Medical image analysis and diagnosis method based on generative artificial intelligence

Abstract

The invention discloses a medical image analysis and diagnosis method based on generative artificial intelligence, and particularly relates to the field of image analysis, comprising the following steps: receiving original image data from X-ray machine equipment, and ensuring sufficient preparation of a patient to reduce interference; then, carrying out scaling processing on the image, unifying the size and keeping the aspect ratio, and meanwhile, carrying out normalization; then, through de-atomization and enhancement processing, the image quality is improved, and halation artifacts are removed; the processed image is converted into a high-dimensional feature vector by using an image convolutional neural network, so that the processing efficiency and accuracy are improved; the feature vector is then sent to a deep learning classifier, and a lesion area is identified and marked through a trained model; finally, a medical diagnosis report is automatically generated according to the diagnosis information, the disease type, the disease severity and medical suggestions are explained in detail, and doctors and patients can understand the disease and follow-up treatment conveniently.

Description

Technical Field

[0001] The present invention relates to the field of image analysis technology, and more specifically, to a medical image analysis and diagnosis method based on generative artificial intelligence. Background Art

[0002] Medical image analysis refers to the process of analyzing and processing medical images (such as X-rays, CT scans, MRI, etc.) using computer science and digital information processing technology. These image data are important bases for doctors to diagnose diseases, plan treatments, monitor treatments, and manage cases.

[0003] Generative AI is a type of AI technology that can learn and generate new data, including generative adversarial networks (GANs), graph convolutional neural networks, etc.; these technologies have achieved remarkable results in image generation, speech synthesis, text creation and other fields; in the field of medical image analysis, generative AI has great application potential and can greatly improve the accuracy and efficiency of image analysis.

[0004] However, traditional medical image analysis still has some shortcomings in actual use. For example, traditional manual analysis methods are not only time-consuming and labor-intensive, but also easily affected by factors such as doctor experience and fatigue, leading to an increased risk of misdiagnosis or missed diagnosis. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a medical image analysis and diagnosis method based on generative artificial intelligence to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Step A1: receiving raw medical image data from a medical imaging device;

[0008] Step A2: scaling the medical image to make the size of the medical image uniform;

[0009] Step A3: performing image enhancement processing on the scaled medical image;

[0010] Step A4: performing a first matrix calculation on the processed medical image using a graph convolutional neural network model, and generating a high-dimensional feature vector corresponding to the medical image;

[0011] Step A5: input the feature vector into a classifier, analyze the image to identify and mark the lesion area through a deep learning algorithm, and generate diagnostic information;

[0012] Step A6: Generate a medical diagnosis report based on the diagnosis information, which includes the disease type, severity of the disease and related medical advice.

[0013] Preferably, in step A1, before undergoing the X-ray examination, the patient needs to prepare according to the doctor's requirements, such as removing metal ornaments, wearing appropriate clothing, etc., to reduce interference with imaging quality;

[0014] The specific method of collecting X-ray medical images by X-ray equipment is as follows:

[0015] Start-up and preheating: Ensure that the X-ray machine is in normal working condition and perform necessary preheating;

[0016] Setting parameters: Set the tube voltage, tube current, and exposure time parameters according to the examination site and patient condition;

[0017] Positioning and exposure: guide the patient to the correct position, ensure that the examination part is in the best imaging position, and then perform the exposure operation;

[0018] Detector reception: X-rays penetrate the human body and are received by the detector, which converts the received ray signals into electrical signals;

[0019] Signal conversion and storage: The electrical signal is further processed, converted into a digital signal, and stored in a computer; the processing includes amplification and filtering of the electrical signal;

[0020] Image acquisition: Extracting stored medical images from a computer.

