Medical image auxiliary diagnosis system based on artificial intelligence

Abstract

The invention discloses a medical image auxiliary diagnosis system based on artificial intelligence, which is suitable for diagnosis assistance of fetal ultrasonic images and comprises an image preprocessing module, a fetal recognition module, an imaging optimization module, a micro structure enhancement module, a characteristic parameter extraction module, a diagnosis assistance module and an image quality control module. The image preprocessing module is used for performing primary processing on the original ultrasonic image; the fetus recognition module is used for recognizing the position and the state of a fetus in the maternal body according to the processed ultrasonic image; the imaging optimization module is used for optimizing imaging conditions for micro-structure imaging by intelligently adjusting the position and angle of a probe and combining three-dimensional reconstruction and image correction technologies according to a fetus recognition result; the micro structure enhancement module is used for identifying and enhancing a micro anatomical structure in the image; the feature parameter extraction module uses a deep learning model to automatically identify and extract key feature parameters in the ultrasonic image. The imaging quality is improved, and the reliability of auxiliary diagnosis is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image assisted diagnosis, and particularly to a medical image assisted diagnosis system based on artificial intelligence. Background Art

[0002] Medical image assisted diagnosis technology refers to a technology that uses different medical imaging devices and methods to obtain relevant information of patients to assist doctors in accurate diagnosis. This technology plays an important role in the medical field, and with the continuous progress and application of the technology, its application scope and auxiliary effect are also constantly expanding and improving, and it is widely used in tumor diagnosis, cardiovascular disease diagnosis, trauma and fracture diagnosis.

[0003] In the existing diagnosis scenarios of obstetrics and pediatrics, traditional medical image assisted diagnosis technology can play the role of monitoring the development of the fetus and the position of the placenta, providing important information for obstetricians. However, compared with tumor diagnosis, cardiovascular disease diagnosis, trauma and fracture diagnosis, the effect is not good. One reason is that the fetal position and amniotic fluid volume affect the imaging effect, and it depends on the doctor's experience and manual operation to adjust the probe position and angle to obtain the best imaging effect. Therefore, it is limited by the doctor's personal skills and experience, and the process is time-consuming. The second is that due to the changes in the fetal position and amniotic fluid volume, the imaging quality is often unstable, which may lead to a decrease in diagnostic accuracy. The third is that small or complex anatomical structures may still be difficult to present clearly, especially when the fetal position is not good or the amniotic fluid volume is insufficient. Therefore, there is a lack of a special enhancement scheme for small structures, resulting in these structures may not be clear enough in the image.

[0004] In summary, it is necessary to propose a medical image assisted diagnosis system based on artificial intelligence to solve the above problems. Summary of the Invention

[0005] An object of the present invention is to solve at least one of the technical problems in the related art to some extent. In view of this, on the one hand, the present invention provides a medical image assisted diagnosis system based on artificial intelligence, which is applicable to the diagnostic assistance of fetal ultrasonic images, and includes an image preprocessing module, a fetal recognition module, an imaging optimization module, a small structure enhancement module, a feature parameter extraction module, a diagnostic assistance module and an image quality control module;

[0006] The image preprocessing module is used to perform preliminary processing on the original ultrasonic image; the fetal recognition module is used to identify the position and state of the fetus in the mother according to the processed ultrasonic image; the imaging optimization module optimizes the imaging conditions for imaging of minute structures by intelligently adjusting the probe position and angle, combining three-dimensional reconstruction and image correction techniques based on the fetal recognition result; the minute structure enhancement module is used to identify and enhance minute anatomical structures in the image; the feature parameter extraction module automatically identifies and extracts key feature parameters in the ultrasonic image using a deep learning model; the diagnostic assistance module gives preliminary diagnostic suggestions based on the extracted feature parameters and a personalized analysis model, and provides uncertainty analysis for the diagnostic suggestions through a Bayesian network; the image quality control module quantitatively scores the imaging effect through a deep learning-based image quality assessment algorithm, monitors the change of image quality in real time, and automatically adjusts the imaging parameters or suggests rescan according to the evaluation result.

[0007] According to an example of the present invention, the image preprocessing module includes an image denoising unit and an image enhancement unit;

[0008] The image denoising unit performs image denoising through an adaptive median filtering algorithm to remove speckle noise and random noise in the ultrasonic image and improve the signal-to-noise ratio of the image;

[0009] The image enhancement unit applies a contrast stretching algorithm to enhance the image contrast.

[0010] According to an example of the present invention, the fetal recognition module includes a body position recognition unit and an amniotic fluid volume assessment unit;

[0011] The body position recognition unit classifies and recognizes the fetal body position through a convolutional neural network, and the recognition information includes at least the fetal facing direction and the positions of the fetal limbs;

[0012] The amniotic fluid volume assessment unit uses image segmentation technology to separate the amniotic fluid area and calculate the amniotic fluid index, and combines a machine learning regression model to predict the influence degree of the amniotic fluid volume on the imaging quality.

