CN117352164A - Multimodal tumor detection and diagnosis platform based on artificial intelligence and its processing method - Google Patents

Abstract

The invention discloses a multi-modal tumor detection and diagnosis platform based on artificial intelligence and a processing method thereof, and relates to the field of information processing technology. The functional information image feature vector of the organ is used as input, and the fusion model with the decision decision as the output fuses the obtained symptom feature vector and image feature vector, and outputs the tumor severity level decision decision to determine the tumor severity level of the patient to be detected. The neural network model based on artificial intelligence comprehensively analyzes the patient's basic information, chief complaint information and medical imaging information, thereby achieving classification and determination of tumor severity levels for patients who have developed target tumors and is to be detected, which can reduce the need for manual imaging based on Determine the probability of error and improve the accuracy of tumor diagnosis.

Description

Multimodal tumor detection and diagnosis platform based on artificial intelligence and its processing method

Technical field

The present invention relates to the field of information processing technology, and in particular to a multi-modal tumor detection and diagnosis platform based on artificial intelligence and its processing method.

Background technique

Multimodal image fusion examination technology (PET/CT+MRI) is a tumor detection technology that has developed rapidly in recent years. PET extracts metabolic information images at the molecular level of subjects, CT extracts anatomical information images of tissue structures, and MRI extracts tissue Organ functional information images integrate information from three high-resolution images with different imaging foundations, which can provide rich information resources for disease diagnosis and achieve the purpose of early detection, depth detection, and complete detection of systemic tumors.

At present, the determination of tumor severity still requires doctors to make based on their own experience and ability to recognize images. Therefore, although multi-modal image fusion examination technology can provide a good diagnostic tool for tumor detection, in the process of determining tumor severity, it is also necessary to determine the severity of tumors. It is very likely that image recognition errors will lead to errors in determining the severity of the patient's tumor, which will affect the patient's subsequent diagnosis and treatment process. To this end, we propose a multimodal tumor detection and diagnosis platform based on artificial intelligence and its processing method.

Contents of the invention

The main purpose of the present invention is to provide a multi-modal tumor detection and diagnosis platform based on artificial intelligence and its processing method, which can effectively solve the problems in the background technology.

In order to achieve the above object, the technical solution adopted by the present invention is:

Multi-modal tumor detection and diagnosis platform based on artificial intelligence, including,

The symptom information collection module is used to collect the basic information parameters and main complaint information parameters of the patients to be tested based on the information in the electronic medical record database of the medical system;

The symptom feature vector extraction module establishes a first neural network model that takes the basic information parameters and chief complaint information parameters of patients in the medical system's electronic medical record database as input, and uses the medically diagnosed tumor severity level as the output, and compares the established The first neural network model is trained, and the number of hidden layers of the model is adjusted according to the error between the training results and the actual alarm results until the accuracy is no less than the first expected value, and the patient to be detected and the target tumor are extracted through the constructed first neural network model. associated symptom feature vectors;

The imaging image acquisition module is used to obtain high-resolution images of different imaging studies, including molecular-level metabolic information images of subjects, tissue structure anatomy information images, and functional images of tissues and organs;

The image feature vector extraction module is used to construct a medical imaging library based on the high-resolution images obtained from different imaging studies. By pre-training various high-resolution imaging data from the medical imaging library as input, the second neural network model is obtained. , and train the established second neural network model, adjust the number of hidden layers of the model according to the error between the training results and the actual alarm results until the accuracy is no less than the second expected value, and extract the to-be-detected data based on the constructed second neural network model The image feature vectors related to the patient and the target tumor include molecular level metabolic information image feature vectors, tissue structure anatomy information image feature vectors, and tissue and organ functional information image feature vectors;

The feature fusion module is used to build a fusion model that takes symptom feature vectors, horizontal metabolism information image feature vectors, tissue structure anatomy information image feature vectors, and tissue and organ functional information image feature vectors as inputs, and uses decision-making as the output, and The acquired symptom feature vectors and image feature vectors are fused according to the constructed fusion model, and the tumor severity level determination decision is output;

The central detection module is used to build a tumor severity determination model and determine the tumor severity level of the patient to be tested based on the obtained tumor severity level determination decision.

