AI-assisted valvular disease assessment system

Abstract

The invention discloses an AI assisted valvular disease assessment system. According to the system, the AI image optimization technology is utilized, the image quality can be improved under different acoustic window conditions, and high-quality image acquisition is guaranteed. A heart standard section is automatically identified through an intelligent section identification function, and an operator is prompted to obtain an optimal measurement view angle. After a proper tangent plane is obtained, the parameter automatic measurement function can automatically measure the shape, the structure and hemodynamic parameters (including the valve peak flow rate, the trans-valve pressure difference, the reflux area and the like) of the valve. Besides, according to the automatic grading evaluation function, an evaluation conclusion is output by automatically applying a clinical grading standard according to a measurement result, and when the result is inconsistent, an operator can be prompted to review, so that the efficiency and the accuracy of valvular disease evaluation are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer-aided diagnosis, and in particular relates to an AI-assisted valvular disease assessment system. Background Art

[0002] Heart valve disease has a high incidence in clinical practice and seriously threatens the health and life of patients. Echocardiography plays a vital role in the diagnosis and severity assessment of heart valve disease.

[0003] However, the traditional echocardiography process faces many challenges. On the one hand, due to the great individual differences among patients, factors such as the physical condition and physiological structure of different patients will have a significant impact on the quality of echocardiographic images. For example, obese patients or patients with special chest structures have poor acoustic window conditions, and the images obtained may have problems such as high noise and low clarity, which increases the difficulty of image acquisition and analysis.

[0004] On the other hand, when evaluating valvular diseases (including aortic stenosis, regurgitation, mitral stenosis, regurgitation and other types), multiple indicators need to be accurately measured. These indicators involve multiple aspects such as valve morphology, structure and hemodynamic parameters, such as valve peak flow velocity, transvalvular pressure gradient, regurgitation area, etc. Moreover, it currently mainly relies on operators to manually grade the disease according to clinical standards. This process is extremely complex and time-consuming, and requires extremely high professional skills and experience from the operator. The operator needs to undergo long-term training and a lot of practice to ensure the accuracy of the evaluation. Even so, it is difficult to avoid errors caused by human factors.

[0005] Therefore, the present invention provides an AI-assisted valvular disease assessment system to solve the problems existing in the prior art. Summary of the invention

[0006] The problem to be solved by the present invention is to provide an AI-assisted valvular disease assessment system in view of the deficiencies in the prior art. The present invention can significantly improve the efficiency of valvular disease assessment in echocardiography through intelligent processing of AI, reduce the errors and time consumption of manual operation, and improve the accuracy and consistency of diagnosis. It not only optimizes the medical service process, but also makes medical resources more effectively utilized, which is of great value for the early diagnosis and treatment of patients with heart disease.

[0007] The present invention provides an AI-assisted valvular disease assessment system, comprising:

[0008] AI-assisted image quality enhancement module, automatic recognition module of standard cardiac sections, automatic measurement tool module of valve parameters, automatic classification logic and output module of disease status, and intelligent prompt and interaction module for special situations.

[0009] Preferably, the AI-assisted image quality enhancement module uses data enhancement technology during training, including random rotation, flipping, scaling and adding noise; the formula for random rotation is:

[0010] I rotated =Rotate(I original ,θ) where, I rotated is the rotated image, I original is the original image, θ is a randomly selected rotation angle;

[0011] The formula for flipping is:

[0012] I flipped =Flip(I original ) Among them, I flipped is the flipped image, I original is the original image;

[0013] The formula for scaling is:

[0014] I scaled =Scale(I original ,s) where I scaled is the scaled image, s is the scaling factor;

[0015] The formula for adding noise is:

[0016] I noisy =I original +n where, I noisy is the image after adding noise, and η is the added noise.

[0017] Preferably, the cardiac standard section automatic recognition module adopts a deep neural network DNN, which combines convolutional neural network CNN feature extraction and recursive neural network RNN ​​sequence processing; the convolutional neural network CNN feature extraction formula is: F=CNN(I), where F is the extracted feature and I is the input cardiac ultrasound image; the recursive neural network RNN ​​sequence processing formula is: S=RNN(F), where S is the processed sequence information and F is the feature sequence extracted from CNN; the formula for standard section recognition is: Y=FC(S), where Y is the recognition result and FC is a fully connected layer used to convert the output of RNN into the final recognition result.

