the corresponding parameters in the operation techniques of skilled physicians in the process of ultrasound scanning. The state transition modeling is carried out in the process to describe the doctor's scanning law to simulate the doctor's operation method.

A method for simulating a doctor's operation technique based on data mining and state learning, characterized in that the steps are as follows:

Step 1: Use the free-arm three-dimensional ultrasound imaging system to record the normal vector N of each scanning point during the ultrasound scanning process of the doctor holding the ultrasound probe in different tissue parts of the human body, and use the force sensor attached to the front of the ultrasound probe to record the scanning process. The pressure value F of the contact surface between the ultrasonic probe of each scanning point and the human tissue site is recorded as Q for the quality evaluation value of the currently scanned ultrasonic image, and the three form a doctor's manipulation characteristic sample x=(N,Q,F) ^T ;

Step 2: Use the method of step 1 for data collection, and set the collected physician manipulation data set as X={x ₁ , x ₂ ,..., x _n }, where n is the total number of samples of the manipulation data set, and each A sample x _i =(N _i ,Q _i ,Fi ) ^T , _i =1,2,...,n;

Use the fuzzy C-means clustering algorithm to cluster the data set X, and obtain two data sets, which are recorded as the common area data set X ₁ and the interest area data set X ₂ respectively. The statistical variance of the data set X ₂ is greater than that of the data set. Statistical variance of X ₁ ;

Step 3: The subset T to be modeled is randomly selected from the data set X ₂ of the region of interest, and the state space S={s ₁ , s ₂ ,...,s _m is formed by counting non-repetitive operation methods in T }, where s _i =(N _i ,Q _i ,F _i ) ^T represents the state of the i-th operation technique, m is the number of state values, i=1,2,...,m;

The one-step transition probability of the physician transferring from the i-th manipulation to the j-th manipulation during the ultrasound scan is calculated according to the following formula:

Among them, P(s _j |s _i ) represents the one-step transition probability from the i-th method to the j-th method, k _j represents the number of times the state is transferred to the s _j state, and K _i represents the state is transferred to itself or the next The total number of times a state; i=1,2,...,m, j=1,2,...,m;

The one-step state transition matrix P of the subset T is calculated according to the following formula:

Wherein, p _ij =P(s _j |s _i ), i=1,2,...,m, j=1,2,...,m;

The initial distribution of state _si is calculated according to the following formula:

Among them, p _i (0) represents the initial distribution of state s _i , ξ _i represents the number of times the manipulation of state s _i occurs, [T] is the number of manipulation samples in subset T, i=1,2,... .,m;

Step 4: Calculate the probability of the occurrence of different operation techniques during the nth operation by the doctor according to the following formula:

Among them, P _i ⁿ is the occurrence probability of the i-th operation technique during the n-th operation;

The operation technique with the highest probability of occurrence is used as the technique for the nth operation by the physician.

The beneficial effects of the present invention are: because parameters such as the pressure at the contact end of the ultrasonic probe and the tissue skin during the operation of the doctor, the ultrasonic image perception, the normal vector of the ultrasonic probe and other parameters are used as the learning objects, the law of the doctor's operation technique can be learned; The data mining method of mean clustering performs cluster analysis on the operation techniques of the region of interest and the common area, which can simplify the complexity of the operation techniques; due to the use of the Markov chain model, the corresponding operation techniques of skilled physicians in the process of ultrasound scanning are analyzed. The dynamic change process of parameters is modeled by state transition to describe the doctor's scanning rules, which can effectively learn the operation methods of ultrasound doctors, and reproduce the operation methods of professional doctors in the actual scanning process, which is convenient for the automatic scanning of robot anthropomorphic ultrasound. to obtain high-quality medical ultrasound images.

Description of drawings

Fig. 1 is the flow chart of the physician's operation technique simulation method based on data mining and state learning of the present invention;

Figure 2 is a schematic diagram of the transfer of the doctor's operation technique.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments, and the present invention includes but is not limited to the following embodiments.

As shown in Figure 1, the present invention provides a method for simulating a physician's operation technique based on data mining and state learning, and its basic implementation process is as follows:

Step 1: Use the free-arm three-dimensional ultrasound imaging system to record the normal vector N of each scanning point during the ultrasound scanning process of the doctor holding the ultrasound probe in different tissue parts of the human body, and use the force sensor attached to the front of the ultrasound probe to record the scanning process. The pressure value F of the contact surface between the ultrasonic probe of each scanning point and the human tissue site is recorded as Q for the quality evaluation value of the currently scanned ultrasonic image, and the three form a doctor's manipulation characteristic sample x=(N,Q,F) ^T ;

Step 2: Use the method of step 1 for data collection, and set the collected physician manipulation data set as X={x ₁ , x ₂ ,..., x _n }, where n is the total number of samples of the manipulation data set, and each A sample x _i =(N _i ,Q _i ,Fi ) ^T , _i =1,2,...,n;