The detection steps of the detection platform include:

Step 1: The symptom information collection module collects the basic information parameters and main complaint information parameters of the patient to be tested from the electronic medical record database of the medical system;

Step 2: Screen the symptom features related to the target tumor, and use the symptom feature vector extraction module to establish the basic information parameters and chief complaint information parameters of the patients in the electronic medical record database of the medical system as input, and use the medically diagnosed tumor severity level as the output The first neural network model is trained, and the number of hidden layers of the model is adjusted according to the error between the training results and the actual alarm results until the accuracy is no less than the first expected value, and through the constructed first neural network model A neural network model extracts symptom feature vectors related to the target tumor of the patient to be detected;

Step 3: Obtain high-resolution images of different imaging studies through the imaging image acquisition module, including molecular-level metabolic information images of the subject, tissue structure and anatomy information images, and functional images of tissues and organs;

Step 4: Screen the image features related to the target tumor, build a medical imaging library based on the high-resolution images of different imaging studies through the image feature vector extraction module, and use various high-resolution imaging data from the medical imaging library as input. Pre-train, obtain the second neural network model, and train the established second neural network model. Adjust the number of hidden layers of the model according to the error between the training results and the actual alarm results until the accuracy is no less than the second expected value. According to the construction The second neural network model extracts image feature vectors related to the target tumor of the patient to be detected, including molecular level metabolic information image feature vectors, tissue structure anatomy information image feature vectors, and tissue and organ functional information image feature vectors;

Step 5: The feature fusion module constructs a fusion model that takes symptom feature vectors, horizontal metabolism information image feature vectors, tissue structure anatomy information image feature vectors, and tissue and organ functional information image feature vectors as inputs, and uses decision-making as the output, and The acquired symptom feature vectors and image feature vectors are fused according to the constructed fusion model, and the tumor severity level determination decision is output;

Step 6: The central detection module builds a tumor severity determination model, and based on the obtained tumor severity level determination decision, the tumor severity level of the patient to be detected is determined and divided into any one of levels I, II, III, and IV.

Further, the diagnosis platform also includes a data storage module, which is connected to the symptom information collection module and the imaging image collection module, and is used to store the basic information parameters and main complaint information parameters of the patient to be detected and the subject. Molecular level metabolic information images, tissue structure anatomy information images and functional images of tissues and organs.

Further, the diagnostic platform also includes a processor and a computer program stored on the data storage module and executable on the processor.

The present invention has the following beneficial effects:

(1) Compared with the existing technology, the technical solution of the present invention collects the basic information parameters and chief complaint information parameters of the patients to be detected based on the information in the electronic medical record database of the medical system, and establishes the basic information parameters and information parameters of the patients who have developed the disease in the electronic medical record database of the medical system. The first neural network model takes the chief complaint information parameter as input and uses the medically diagnosed tumor severity level as output to extract the symptom feature vectors of the patient to be detected and related to the target tumor, and builds a medical imaging library based on the high-resolution images obtained with different imaging techniques. By pre-training various high-resolution imaging data from the medical imaging library as input, the second neural network model is obtained to extract image feature vectors related to the patient to be detected and the target tumor, and construct an image based on symptom feature vectors and horizontal metabolic information. Feature vectors, tissue structure anatomical information image feature vectors, and tissue and organ functional information image feature vectors are used as inputs. The fusion model with decision-making as the output fuses the obtained symptom feature vectors and image feature vectors to output the tumor severity level. Decision making determines the severity level of the tumor of the patient to be detected. The neural network model based on artificial intelligence comprehensively analyzes the patient's basic information, chief complaint information and medical imaging information, thereby realizing the diagnosis of the tumor of the patient to be detected who has developed the target tumor. Classification and determination of severity levels can reduce the probability of errors due to manual judgment based on images and improve the accuracy of tumor diagnosis.