[0018] Preferably, the valve parameter automatic measurement tool module adopts the Canny edge detection algorithm with a dual threshold strategy to dynamically adjust the threshold according to the local grayscale change and gradient information of the image; at the same time, in the shape model fitting process, an elastic deformation model is introduced to allow physiological variation of the valve shape within a certain range; the gradient amplitude calculation formula is:

[0019] Among them, G x (x,y) and G y (x, y) are the gradient magnitudes of the image in the x and y directions respectively;

[0020] The gradient direction calculation formula is:

[0021]

[0022] The dual threshold strategy formula is:

[0023] Among them, T1 is the low threshold and T2 is the high threshold;

[0024] The formula of elastic deformation energy function is:

[0025] Where u is the displacement field, Ω is the deformation area, λ and μ are Lamé parameters, and E is the Green-Lagrange strain tensor;

[0026] The strain energy function formula is:

[0027] W=c1(I1-3) 3 + c 2(I2-3) 2 +c3(I1-3)+c4(e Q -1) where c1, c2, c3, c4 are material coefficients, I1 and I2 are the first and second principal invariants, and Q is the Cauchy-Green tensor.

[0028] Preferably, the disease state automatic classification logic and output module adopts fuzzy membership function and fuzzy rules; the fuzzy membership function formula is:

[0029] Among them, a, b and c are the parameters that define the fuzzy set of the linguistic variable A, and n is the shape parameter;

[0030] The fuzzy rule formula is:

[0031] R i :IFx1is A i1 AND x2is A i2 THEN y is B i Among them, R i is the i-th fuzzy rule, x1 and x2 are input variables, A i1 and A i2 is the fuzzy set corresponding to x1 and x2, y is the output variable, B i is the fuzzy set corresponding to y.

[0032] Preferably, the special situation intelligent prompt and interaction module can be integrated with the hospital information system HIS and the medical image archiving and communication system PACS; the information retrieval formula is:

[0033] I=HIS∪PACS, where I is the collection of complete medical records, previous examination reports, and image data of the patient retrieved from HIS and PACS;

[0034] The special case identification formula is: S = IdentifySpecialCases (I) where S is the set of identified special cases, and IdentifySpecialCases is a function used to analyze the information set I and identify special cases;

[0035] The formula for generating smart tips is: T = GenerateTips(S) where T is the set of smart tips generated for the operator, and GenerateTips is a function that generates corresponding tips based on the identified special case S;

[0036] The formula for multi-language support is: L=UserLanguage, where L is the user's language preference, which is used to ensure that the interactive interface supports multi-language display. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is the system architecture diagram;

[0039] Figure 2 It is a work flow chart;

[0040] Figure 3 This is a schematic diagram of the key operation interface. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be described clearly and completely in conjunction with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Example 1

[0043] The embodiment of the present invention provides an AI-assisted valvular disease assessment system, which specifically includes:

[0044] For the convolutional neural network (CNN) in the AI-assisted image quality enhancement module, we first obtain the model parameters pre-trained on a large natural image dataset (. These parameters contain rich knowledge of image feature extraction. Then, we collect a large number of cardiac ultrasound image datasets, including ultrasound images of patients of different ages, genders, and conditions, covering images under various acoustic window conditions, as fine-tuning training data.

[0045] During training, the pre-trained model parameters are loaded into the CNN as initialization values. Next, data augmentation techniques are applied to the cardiac ultrasound image dataset. For example, each image is randomly rotated (angle range is -30° to 30°), flipped horizontally and vertically, scaled (scaling factor is between 0.8 and 1.2), and a certain degree of Gaussian noise is added (mean is 0, standard deviation is between 0.01 and 0.05). In this way, the diversity of the training data is greatly increased.

[0046] In actual echocardiography, when a new ultrasound image of the heart valve is acquired, the trained AI-assisted image quality enhancement module can process the image in real time. For situations with poor acoustic window conditions, blurry and noisy images (such as images of obese patients), the module can effectively remove noise, enhance the edges and details of the image, and significantly improve the image quality to meet the image quality requirements of subsequent analysis.

[0047] Example 2

[0048] The embodiment of the present invention provides an AI-assisted valvular disease assessment system, which specifically includes:

[0049] The deep neural network DNN in the automatic recognition algorithm of standard cardiac sections adopts a hybrid architecture of CNN and RNN. For the CNN part, the appropriate convolution kernel size and number of layers are designed to effectively extract the local spatial features of cardiac ultrasound images, such as the texture and shape features of structures such as heart valves and myocardium. For the RNN part, the appropriate number of neurons and connection methods are set according to the sequence information characteristics of the cardiac structure in the ultrasound image.