[0021] Preferably, in step A2, in the process of processing X-ray images, due to differences in equipment and acquisition environment, the size of the image often varies; in the image equipment differences, different X-ray machines use different imaging technologies and sensor sizes, resulting in inconsistent output image sizes; in the acquisition environment, slight changes in shooting distance, angle, and patient position may affect the size of the final image; the acquired medical image is scaled; the medical image is defined as matrix A x , y ; The calculation method of the conversion formula is as follows:

[0022] B(A x，y )=xC x，y *yA x，y , where B(A x，y ) is represented as the transformed medical image matrix, x is represented as the vertical dimension of the matrix, y is represented as the horizontal dimension of the matrix, C x，y Represented as a matrix with x vertical dimension and y horizontal dimension;

[0023] After scaling, all images are adjusted to a uniform resolution, a standard width and height; during the resizing process, the aspect ratio of the original image must be kept unchanged to avoid deformation caused by stretching or compression; the image is normalized so that the pixel value falls within the range of 0 to 1.

[0024] Preferably, in step A3, the method for performing image enhancement processing on the medical image is to perform defogging processing on the image. In medical images, the "halo artifact" in the X-ray image is similar to the "fog" noise in the natural image. This can effectively remove the problem of "halo artifact" in the image. The defogging method is specifically as follows:

[0025] Among them, E represents the medical image after defogging, F represents the medical image without the past fogging treatment, G represents the transmittance, and H represents the global atmospheric illumination;

[0026] When taking medical images in low-light environments, the pixel values ​​in the image are affected and reduced due to insufficient lighting conditions; in this case, the light brightness is not enough to provide adequate illumination, resulting in reduced details and contrast in the image;

[0027] In addition, when light propagates through the atmosphere, scattering occurs; scattering refers to the interaction of light with gas molecules, particulate matter, etc. in the atmosphere during propagation, causing the direction of the light to change; this scattering phenomenon will cause the pixel value in the target area to decrease, making the objects in the image blurred or lose clarity.

[0028] Preferably, in step A4, the image representation is simplified by aggregating pixel blocks in the original image into a single pixel node; the aggregated pixel node set is represented as a, which is a compact representation of the processed original image; in the aggregation process, we first determine the pixel point in each pixel block that has the smallest difference with the aggregated pixel node value; this pixel point is considered to be the pixel that best represents the characteristics of the block, and is therefore selected as the representative pixel point of the block; this method ensures that even when the data scale is greatly reduced, the important visual information of the original image can be retained as much as possible;

[0029] The calculation method of the first matrix is:

[0030] Among them, K represents the first matrix, J represents the matrix of size N*N, M represents the degree matrix, and P represents the adjacency matrix of graph structure data;

[0031] In this way, the efficiency and accuracy of subsequent image processing tasks are improved; for example, in applications such as image recognition and image classification, this aggregated pixel node representation can be used as an input feature to help the model learn and recognize image content faster.

[0032] Preferably, in step A5, the feature vector and the corresponding lesion area label are input into the classifier, wherein the lesion area label input is annotated by a medical expert and the model is trained by a gradient descent optimization algorithm; during the training process, the model will learn how to predict the location and type of the lesion area based on the feature vector; the trained model is used to segment the new medical image to distinguish the lesion area from normal tissue in the image; this usually involves pixel-level classification, i.e. assigning each pixel to the most likely category, wherein the most likely category includes normal and lesion; different colors or contours are used to mark the lesion area on the segmented image so that the doctor or patient can intuitively see the location and range of the lesion.

[0033] Preferably, in step A6, the diagnostic information output from the classifier is verified, including key data such as the location, size, and type of the lesion area, to ensure that these data are consistent with the results of the image analysis and are accurate; the patient's personal information is checked, including name, gender, age, and medical record number, to ensure that it matches the image data to avoid information dislocation; based on the characteristics of the lesion area, the lesion area characteristics include shape, size, location, and texture; the specific disease type is determined in combination with the medical knowledge base and expert system; if the lesion characteristics are atypical or there are multiple possibilities, the report will indicate "suspected" or "pending further examination", and a list of possible diseases and their probabilities will be given; then the evaluation standard pathological grading is used, combined with the characteristics of the lesion area, the patient's medical history, physical signs and laboratory test results, to assess the severity of the disease; if necessary, the opinions of clinicians can be introduced for review to ensure the accuracy of the evaluation results.