Description of drawings

Figure 1 is a structural block diagram of the multi-modal tumor detection and diagnosis platform based on artificial intelligence of the present invention;

Figure 2 is a detection flow chart of the multi-modal tumor detection and diagnosis platform based on artificial intelligence of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the specific embodiments. The drawings are only for illustrative purposes and represent only schematic diagrams rather than actual diagrams. They cannot be understood as limitations of the present invention. In order to better illustrate the present invention, Specific embodiments, some components in the drawings may be omitted, enlarged or reduced, which do not represent the size of the actual product.

Example 1

The overall structure diagram of the multi-modal tumor detection and diagnosis platform based on artificial intelligence is shown in Figure 1, and the detection flow chart of the multi-modal tumor detection and diagnosis platform based on artificial intelligence is shown in Figure 2.

The detection steps of the detection platform include,

Step 1: The symptom information collection module collects the basic information parameters and main complaint information parameters of the patient to be tested from the electronic medical record database of the medical system. The basic information parameters include the patient's name, gender, age, occupation, marriage, ethnicity, place of origin, and work. Unit and current address; main complaint information parameters include incidence, disease process, diagnosis and treatment;

Step 2: Screen the symptom features related to the target tumor. The steps of the screening method for the symptom features related to the target tumor are:

Step 21), count the number of patients with the target tumor disease found in the electronic medical record database of the medical system;

Step 22), collect basic information parameters and main complaint information parameters of patients with the target tumor disease, and obtain various symptom characteristics in the basic information parameters and main complaint information parameters of all patients with the target tumor disease;

Step 23), use each symptom feature as a classification attribute to build a decision tree model, and calculate the incidence probability P _t of any symptom feature based on the decision tree model. The calculation formula is: Among them, N is the number of patients with the target tumor disease who have been found in the electronic medical record database; N _t is the number of patients with the k-th symptom characteristic of the target tumor that has been found in the electronic medical record database;

Step 24), after obtaining the incidence probability P _t , use the numerical value of the incidence probability P _t to create a sample set, obtain the mean and standard deviation of the sample set, and use the mean and standard deviation to standardize the data. The standardization formula is: In this formula, z is the standard parameter, σ is the variance of the sample data, and μ is the mean of the sample data. After the standardization is completed, the standard parameter is used Adjust the numerical interval to [0,1], and use the function value of f(k) to classify the coincidence rate. The classification mechanism is:

when When , the incidence probability P _t is classified as level one;

when When , the incidence probability P _t is classified as level two;

Among them, f(k)min and f(k)max are the minimum and maximum values of the function value of f(k) respectively;

Step 25), screen the symptom features corresponding to the incidence probability P _t classified as one level as symptom features related to the target tumor;

Through the symptom feature vector extraction module, a first neural network model is established that takes the basic information parameters and chief complaint information parameters of patients with the disease in the electronic medical record database of the medical system as input, and uses the medically diagnosed tumor severity level as the output, where the tumor severity The grade is determined based on the tumor evaluation index parameters. The tumor evaluation index includes the morphological index, size index, position index and surrounding tissue relationship of the tumor. It is divided into four different levels: I, II, III and IV. The first neural network model established Carry out training and adjust the number of hidden layers of the model according to the error between the training results and the actual alarm results until the accuracy is no less than the first expected value, where the calculation formula of the first expected value is,

Among them, E(Y) ₁ represents the first expected value; N ₁ represents the number of input samples of the first neural network; f(X _i ) ₁ represents the output function of the first neural network _; The i-th output sample is used to extract the symptom feature vector of the patient to be detected related to the target tumor through the constructed first neural network model;

Step 3: Obtain high-resolution images of different imaging studies through the imaging image acquisition module, including molecular-level metabolic information images of the subject, tissue structure and anatomy information images, and functional images of tissues and organs;

Step 4: Screen image features related to the target tumor. The screening method for image features related to the target tumor is,

Step 41), count the number of patients with the target tumor disease found in the medical imaging database;

Step 42), collect imaging images of patients with the target tumor disease, and obtain various imaging features in the imaging images of all patients with the target tumor disease;