[0050] In actual applications, ultrasound images processed by the AI-assisted image quality enhancement module are input into the standard cardiac section automatic recognition algorithm. For example, when examining a patient suspected of valvular heart disease, the algorithm first uses CNN to extract local spatial features in the image, and then RNN processes the sequence information of these features to capture the contextual relationship and dynamic changes between different regions.

[0051] The test was conducted on 100 cardiac ultrasound images of different patients and compared with standard sections manually annotated by experienced ultrasound physicians. The results showed that for complex cardiac structures and dynamically changing image scenes, the algorithm's recognition accuracy for standard cardiac sections reached over 90%, and it was able to accurately identify key standard sections such as the mitral valve horizontal short-axis section and the aortic valve long-axis section, providing an accurate perspective for subsequent measurements.

[0052] Example 3

[0053] The embodiment of the present invention provides an AI-assisted valvular disease assessment system, which specifically includes:

[0054] In the valve parameter automatic measurement tool set, for the improved Canny edge detection algorithm, the threshold is dynamically adjusted according to the local grayscale changes and gradient information of the image. When measuring the valve edge, ultrasound images with different lighting conditions and image quality are processed. For example, when analyzing the ultrasound image of a patient with mitral regurgitation, the algorithm automatically adjusts the threshold according to the grayscale differences and gradient changes of the tissue around the valve in the image, and accurately detects the edge of the mitral valve.

[0055] In the process of shape model fitting, the elastic deformation model is introduced. Taking the aortic valve as an example, considering the physiological variation of the aortic valve of different individuals, the elastic deformation model allows the valve shape to vary within a certain range. By learning and modeling aortic valve images of a large number of different patients, when measuring the aortic valve morphological parameters (such as leaflet length, width, etc.) of new patients, the actual shape of the valve can be fitted more accurately, improving the accuracy of the measurement.

[0056] In the measurement of valve morphology and structural parameters of 50 patients with valvular heart disease, the measurement results of this automated measurement toolset had an error of less than 5% compared with traditional manual measurement methods. The measurement of hemodynamic parameters (such as valve peak flow velocity, transvalvular pressure gradient, regurgitant area, etc.) can also reach a clinically acceptable high precision range.

[0057] Example 4

[0058] The embodiment of the present invention provides an AI-assisted valvular disease assessment system, which specifically includes:

[0059] The fuzzy membership function and fuzzy rules of the fuzzy logic classifier in the disease state automatic classification logic and output system are constructed based on a large amount of clinical case data (more than 1,000 cases) and expert experience. These data cover various types and severity of heart valve diseases. For example, for the classification of mitral stenosis, the corresponding fuzzy membership function and fuzzy rules are determined according to the different ranges of parameters such as valve area and blood flow velocity.

[0060] In actual applications, the system continuously collects new clinical case data. When encountering new case data or expert feedback, the system automatically adjusts the fuzzy membership function and fuzzy rules. For example, when it is found that the manifestation of valvular disease in a patient group in a certain area is somewhat different from the existing grading standards, the system will adjust according to the new case data to better adapt to the actual situation.

[0061] 200 patients with heart valve disease were diagnosed and graded, and the system's output of the classification results was compared with the diagnosis results of the expert committee. The results showed that the system's classification accuracy reached 92%, and it could accurately output the disease type (such as aortic stenosis, mitral regurgitation, etc.), classification results (mild, moderate, severe) and description of the condition, providing a reliable basis for clinical treatment.

[0062] Example 5

[0063] The embodiment of the present invention provides an AI-assisted valvular disease assessment system, which specifically includes:

[0064] Intelligent prompts and interactive interfaces for special situations are seamlessly integrated with the hospital information system HIS and the medical image archiving and communication system PACS. In daily examinations, when examining a patient with valvular heart disease, the system can automatically obtain the patient's basic information, medical history, medication status, etc. from HIS, and obtain the patient's previous ultrasound images, examination reports and other information from PACS.

[0065] During the evaluation process, the measurement results and classification are monitored in real time. For example, when there are special situations such as abnormal measurement parameters (such as a sudden increase in the valve regurgitation area but no significant changes in other parameters) or the classification results do not match the patient's medical history (such as the patient's condition is stable but the classification results show a sudden worsening of the condition), the system pops up a warning window on the interactive interface.

