Medical image curative effect evaluation system based on artificial intelligence

Abstract

The invention discloses a medical image curative effect evaluation system based on artificial intelligence, and the system is characterized in that an acquisition module obtains a clinical vascular medical image, builds an image database according to the clinical vascular medical image, and builds a high-resolution vascular three-dimensional image according to the image data in the image database; a positioning recognition module receives the blood vessel three-dimensional image and inputs the blood vessel three-dimensional image into a trained blood vessel positioning model, and the blood vessel positioning model automatically recognizes the blood vessel position and level in the blood vessel three-dimensional image; the evaluation module is used for cutting and calculating the blood vessel three-dimensional image according to the recognized blood vessel position and level and a preset size, and evaluating whether the lesion position and lesion type of the blood vessel with blood vessel stenosis, blood vessel dilatation and blood vessel thickening occur or not according to a calculation result; the technical problem that the prior art is difficult to accurately evaluate the curative effect of vascular diseases is solved.

Description

Technical Field

[0001] This application relates to the technical field of medical efficacy evaluation, and particularly to a medical image efficacy evaluation system based on artificial intelligence. Background Art

[0002] Nowadays, vascular diseases occur more and more frequently, and it is difficult to accurately evaluate the efficacy in the treatment of vascular diseases.

[0003] Existing technologies use methods such as CTA, contrast-enhanced vascular ultrasound, PET-CT, and pathology to observe the changes before and after the treatment of vascular diseases, which are commonly used clinical imaging follow-up methods for vascular diseases. However, these methods have certain limitations, that is, a large amount of time is required for imaging evaluation, and they cannot accurately reflect the changes before and after the treatment of vascular diseases, making it difficult to evaluate the efficacy of vascular diseases. Summary of the Invention

[0004] An embodiment of this application provides a medical image efficacy evaluation system based on artificial intelligence, which is used to solve the technical problems that a large amount of time is required for the prior art to evaluate the treatment before and after vascular diseases, and it cannot accurately reflect the changes before and after the treatment of vascular diseases, making it difficult to evaluate the efficacy of vascular diseases.

[0005] In view of this, this application provides a medical image efficacy evaluation system based on artificial intelligence, including an acquisition module, a positioning and recognition module, and an evaluation module;

[0006] The acquisition module is used to obtain clinical vascular medical images, form an image database according to the clinical vascular medical images, and then construct a high-resolution three-dimensional vascular image based on the image data in the image database;

[0007] The positioning and recognition module is used to receive the three-dimensional vascular image and input the three-dimensional vascular image into a trained vascular positioning model, and the vascular positioning model automatically recognizes the vascular position and hierarchy in the three-dimensional vascular image;

[0008] The evaluation module is used to cut and calculate the three-dimensional vascular image according to the recognized vascular position and hierarchy according to a preset size, and evaluate the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, and vascular thickening according to the calculation results.

[0009] Optionally, the evaluation of the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, and vascular thickening according to the calculation results specifically includes:

[0010] Cut the three-dimensional vascular image into equal-sized segments according to the vascular position and hierarchy according to a preset size to obtain five adjacent consecutive vascular segments;

[0011] Then, calculate the average lumen diameter and wall thickness of the five consecutive vascular segments through imaging methods, compare the average lumen diameter and wall thickness of adjacent consecutive five vascular segments, take the opening from the left ventricle to the ascending aorta as the reference value, and calculate the change rate of the average lumen diameter and wall thickness.

[0012] Set different change rate thresholds for vascular stenosis, vascular dilation, and vascular thickening. If the calculated change rate of the average lumen diameter and wall thickness is greater than the corresponding change rate threshold, it is judged and evaluated that vascular stenosis, vascular dilation, or vascular thickening has occurred, and then output the lesion location and lesion type of the blood vessel with vascular stenosis, vascular dilation, or vascular thickening.

[0013] Optionally, the three-dimensional vascular image is a high-resolution cross-sectional image of MRA vascular scan.

[0014] Optionally, the vascular positioning model adopts a neural network algorithm.

[0015] Optionally, the positioning and recognition module is further configured to output the lesion degree, and the lesion degree is reflected by the difference between the change rate obtained by real-time calculation and the change rate threshold.