Step 43), use each image feature as a classification attribute to build a decision tree model, and calculate the occurrence probability P _m of any image feature based on the decision tree model. The calculation formula is: Among them, N ₀ is the number of patients with the target tumor disease who have been found in the medical imaging database; N _m is the number of patients with the m-th symptom characteristic of the target tumor that has been found in the medical imaging database;

Step 44), after obtaining the occurrence probability P _m , use the numerical value of the occurrence probability P _m to create a sample set, obtain the mean and standard deviation of the sample set, and use the mean and standard deviation to standardize the data. The standardization formula is In this formula, z is the standard parameter, σ is the variance of the sample data, and μ is the mean of the sample data. After the standardization is completed, the standard parameter is used Adjust the numerical interval to [0,1], and use the function value of f(k) to classify the coincidence rate. The classification mechanism is:

when When , the occurrence probability P _m is classified as level one;

when When, the occurrence probability P _m is classified as level two;

Among them, f(k)min and f(k)max are the minimum and maximum values of the function value of f(k) respectively;

Step 45), screen the image features corresponding to the first-level occurrence probability P _m as image features related to the target tumor, and construct a medical imaging library based on the high-resolution images of different imaging studies obtained through the image feature vector extraction module. Pre-train various types of high-resolution imaging data from the medical imaging library as input to obtain a second neural network model, train the established second neural network model, and adjust the model based on the error between the training results and the actual alarm results. The number of hidden layers until the accuracy is no less than the second expected value, where the calculation formula of the second expected value is,

Among them, E(Y) ₂ represents the second expected value; N ₂ represents the number of input samples of the second neural network model; f(X _i ) ₂ represents the output function of the second neural network model; X _i2 represents the second neural network model For the i-th output sample, the image feature vectors related to the patient to be detected and the target tumor are extracted according to the second neural network model constructed, including image feature vectors of molecular level metabolic information, image feature vectors of tissue structure and anatomy information, and functions of tissues and organs. Learning information image feature vector;

Step 5: The feature fusion module constructs a fusion model that takes symptom feature vectors, horizontal metabolism information image feature vectors, tissue structure anatomy information image feature vectors, and tissue and organ functional information image feature vectors as inputs, and uses decision-making as the output. In the embodiment, the splicing feature fusion method without learning cost is taken as an example. Let the symptom feature vector be X1, the horizontal metabolic information image feature vector be X2, the tissue structure and anatomy information image feature vector be The information image feature vector is X4. The four are matrix stacked to form the final output feature X = [X1,

Step 6: The central detection module builds a tumor severity determination model. Taking the threshold method as an example, by setting the threshold reference range, when the feature value is within a certain threshold range, it is determined to be a certain severity level. According to the obtained tumor severity level The determination decision is to determine the tumor severity level of the patient to be detected and classify it into any one of levels I, II, III, and IV.

The basic principles and main features of the present invention and the advantages of the present invention have been shown and described above. Those skilled in the industry should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have other aspects. Various changes and modifications are possible, which fall within the scope of the claimed invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims (10)

1. The multi-mode tumor detection and diagnosis platform based on artificial intelligence is characterized by comprising,

the symptom information acquisition module is used for acquiring basic information parameters and complaint information parameters of a patient to be detected according to information of the medical system electronic medical record database;

the symptom feature vector extraction module is used for establishing a first neural network model which takes basic information parameters and complaint information parameters of a patient suffering from a disease in an electronic medical record database of a medical system as input and takes the grade of the severity of a tumor in medical diagnosis as output, training the established first neural network model, adjusting the number of hidden layers of the model to be not lower than a first expected value according to errors of a training result and an actual alarm result, and extracting symptom feature vectors related to a target tumor of the patient to be detected through the established first neural network model;

the imaging image acquisition module is used for acquiring high-resolution images of different imaging, including a molecular level metabolism information image, a tissue structure anatomical information image and a functional image of a tissue organ of a subject;