[0066] The warning window displays a detailed description of the abnormal situation, possible cause analysis (such as measurement error, disease change, etc.), and provides relevant patient complete medical history information, previous examination reports and links to image data. The interactive interface supports multi-language display (such as English, Chinese, French, etc.), which is convenient for medical staff with different language backgrounds to operate. Operators can make further reviews and diagnostic decisions based on the prompts, such as re-measurement, adjustment of parameters or requesting expert consultation.

Claims

1. An AI-assisted valvular disease assessment system, characterized in that: include: AI-assisted image quality enhancement module, automatic recognition module of standard cardiac sections, automatic measurement tool module of valve parameters, automatic classification logic and output module of disease status, and intelligent prompt and interaction module for special situations.

2. The AI-assisted valvular disease assessment system according to claim 1, characterized in that: The AI-assisted image quality enhancement module uses data enhancement techniques during training, including random rotation, flipping, scaling, and adding noise; the formula for random rotation is: I rotated =Rotate(i original , θ) where, I rotated is the rotated image, I original is the original image, θ is a randomly selected rotation angle; The formula for flipping is: I flipped =Flip(I original ) Among them, I flipped is the flipped image, I original is the original image; The formula for scaling is: I scaled =Scale(I original , s) where I scaled is the scaled image, s is the scaling factor; The formula for adding noise is: I noisy =I original +n where, I noisy is the image after adding noise, and η is the added noise.

3. The AI-assisted valvular disease assessment system according to claim 1, characterized in that: The cardiac standard section automatic recognition module adopts a deep neural network DNN, which combines convolutional neural network CNN feature extraction and recurrent neural network RNN ​​sequence processing; the convolutional neural network CNN feature extraction formula is: F = CNN (I), where F is the extracted feature and I is the input cardiac ultrasound image; the recurrent neural network RNN ​​sequence processing formula is: S = RNN (F), where S is the processed sequence information and F is the feature sequence extracted from CNN; the formula for standard section recognition is: Y = FC (S), where Y is the recognition result and FC is a fully connected layer used to convert the output of RNN into the final recognition result.

4. The AI-assisted valvular disease assessment system according to claim 1, characterized in that: The valve parameter automatic measurement tool module adopts the Canny edge detection algorithm with a dual threshold strategy to dynamically adjust the threshold according to the local grayscale change and gradient information of the image; at the same time, in the shape model fitting process, an elastic deformation model is introduced to allow physiological variation of the valve shape within a certain range; the gradient amplitude calculation formula is: Among them, G x (x,y) and G y (x, y) are the gradient magnitudes of the image in the x and y directions respectively; The gradient direction calculation formula is: The dual threshold strategy formula is: Among them, T1 is the low threshold and T2 is the high threshold; The formula of elastic deformation energy function is: Where u is the displacement field, Ω is the deformation area, λ and μ are Lamé parameters, and E is the Green-Lagrange strain tensor; The strain energy function formula is: W=c1(I1-3) 3 +c2(I2-3) 2 +c3(I1-3)+c4(e Q -1) where c1, c2, c3, c4 are material coefficients, I1 and I2 are the first and second principal invariants, and Q is the Cauchy-Green tensor.

5. The AI-assisted valvular disease assessment system according to claim 1, characterized in that: The disease state automatic classification logic and output module adopts fuzzy membership function and fuzzy rules; the fuzzy membership function formula is: Among them, a, b and c are the parameters that define the fuzzy set of the linguistic variable A, and n is the shape parameter; The fuzzy rule formula is: R i :IF x1 is A i1 AND x2 is Ai2 THEN y is B i Among them, R i is the i-th fuzzy rule, x1 and x2 are input variables, A i1 and A i2 is the fuzzy set corresponding to x1 and x2, y is the output variable, B i is the fuzzy set corresponding to y.

6. The AI-assisted valvular disease assessment system according to claim 1, characterized in that: The special situation intelligent prompt and interaction module can be integrated with the hospital information system HIS and the medical image archiving and communication system PACS; the information retrieval formula is: I=HIS∪PACS, where I is the collection of complete medical records, previous examination reports, and image data of the patient retrieved from HIS and PACS; The special case identification formula is: S = IdentifySpecialCases (I) where S is the set of identified special cases, and IdentifySpecialCases is a function used to analyze the information set I and identify special cases; The formula for generating smart tips is: T = GenerateTips(S) where T is the set of smart tips generated for the operator, and GenerateTips is a function that generates corresponding tips based on the identified special case S; The formula for multi-language support is: L=UserLanguage, where L is the user's language preference, which is used to ensure that the interactive interface supports multi-language display.

Images