[0034] Technical effects and advantages of the present invention:

[0035] The present invention receives raw image data from a medical imaging device and performs preprocessing such as scaling, noise removal, and image enhancement. Next, the processed image is subjected to feature extraction using a graph convolutional neural network model to generate a high-dimensional feature vector. These feature vectors are then input into a classifier, and the image is analyzed and identified using a deep learning algorithm to mark the lesion area and generate diagnostic information. Finally, a medical diagnosis report containing the disease type, severity of the disease, and related medical advice is generated based on the diagnostic information; this method can quickly identify the lesion area in the image, and greatly reduce the risk of misdiagnosis and missed diagnosis caused by fatigue of doctors due to long-term image analysis, thereby reducing the probability of misdiagnosis and missed diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0038] See also Figure 1 As shown, the present invention provides a medical image analysis and diagnosis method based on generative artificial intelligence, and the method is:

[0039] Step A1: receiving raw medical image data from a medical imaging device;

[0040] In step A1, before undergoing the X-ray examination, the patient needs to prepare according to the doctor's requirements, such as removing metal ornaments, wearing appropriate clothing, etc., to reduce interference with the imaging quality;

[0041] The specific method of collecting X-ray medical images by X-ray equipment is as follows:

[0042] Start-up and preheating: Ensure that the X-ray machine is in normal working condition and perform necessary preheating;

[0043] Setting parameters: Set the tube voltage, tube current, and exposure time parameters according to the examination site and patient condition;

[0044] Positioning and exposure: guide the patient to the correct position, ensure that the examination part is in the best imaging position, and then perform the exposure operation;

[0045] Detector reception: X-rays penetrate the human body and are received by the detector, which converts the received ray signals into electrical signals;

[0046] Signal conversion and storage: The electrical signal is further processed, converted into a digital signal, and stored in a computer; the processing includes amplification and filtering of the electrical signal;

[0047] Image acquisition: Extracting stored medical images from a computer.

[0048] Step A2: scaling the medical image to make the size of the medical image uniform;

[0049] In step A2, in the process of processing X-ray images, due to differences in equipment and acquisition environment, the size of the image often varies; in the image equipment differences, different X-ray machines use different imaging technologies and sensor sizes, resulting in inconsistent output image sizes; in the acquisition environment, slight changes in shooting distance, angle, and patient position may affect the size of the final image; the acquired medical image is scaled; the medical image is defined as matrix A x , y , the calculation method of the conversion formula is as follows:

[0050] B(A x，y )=xC x，y *yA x，y , where B(A x，y ) is represented as the transformed medical image matrix, x is represented as the vertical dimension of the matrix, y is represented as the horizontal dimension of the matrix, C x，y Represented as a matrix with x vertical dimension and y horizontal dimension;

[0051] The calculation method of the scaling function is as follows:

[0052] D=Q(B(A x，y )), where D represents the medical image matrix after the scaling function, Q(·) represents the scaling function, and B(A x，y ) is represented as the transformed medical image matrix;

[0053] After scaling, all images are adjusted to a uniform resolution, a standard width and height; during the resizing process, the aspect ratio of the original image must be kept unchanged to avoid deformation caused by stretching or compression; the image is normalized so that the pixel value falls within the range of 0 to 1.