[0016] Optionally, it further includes a comprehensive evaluation module, which is configured to screen indicators according to the medical and biological information indicators of the patient, optimize and combine the screened indicators with the imaging findings output by the positioning and recognition module, then automatically analyze the vascular image and vascular pathology according to the optimized combination, and evaluate the nature of the vascular wall lesion according to the analysis result.

[0017] Optionally, it further includes a trend prediction module, which is configured to perform the same processing on the data before and after treatment of the same patient, calculate the change rate of the vascular wall thickness and vascular lumen diameter at the same part before and after treatment, and evaluate the change of the blood vessel based on the preset thresholds for the change of the vascular lumen and vascular wall.

[0018] Optionally, the acquisition module includes a medical image processing module, and the medical image processing module includes a training module, a structuring module, and an information extraction module;

[0019] The training module is configured to establish a medical image natural language model, and construct training data and train a loss function based on image anatomy and diagnostics;

[0020] The structuring module is configured to generate a report structuring template based on the natural language model and adaptive ontology learning;

[0021] The information extraction module is used to perform policy-based automated hierarchical information extraction and content verification on the input report text according to the report structured template.

[0022] As can be seen from the above technical solutions, the embodiments of the present application have the following advantages:

[0023] A medical image efficacy evaluation system based on artificial intelligence provided by the present application, compared with the prior art, the present invention acquires clinical vascular medical images through an acquisition module, forms an image database according to the clinical vascular medical images, and then constructs a high-resolution three-dimensional vascular image based on the image data in the image database; the positioning and recognition module receives the three-dimensional vascular image and inputs the three-dimensional vascular image into a trained vascular positioning model, and the vascular positioning model automatically recognizes the vascular position and hierarchy in the three-dimensional vascular image; the evaluation module cuts and calculates the three-dimensional vascular image according to the recognized vascular position and hierarchy according to a preset size, and evaluates the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, and vascular thickening according to the calculation results, realizing automatic processing after the vascular image is extracted, intelligently evaluating the efficacy changes before and after the treatment of vascular diseases, saving the evaluation time, improving the evaluation efficiency of doctors, and improving the accuracy of the evaluation.

[0024] The present application also establishes a medical image natural language model through a training module, constructs training data and trains a loss function based on image anatomy and diagnostics; a structuring module generates a report structured template based on the natural language model and adaptive ontology learning; an information extraction module, which is used to perform policy-based automated hierarchical information extraction and content verification on the input report text according to the report structured template, realizes the automatic generation of a tree-structured template of the report, realizes the structured decomposition of the report text in the image data of vascular disease patients, and thus helps medical institutions to expand in functions such as medical information retrieval, analysis, evaluation, and artificial intelligence learning, thereby indirectly reducing the work intensity of medical staff and improving the operation efficiency of medical institutions. Brief Description of the Drawings

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0026] Figure 1 It is a system block diagram of a medical image efficacy evaluation system based on artificial intelligence provided in the embodiments of the present application. Detailed Embodiments

[0027] To enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application.

[0028] For ease of understanding, please refer to Figure 1 , an embodiment of a medical image efficacy evaluation system based on artificial intelligence provided by this application, includes an acquisition module, a positioning and recognition module, and an evaluation module; the acquisition module is used to obtain clinical vascular medical images, form an image database according to the clinical vascular medical images, and then construct a high-resolution three-dimensional vascular image based on the image data in the image database. The clinical vascular medical images include MRI images, and the three-dimensional vascular image is a high-resolution cross-sectional image of MRA vascular scanning.

[0029] The positioning and recognition module is used to receive the immunohistochemical pathological images in the three-dimensional vascular image and the corresponding clinical vascular medical images, and input the three-dimensional vascular image into a trained vascular positioning model. The vascular positioning model automatically recognizes the vascular position and layer in the three-dimensional vascular image. Among them, the process of constructing a training set when training the vascular positioning model includes: obtaining the three-dimensional vascular image, marking the vascular position on the level of systematic anatomy based on characteristic vascular anatomical segmentation points, and marking the intima, media, and adventitia layers of the blood vessels according to the differences in the three-layer membrane structure on the level of local anatomy to obtain a training set; using the training set to repeatedly train the vascular positioning model to optimize the recognition and positioning ability of the vascular positioning model; the vascular positioning model adopts a neural network algorithm.