Use the fuzzy C-means clustering algorithm to cluster the dataset X, mine the distribution information in the manipulation data, and obtain two datasets, which are respectively recorded as the common area dataset X ₁ and the interest area dataset X ₂ . The statistical variance of the set X ₂ is greater than the statistical variance of the data set X ₁ , that is, if var(X ₁ )<var(X ₂ ), then it is judged that X ₁ is a common area data set, and X ₂ is a region of interest data set;

Step 3: Randomly select the subset T to be modeled from X ₂ , and count the non-repetitive operation techniques from T to form a state space S={s ₁ , s ₂ ,...,s _m }, where s _i =(N _i ,Q _i ,Fi ) ^T represents the state of the _i -th operation technique, m is the number of state values, i=1,2,...,m;

Since the doctor's scanning operation is a process of continuous change, when encountering a current image quality, the doctor will judge according to his own experience, what operation method to take at the next scanning point along the scanning path will achieve better results Effect. However, due to the difference between different scanned persons and different scanning habits of doctors, the next steps to be used will also be different. That is to say, the selection of the next operation method is based on the current operation method and is selected according to a certain probability. As shown in Figure 2, each circle represents an operation method, the arrow is the transfer direction of the method, and p _ij represents the The probability value of the i-th approach transitioning to the j-th approach. This is exactly the Markov property. Therefore, the present application adopts the Markov chain model to model the doctor's operation technique, and simulates the dynamic process of the doctor's ultrasound scanning, so as to realize the automatic scanning of the anthropomorphic robot ultrasound.

The one-step transition probability of the physician transferring from the i-th manipulation to the j-th manipulation during the ultrasound scan is calculated according to the following formula:

Among them, P(s _j |s _i ) represents the one-step transition probability from the i-th method to the j-th method, k _j represents the number of times the state is transferred to the s _j state, and K _i represents the state is transferred to itself or the next The total number of times a state; i=1,2,...,m, j=1,2,...,m;

The one-step state transition matrix P of the subset T is calculated according to the following formula:

Wherein, p _ij =P(s _j |s _i ), i=1,2,...,m, j=1,2,...,m;

The initial distribution of state _si is calculated according to the following formula:

Among them, p _i (0) represents the initial distribution of state s _i , ξ _i represents the number of times the manipulation of state s _i occurs, [T] is the number of manipulation samples in subset T, i=1,2,... .,m;

Step 4: Calculate the probability of occurrence of different operation techniques during the nth operation by the doctor according to the following formula:

Among them, P _i ⁿ is the occurrence probability of the i-th operation technique during the n-th operation;

The operation technique corresponding to the maximum value of P _i ⁿ (i=1, 2, .

Claims (1)

1. a physician operating technique simulation method based on data mining and state learning, is characterized in that the steps are as follows: Step 1: Use the free-arm three-dimensional ultrasound imaging system to record the normal vector N of each scanning point during the ultrasound scanning process of the doctor holding the ultrasound probe in different tissue parts of the human body, and use the force sensor attached to the front of the ultrasound probe to record the scanning process. The pressure value F of the contact surface between the ultrasonic probe of each scanning point and the human tissue site is recorded as Q for the quality evaluation value of the currently scanned ultrasonic image, and the three form a doctor's manipulation characteristic sample x=(N,Q,F) ^T ; Step 2: Use the method of step 1 for data collection, and set the collected physician manipulation data set as X={x ₁ , x ₂ ,..., x _n }, where n is the total number of samples of the manipulation data set, and each A sample x _i =(N _i ,Q _i ,Fi ) ^T , _i =1,2,...,n; Use the fuzzy C-means clustering algorithm to cluster the data set X, and obtain two data sets, which are recorded as the common area data set X ₁ and the interest area data set X ₂ respectively. The statistical variance of the data set X ₂ is greater than that of the data set. Statistical variance of X ₁ ; Step 3: The subset T to be modeled is randomly selected from the data set X ₂ of the region of interest, and the state space S={s ₁ , s ₂ ,...,s _m is formed by counting non-repetitive operation methods in T }, where s _i =(N _i ,Q _i ,F _i ) ^T represents the state of the i-th operation technique, m is the number of state values, i=1,2,...,m; The one-step transition probability of the physician transferring from the i-th manipulation to the j-th manipulation during the ultrasound scan is calculated according to the following formula:

Among them, P(s _j |s _i ) represents the one-step transition probability from the i-th method to the j-th method, k _j represents the number of times the state is transferred to the s _j state, and K _i represents the state is transferred to itself or the next The total number of times a state; i=1,2,...,m, j=1,2,...,m; The one-step state transition matrix P of the subset T is calculated according to the following formula:

Wherein, p _ij =P(s _j |s _i ), i=1,2,...,m, j=1,2,...,m; The initial distribution of state _si is calculated according to the following formula:

Among them, p _i (0) represents the initial distribution of state s _i , ξ _i represents the number of times the manipulation of state s _i occurs, [T] is the number of manipulation samples in subset T, i=1,2,... .,m; Step 4: Calculate the probability of occurrence of different operation techniques during the nth operation by the doctor according to the following formula:

Among them, P _i ⁿ is the occurrence probability of the i-th operation technique during the n-th operation; The operation technique with the highest probability of occurrence is used as the technique for the nth operation by the physician.

Images