[0054] Step A3: performing image enhancement processing on the scaled medical image;

[0055] In step A3, the method for performing image enhancement processing on the medical image is to perform defogging processing on the image. In medical images, the "halo artifact" in the X-ray image is similar to the "fog" noise in the natural image. This can effectively remove the "halo artifact" problem existing in the image. The defogging method is specifically as follows:

[0056] Among them, E represents the medical image after defogging, F represents the medical image without the past fogging treatment, G represents the transmittance, and H represents the global atmospheric illumination;

[0057] The transmittance is calculated as follows:

[0058] G=e α, where G represents the transmittance, e represents the natural constant, and α represents the scattering coefficient of the atmosphere;

[0059] When taking medical images in low-light environments, the pixel values ​​in the image are affected and reduced due to insufficient lighting conditions; in this case, the light brightness is not enough to provide adequate illumination, resulting in reduced details and contrast in the image;

[0060] In addition, when light propagates through the atmosphere, scattering occurs. Scattering refers to the interaction of light with gas molecules and particles in the atmosphere during propagation, causing the direction of the light to change. This scattering phenomenon will cause the pixel value in the target area to decrease, making the objects in the image blurry or lose clarity.

[0061] Insufficient lighting and light scattering are two important factors that affect image quality. When medical images are taken in an environment with insufficient light, the pixel value will be reduced, and the details and contrast will be weakened. When light is scattered when it propagates through the atmosphere, it will also cause the pixel value in the target area to decrease, making the image blurred or losing clarity. Therefore, when taking medical images, choose an environment with sufficient light and pay attention to avoid the influence of light scattering to obtain better image quality.

[0062] Step A4: performing a first matrix calculation on the processed medical image using a graph convolutional neural network model, and generating a high-dimensional feature vector corresponding to the medical image;

[0063] In step A4, the image representation is simplified by aggregating pixel blocks in the original image into a single pixel node; the aggregated pixel node set is represented as a, which is a compact representation of the processed original image; in the aggregation process, we first determine the pixel point in each pixel block that has the smallest difference with the aggregated pixel node value; this pixel point is considered to be the pixel that best represents the characteristics of the block, and is therefore selected as the representative pixel point of the block; this method ensures that even when the data scale is greatly reduced, the important visual information of the original image can be retained as much as possible;

[0064] The calculation method of the first matrix is:

[0065] Among them, K represents the first matrix, J represents the matrix of size N*N, M represents the degree matrix, and P represents the adjacency matrix of graph structure data;

[0066] The specific calculation method of graph structure data is as follows:

[0067] W = (a, b), where W represents graph structure data, a represents the set of aggregated pixel nodes, and b represents the set of edges;

[0068] In this way, the efficiency and accuracy of subsequent image processing tasks are improved; for example, in applications such as image recognition and image classification, this aggregated pixel node representation can be used as an input feature to help the model learn and recognize image content faster;

[0069] The first matrix is ​​transformed into a 1×1×512 feature vector through convolution operations and flattening operations, and the feature vector is input into the classifier.

[0070] Step A5: input the feature vector into a classifier, analyze the image to identify and mark the lesion area through a deep learning algorithm, and generate diagnostic information;

[0071] In step A5, the feature vector and the corresponding lesion area label are input into the classifier, wherein the lesion area label input is annotated by a medical expert and the model is trained by a gradient descent optimization algorithm; during the training process, the model will learn how to predict the location and type of the lesion area based on the feature vector; the trained model is used to segment the new medical image to distinguish the lesion area from the normal tissue in the image; this usually involves pixel-level classification, that is, assigning each pixel to the most likely category, wherein the most likely category includes normal and lesion; different colors or contours are used to mark the lesion area on the segmented image so that the doctor or patient can intuitively see the location and range of the lesion.

[0072] The calculation method of the gradient descent update rule is as follows:

[0073] β 0 =β-γVJ(β), where β 0 is represented as the updated parameter vector, β is represented as the parameter vector before the update, γ is represented as the learning rate, J(β) is represented as the loss function, and VJ(β) is represented as the loss function for the parameter β 0 The gradient of

[0074] Different colors or contour lines are used to mark the lesion area on the segmented image so that doctors or patients can intuitively see the location and range of the lesion.