[0030] The evaluation module is used to cut and calculate the three-dimensional vascular image according to the recognized vascular position and layer according to a preset size, and evaluate whether there are vascular stenosis, vascular dilation, and the lesion position and lesion type of the blood vessels with vascular thickening according to the calculation results; realize automatic processing after the vascular image is extracted, intelligently evaluate the efficacy changes before and after the treatment of vascular diseases, save the evaluation time, improve the evaluation efficiency of doctors, and improve the accuracy of the evaluation.

[0031] As a further solution of the present invention, the lesion location and lesion type of the blood vessel for evaluating whether blood vessel stenosis, blood vessel dilation, and blood vessel thickening occur according to the calculation result specifically include: performing equal-size cutting on the three-dimensional blood vessel image according to a preset size according to the blood vessel position and layer to obtain five adjacent consecutive blood vessel segments; then calculating the average lumen diameter and the average wall thickness of the five consecutive blood vessel segments by an imaging method, comparing the average lumen diameter and the average wall thickness of the five adjacent consecutive blood vessel segments, using the opening from the left ventricle to the ascending aorta as a reference value, and calculating the change rate of the average lumen diameter and the average wall thickness; setting different change rate thresholds for blood vessel stenosis, blood vessel dilation, and blood vessel thickening, and if the calculated change rate of the average lumen diameter and the average wall thickness is greater than the corresponding change rate threshold, it is determined and evaluated that blood vessel stenosis, blood vessel dilation, or blood vessel thickening has occurred, and then the lesion location and lesion type of the blood vessel with blood vessel stenosis, blood vessel dilation, or blood vessel thickening are output.

[0032] Further, the positioning and recognition module is further configured to output the lesion degree, and the lesion degree is reflected by the difference between the change rate obtained by real-time calculation and the change rate threshold.

[0033] Further, a comprehensive evaluation module is further included. The comprehensive evaluation module is configured to perform index screening according to the medical and biological information indexes of the patient, optimize and combine the screened indexes with the imaging manifestations output by the positioning and recognition module, then automatically analyze the blood vessel image and blood vessel pathology according to the optimized combination, and evaluate the lesion nature of the blood vessel wall according to the analysis result.

[0034] Further, a trend prediction module is further included. The trend prediction module is configured to perform the same processing on the data before and after treatment of the same patient, calculate the change rate of the blood vessel wall thickness and the blood vessel lumen diameter at the same part before and after treatment, and evaluate the change of the blood vessel based on the preset thresholds for the change of the blood vessel lumen and the blood vessel wall.

[0035] Specifically, the trend prediction unit performs the same processing on the data before and after treatment of the same patient, calculates the change rate of the wall thickness and the lumen diameter at the same part before and after treatment, sets the thresholds for the change of the lumen and the wall (for example, if the lumen diameter decreases by more than 25%, stenosis progression is considered), and evaluates the change of the blood vessel. The trend prediction unit then combines the medical and biological information of the patient, and uses screening factors affecting the blood vessel prognosis such as regression models and decision tree models to construct a prediction model. For new cases, input the data included in the analysis, and the probability of this prognosis can be obtained, and then the blood vessel prognosis can be accurately evaluated.

[0036] Furthermore, the acquisition module includes a medical image processing module, and the medical image processing module includes a training module, a structuring module, and an information extraction module; the training module is used to establish a medical image natural language model, construct training data based on image anatomy and diagnostics, and train a loss function; the structuring module is used to generate a report structuring template based on the natural language model and adaptive ontology learning; the information extraction module is used to perform policy-based automated hierarchical information extraction and content verification on the input report text according to the report structuring template, realize the automated generation of a tree-like structuring template for the report, and realize the structured decomposition of the report text in the image data of vascular disease patients, thereby contributing to the expansion of functions such as medical information retrieval, analysis, evaluation, and artificial intelligence learning in medical institutions, thus indirectly reducing the work intensity of medical staff and improving the operation efficiency of medical institutions.