the image feature vector extraction module is used for constructing a medical image library according to the acquired high-resolution images of different imagers, obtaining a second neural network model by pre-training various high-resolution image data of the medical image library as input, training the established second neural network model, adjusting the number of hidden layers of the model to be not lower than a second expected value according to errors of a training result and an actual alarm result, and extracting image feature vectors related to a target tumor of a patient to be detected according to the established second neural network model, wherein the image feature vectors comprise molecular level metabolic information image feature vectors, tissue structure anatomical information image feature vectors and tissue organ functional information image feature vectors;

the feature fusion module is used for constructing a fusion model taking a symptom feature vector, a horizontal metabolism information image feature vector, a tissue structure anatomical information image feature vector and a tissue organ functional information image feature vector as inputs and taking a decision as output, fusing the acquired symptom feature vector and the image feature vector according to the constructed fusion model, and outputting a tumor severity level decision;

the center detection module is used for constructing a tumor severity judging model and judging the tumor severity grade of the patient to be detected according to the obtained tumor severity grade judging decision.

2. The artificial intelligence based multimodal tumor detection diagnostic platform of claim 1 wherein the first expected value and the second expected value are calculated as,

wherein E (Y) _j Represents the j thA desired value; n (N) _j Representing the number of input samples of the jth neural network; f (X) _i ) _j An output function representing a jth neural network; x is X _ij Represents the ith output sample of the jth neural network, j=1, 2.

3. The artificial intelligence based multimodal tumor detection diagnostic platform of claim 1 wherein the basic information parameters include patient name, gender, age, occupation, marital, ethnicity, native, work unit, present address; the main complaint information parameters comprise morbidity, pathological changes and diagnosis and treatment conditions.

4. The artificial intelligence based multi-modal tumor detection and diagnosis platform of claim 1 wherein the tumor severity level is determined based on tumor assessment index parameters, the tumor assessment index including a morphology index, a size index, a location index, and a surrounding tissue relationship index of the tumor.

5. The artificial intelligence based multi-modal tumor detection and diagnosis platform of claim 1 wherein the tumor severity level decision classifies the tumor severity of the patient to be detected into four different levels i, ii, iii, iv based on the tumor severity level.

6. The artificial intelligence-based multi-modal tumor detection and diagnosis platform according to claim 1, further comprising a data storage module, wherein the storage module is connected with the symptom information acquisition module and the imaging image acquisition module, and is used for storing basic information parameters and complaint information parameters of a patient to be detected, and molecular level metabolic information images, anatomical information images of tissue structures and functional images of tissue organs of a subject.

7. The artificial intelligence based multimodal tumor detection diagnostic platform of claim 1 further comprising a processor and a computer program stored on the data storage module and executable on the processor, wherein the processor is capable of performing the functions of any of the modules of claim 1 when the processor executes the program.

8. The multi-modal tumor detection and diagnosis platform based on artificial intelligence according to claim 1, wherein the screening method of symptom characteristics related to the target tumor is that,

step 1), counting the number of found target tumor patients in an electronic medical record database of a medical system;

step 2), basic information parameters and complaint information parameters of the found target tumor patients are collected, and various symptom characteristics in the basic information parameters and the complaint information parameters of all the found target tumor patients are obtained;

step 3), constructing a decision tree model by taking each symptom characteristic as a classification attribute, and calculating the probability P of occurrence under any symptom characteristic according to the decision tree model _t The calculation formula is as follows:n is the number of the patients with the target tumor found in the electronic medical record database; n (N) _t The number of patients with the kth symptom characteristic of the occurrence of the target tumor is found in the electronic medical record database;

step 4), obtaining the incidence probability P _t Then, the probability of occurrence P is used _t A sample set is created, the mean value and the standard deviation in the sample set are obtained, the data are standardized by the mean value and the standard deviation, and the standardized formula is thatIn the formula, z is a standard parameter, sigma is the variance of sample data, mu is the mean value of the sample data, and after normalization is completed, the standard parameter is utilizedAdjust the value interval to [0 ],1]The coincidence rate is classified by using the function value of f (k), and the classification mechanism is as follows:

when (when)Probability of onset P _t Classifying into a first class;

when (when)Probability of onset P _t Classifying into a second stage;

wherein f (k) min and f (k) max are the minimum and maximum values of the function values of f (k), respectively;

step 5), screening the incidence probability P classified as first order _t The corresponding symptom characteristic is a symptom characteristic related to the target tumor.