[0075] Step A6: Generate a medical diagnosis report based on the diagnosis information, which includes the disease type, severity of the disease and related medical advice;

[0076] In step A6, the diagnostic information output from the classifier is verified, including key data such as the location, size, and type of the lesion area, to ensure that these data are consistent with the results of the image analysis and are accurate; the patient's personal information, including name, gender, age, and medical record number, is checked to ensure that it matches the image data to avoid information dislocation; based on the characteristics of the lesion area, the lesion area characteristics include shape, size, location, and texture; the specific disease type is determined in combination with the medical knowledge base and expert system; if the lesion characteristics are atypical or there are multiple possibilities, the report will indicate "suspected" or "pending further examination" and provide a list of possible diseases and their probabilities; then the evaluation standard pathological grading is used to evaluate the severity of the disease in combination with the characteristics of the lesion area, the patient's medical history, physical signs, and laboratory test results; if necessary, the opinions of clinicians can be introduced for review to ensure the accuracy of the evaluation results:

[0077] Develop personalized medical advice based on the type and severity of the disease, combined with current medical standards and best practice guidelines; including treatment plans, which include surgery, radiotherapy, chemotherapy, targeted therapy, follow-up plans, lifestyle adjustments, etc.; emphasize the importance of following doctor's orders, and remind patients and their families of possible complications and coping measures;

[0078] Integrate the above information into a medical diagnosis report with clear structure and accurate language; the report should include basic information of the patient, examination methods, diagnosis information (disease type, severity of the disease), medical advice and precautions, etc.

[0079] After the report is completed, it must be reviewed by a medical expert with relevant qualifications to ensure the accuracy and professionalism of the report;

[0080] Release the approved diagnostic reports to patients and their families in a timely manner, and provide necessary explanations and consulting services;

[0081] Archive the diagnostic report and related imaging data for future reference and follow-up;

[0082] According to the patient's specific situation and medical advice, a follow-up plan is developed and follow-up is conducted on time to monitor changes in the condition and adjust the treatment plan.

[0083] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A medical image analysis and diagnosis method based on generative artificial intelligence, characterized in that: include: Step A1: receiving raw medical image data from a medical imaging device; Step A2: scaling the medical image to make the size of the medical image uniform; Step A3: performing image enhancement processing on the scaled medical image; Step A4: performing a first matrix calculation on the processed medical image using a graph convolutional neural network model, and generating a high-dimensional feature vector corresponding to the medical image; Step A5: input the feature vector into a classifier, analyze the image to identify and mark the lesion area through a deep learning algorithm, and generate diagnostic information; Step A6: Generate a medical diagnosis report based on the diagnosis information, which includes the disease type, severity of the disease and related medical advice.

2. The medical image analysis and diagnosis method based on generative artificial intelligence according to claim 1, characterized in that: In step A1, before undergoing the X-ray examination, the patient needs to prepare according to the doctor's requirements, such as removing metal ornaments, wearing appropriate clothing, etc., to reduce interference with the imaging quality; The specific method of collecting X-ray medical images by X-ray equipment is as follows: Start-up and preheating: Ensure that the X-ray machine is in normal working condition and perform necessary preheating; Setting parameters: Set the tube voltage, tube current, and exposure time parameters according to the examination site and patient condition; Positioning and exposure: guide the patient to the correct position, ensure that the examination part is in the best imaging position, and then perform the exposure operation; Detector reception: X-rays penetrate the human body and are received by the detector, which converts the received ray signals into electrical signals; Signal conversion and storage: The electrical signal is further processed, converted into a digital signal, and stored in a computer; the processing includes amplification and filtering of the electrical signal; Image acquisition: Extracting stored medical images from a computer.

3. The medical image analysis and diagnosis method based on generative artificial intelligence according to claim 1, characterized in that: In step A2, during the process of processing X-ray images, the image sizes often differ due to differences in equipment and acquisition environments. Among the differences in image equipment, different X-ray machines use different imaging technologies and sensor sizes, resulting in inconsistent output image sizes. In the acquisition environment, slight changes in shooting distance, angle, and patient position may affect the size of the final image. The acquired medical image is scaled.