[0037] Specifically, the training module adds a cosine similarity loss to the input text features according to the association relationship of the image description, research conclusion, and diagnostic report triple or default binary group to constrain the consistency of its associated description; constructs a pre-training fine-tuning training framework integrating book literature and image reports through sample ratio and loss weight design; the structuring module uses the natural language model to complete word segmentation and feature extraction of all information of the nodes; then performs frequency statistics and sorting on all word segments; clusters all word segments by setting a similarity threshold; mines the next-level node names according to frequency topk and similarity topk, and updates the anatomy ontology; then selects some nodes after topk, and updates the anatomy ontology through annotation and confirmation by annotators; the information extraction module generates policy-based retrieval keywords during information extraction and extracts information based on the report text segment of the parent node. Among them, the content verification in the information extraction module sets inspection policies for the retrieved content of the current node, including: physical description; whether the truth of the current node information is included in the information of the parent and grandparent nodes; the current node information does not contain any retrieval word characters, and this information has already appeared in the information of other sibling nodes, and other sibling label characters appear in this information; if the information appears in multiple nodes, determine whether to retain the node information according to whether retrieval word characters appear and the priority of the nodes, realize the structured decomposition of the report text in the image data of vascular disease patients, contribute to the expansion of functions such as medical information retrieval, analysis, evaluation, and artificial intelligence learning in medical institutions, thus indirectly reducing the work intensity of medical staff and improving the operation efficiency of medical institutions.

[0038] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An artificial intelligence-based medical image treatment effect evaluation system, characterized in that It includes a collection module, a positioning and recognition module, and an evaluation module; The collection module is used to obtain clinical vascular medical images, form an image database based on the clinical vascular medical images, and then construct a high-resolution three-dimensional vascular image according to the image data in the image database; The positioning and recognition module is used to receive the three-dimensional vascular image and input the three-dimensional vascular image into a trained vascular positioning model, and the vascular positioning model automatically recognizes the vascular position and hierarchy in the three-dimensional vascular image; The evaluation module is used to cut and calculate the three-dimensional vascular image according to the recognized vascular position and hierarchy according to a preset size, and evaluate the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, and vascular thickening according to the calculation results.

2. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, characterized in that The evaluation of the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, and vascular thickening according to the calculation results specifically includes: Cut the three-dimensional vascular image into equal-sized segments according to the vascular position and hierarchy according to a preset size to obtain five adjacent consecutive vascular segments; Then calculate the average lumen diameter and wall thickness of the five consecutive vascular segments by imaging methods, compare the average lumen diameter and wall thickness of the five adjacent consecutive vascular segments, and use the opening from the left ventricle to the ascending aorta as the reference value to calculate the change rate of the average lumen diameter and wall thickness; Set different change rate thresholds for vascular stenosis, vascular dilation, and vascular thickening. If the calculated change rate of the average lumen diameter and wall thickness is greater than the corresponding change rate threshold, it is judged and evaluated that vascular stenosis, vascular dilation, or vascular thickening has occurred, and then the lesion position and lesion type of the blood vessels with vascular stenosis, vascular dilation, or vascular thickening are output.

3. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, characterized in that, The three-dimensional vascular image is a high-resolution cross-sectional MRA vascular scan image.

4. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, characterized in that, The vascular positioning model uses a neural network algorithm.

5. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, characterized in that, The positioning and recognition module is also used to output the lesion degree, and the lesion degree is reflected by the difference between the change rate calculated in real time and the change rate threshold.

6. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, wherein It also includes a comprehensive evaluation module, which is used to screen indicators according to the patient's medical and biological information indicators, optimize the combination of the selected indicators and the imaging manifestations output by the positioning and recognition module, then automatically analyze the vascular image and vascular pathology according to the optimized combination, and evaluate the nature of the vascular wall lesion according to the analysis results.

7. An artificial intelligence-based medical image efficacy evaluation system according to claim 1, characterized in that, It also includes a trend prediction module, which is used to perform the same processing on the data before and after treatment of the same patient, calculate the change rate of the vascular wall thickness and vascular lumen diameter before and after treatment at the same site, and evaluate the change of the blood vessels based on the preset thresholds for vascular lumen and vascular wall changes.

8. The medical image efficacy evaluation system based on artificial intelligence according to claim 1, wherein The collection module includes a medical image processing module, and the medical image processing module includes a training module, a structuring module, and an information extraction module; The training module is used to establish a natural language model for medical images, and construct training data and train a loss function based on image anatomy and diagnostics; The structured module is used to generate a report structured template based on the natural language model and adaptive ontology learning; The information extraction module is used to perform policy-based automated step-by-step information extraction and content verification on the input report text according to the report structured template.

Images