9. The multi-modal tumor detection and diagnosis platform based on artificial intelligence according to claim 1, wherein the screening method of image features related to target tumor is that,

step 1), counting the number of found target tumor patients in a medical image library;

step 2), acquiring imaging images of the found target tumor patients, and acquiring various image features in the imaging images of all the found target tumor patients;

step 3), constructing a decision tree model by taking each image feature as a classification attribute, and calculating the occurrence probability P under any image feature according to the decision tree model _m The calculation formula is as follows:wherein N is ₀ The number of the patients with the target tumor found in the medical image library; n (N) _m The number of patients characterized by the m-th symptom of the occurrence of the target tumor in the medical image library;

step 4), obtaining the occurrence probability P _m Thereafter, the probability of occurrence P is utilized _m Creating a sample set of values of (a) and obtaining the sample setThe data is normalized by the mean and standard deviation, and the normalization formula isIn the formula, z is a standard parameter, sigma is the variance of sample data, mu is the mean value of the sample data, and after normalization is completed, the standard parameter is utilizedAdjust the value interval to [0,1 ]]The coincidence rate is classified by using the function value of f (k), and the classification mechanism is as follows:

when (when)Probability of occurrence P _m Classifying into a first class;

when (when)Probability of occurrence P _m Classifying into a second stage;

wherein f (k) min and f (k) max are the minimum and maximum values of the function values of f (k), respectively;

step 5), screening the occurrence probability P classified as one level _m The corresponding image features are those associated with the target tumor.

10. The artificial intelligence based multi-modality tumor detection and diagnosis platform of claim 1, wherein the detecting step of the detection platform includes,

the method comprises the steps that firstly, a symptom information acquisition module acquires basic information parameters and complaint information parameters of a patient to be detected from an electronic medical record database of a medical system;

screening symptom characteristics related to a target tumor, establishing a first neural network model which takes basic information parameters and main complaint information parameters of a patient suffering from the disease in an electronic medical record database of a medical system as input and the severity level of the tumor in medical diagnosis as output through a symptom characteristic vector extraction module, training the established first neural network model, adjusting the number of hidden layers of the model to be not lower than a first expected value according to the training result and the error of an actual alarm result, and extracting symptom characteristic vectors related to the target tumor of the patient to be detected through the established first neural network model;

step three, obtaining high-resolution images of different imageology through an imaging image acquisition module, wherein the high-resolution images comprise a molecular level metabolism information image, a tissue structure anatomy information image and a functional image of a tissue organ of a subject;

screening image features related to a target tumor, constructing a medical image library according to the acquired high-resolution images of different imagers through an image feature vector extraction module, pre-training various high-resolution image data of the medical image library as input to obtain a second neural network model, training the established second neural network model, adjusting the number of hidden layers of the model to be not lower than a second expected value according to errors of a training result and an actual alarm result, and extracting image feature vectors related to the target tumor of a patient to be detected according to the constructed second neural network model, wherein the image feature vectors comprise molecular level metabolic information image feature vectors, tissue structure anatomical information image feature vectors and tissue organ functional information image feature vectors;

step five, a feature fusion module constructs a fusion model taking symptom feature vectors, horizontal metabolism information image feature vectors, tissue structure anatomical information image feature vectors and functional information image feature vectors of tissues and organs as input and taking decision as output, fuses the acquired symptom feature vectors and the image feature vectors according to the constructed fusion model, and outputs a tumor severity level decision;

step six, the central detection module builds a tumor severity judging model, and judges and classifies the tumor severity grade of the patient to be detected into any one grade of I, II, III and IV according to the obtained tumor severity grade judging decision.