4. The method for medical image analysis and diagnosis based on generative artificial intelligence according to claim 3, characterized in that: A medical image is defined as a matrix A x，y , the calculation method of the conversion formula is as follows: B(A x，y )=xC x，y *yA x，y , where B(A x，y ) is represented as the transformed medical image matrix, x is represented as the vertical dimension of the matrix, y is represented as the horizontal dimension of the matrix, C x，y Represented as a matrix with x vertical dimension and y horizontal dimension; The calculation method of the scaling function is as follows: D=Q(B(A x，y )), where D represents the medical image matrix after the scaling function, Q(·) represents the scaling function, and B(A x，y ) is represented as the transformed medical image matrix.

5. The medical image analysis and diagnosis method based on generative artificial intelligence according to claim 1, characterized in that: In step A3, the method for performing image enhancement processing on the medical image is to perform defogging processing on the image; the defogging processing method is specifically: Among them, E represents the medical image after defogging, F represents the medical image without the past fogging treatment, G represents the transmittance, and H represents the global atmospheric illumination; When medical images are taken in low-light environments, the pixel values ​​in the image are affected and reduced due to insufficient lighting conditions; in this case, the light brightness is not sufficient to provide adequate illumination, resulting in reduced details and contrast in the image.

6. The method for medical image analysis and diagnosis based on generative artificial intelligence according to claim 5, characterized in that: The transmittance is calculated as follows: G=e α , where G represents the transmittance, e represents the natural constant, and α represents the scattering coefficient of the atmosphere.

7. The method for medical image analysis and diagnosis based on generative artificial intelligence according to claim 1, characterized in that: In step A4, the image representation is simplified by aggregating pixel blocks in the original image into a single pixel node; the aggregated pixel node set is represented as a; and the calculation method of the first matrix is: The calculation method of the first matrix is: Among them, K represents the first matrix, J represents the matrix of size N*N, M represents the degree matrix, and P represents the adjacency matrix of graph structure data.

8. The method for medical image analysis and diagnosis based on generative artificial intelligence according to claim 7, characterized in that: The specific calculation method of graph structure data is as follows: W = (a, b), where W represents graph structure data, a represents the set of aggregated pixel nodes, and b represents the set of edges; The first matrix is ​​transformed into a 1×1×512 feature vector through convolution operations and flattening operations, and the feature vector is input into the classifier.

9. The method for medical image analysis and diagnosis based on generative artificial intelligence according to claim 1, characterized in that: In step A5, the feature vector and the corresponding lesion region label are input into the classifier, wherein the lesion region label input is annotated by a medical expert, and the model is trained by a gradient descent optimization algorithm; the calculation method of the gradient descent update rule is specifically as follows: β0=β-γVJ(β), where β0 represents the updated parameter vector, β represents the parameter vector before updating, γ represents the learning rate, J(β) represents the loss function, and VJ(β) represents the gradient of the loss function with respect to the parameter β0.

10. The medical image analysis and diagnosis method based on generative artificial intelligence according to claim 1, characterized in that: In step A6, the diagnostic information output from the classifier is verified, including key data such as the location, size, and type of the lesion area, to ensure that these data are consistent with the results of the image analysis and are accurate; the patient's personal information, including name, gender, age, and medical record number, is checked to ensure that it matches the image data to avoid information misalignment; based on the characteristics of the lesion area, the lesion area characteristics include shape, size, location, and texture; the specific disease type is determined in combination with the medical knowledge base and expert system; if the lesion characteristics are atypical or there are multiple possibilities, the report will indicate "suspected" or "pending further examination", and a list of possible diseases and their probabilities will be given; Then, the severity of the disease is assessed using the pathological grading of the evaluation criteria, combined with the characteristics of the lesion area, the patient's medical history, physical signs, and laboratory test results; If necessary, clinician's opinions can be introduced for review to ensure the accuracy of the assessment results.

Images