[0013] According to an example of the present invention, the imaging optimization module includes a dynamic imaging adjustment unit and an image reconstruction and correction unit;

[0014] The dynamic imaging adjustment unit optimizes the imaging conditions by adjusting the probe position and angle according to the body position recognition result;

[0015] The image reconstruction and correction unit applies three-dimensional reconstruction technology to construct a three-dimensional model of the fetus.

[0016] According to an example of the present invention, the minute structure enhancement module includes a micro-structure recognition unit and a detail enhancement and rendering unit;

[0017] The micro-structure recognition unit uses a fine segmentation network to recognize and extract tiny anatomical structures, and the tiny anatomical structures at least include heart valve structures and facial detail structures;

[0018] The detail enhancement rendering unit performs high-definition rendering on the recognized tiny anatomical structures through a conditional generative adversarial network.

[0019] According to an example of the present invention, the feature parameter extraction module includes a parameter extraction unit and a personalized analysis model;

[0020] The parameter extraction unit automatically recognizes and extracts key feature parameters in the ultrasonic image by using a deep learning model, and the key feature parameters at least include heart size parameters and blood vessel diameter parameters, which are used to achieve multi-scale feature fusion;

[0021] The personalized analysis model is used to provide a personalized analysis model based on multiple factors such as race and nutritional level, and adopts an ensemble learning method to fuse various biometric data.

[0022] According to an example of the present invention, the diagnostic assistance module further includes a preliminary diagnosis suggestion unit and a diagnostic evaluation unit;

[0023] The preliminary diagnosis suggestion unit gives a preliminary diagnosis suggestion based on the extracted feature parameters and the personalized analysis model through a decision tree or random forest algorithm;

[0024] The diagnostic evaluation unit uses a Bayesian network to evaluate the confidence level of the diagnosis suggestion and provides uncertainty analysis.

[0025] According to an example of the present invention, an image quality control module is further included;

[0026] The image quality control module includes a quality evaluation unit and an image optimization feedback unit;

[0027] The quality evaluation unit uses an image quality evaluation algorithm based on deep learning to quantitatively score the imaging effect;

[0028] The image optimization feedback unit adjusts the imaging parameters according to the quality evaluation result or suggests rescan to ensure that the image quality meets the standard, and establishes an image quality database for continuously optimizing the image processing algorithm.

[0029] According to an example of the present invention, an image presentation module is further included;

[0030] The image presentation module constructs an experience knowledge base based on a knowledge graph for retrieving and matching similar cases, and uses deep learning style transfer to convert the image diagnosis cases in historical experience into consistent and standardized image presentations, and provides an interactive interface to allow fine-tuning of the image presentation.

[0031] On the other hand, the present invention also provides a workflow of a medical image assisted diagnosis system based on artificial intelligence, which is applicable to the diagnosis assistance of fetal ultrasound images. The workflow includes the following steps:

[0032] S1. Remove the speckle noise and random noise of the ultrasound image through the adaptive median filtering algorithm, and further refine the image using a deep learning denoising network; improve the signal-to-noise ratio by training and learning the detailed features in the image; apply the contrast stretching algorithm to enhance the image contrast and make the fetal structure clearer;

[0033] S2. Use a convolutional neural network to classify and identify the fetal position, including the fetal facing direction and the positions of the fetal limbs; accurately locate each part of the fetus using a pose estimation algorithm by training and learning the fetal features in the image to obtain the position recognition result; use image segmentation technology to separate the amniotic fluid area in the image, calculate the amniotic fluid index, and combine a machine learning regression model to predict the influence degree of the amniotic fluid volume on the imaging quality to obtain the amniotic fluid volume evaluation result;

[0034] S3. According to the position recognition result and the amniotic fluid volume evaluation result in step S2, optimize the imaging conditions by adjusting the probe position and angle, apply 3D reconstruction technology to construct a fetal 3D model, and use an image correction algorithm to eliminate the image distortion caused by the fetal position or amniotic fluid to obtain a corrected image;

[0035] S4. Identify and extract the tiny anatomical structures in the image through a fine segmentation network, achieve accurate identification by training and learning the tiny structure features in the image, and then perform high-definition rendering on the identified tiny anatomical structures through a conditional generative adversarial network, and enhance the clarity of the structure edges through edge enhancement to obtain the finally enhanced image;

[0036] S5. Automatically identify and extract the key feature parameters in the ultrasound image using a deep learning model, and the model output is a set of feature parameters; achieve accurate extraction by training and learning the feature parameters in the image; combine multi-factor integrated learning methods such as race and nutritional level to fuse various biometric data, construct a personalized analysis model, and apply transfer learning technology to continuously optimize the model according to new case data;

[0037] S6. Based on the feature parameters extracted in step S5 and the personalized analysis model, give preliminary diagnostic suggestions through a decision tree or random forest algorithm;

[0038] S7. Through an image quality assessment algorithm based on deep learning, quantitatively score the imaging effect, implement dynamic quality control, monitor the change of image quality in real time, adjust the imaging parameters according to the quality assessment results or recommend re-scanning, calculate the optimal imaging parameters through the image quality assessment algorithm, and then establish an image quality database for continuous optimization of the image processing algorithm;

[0039] S8. Construct an experience knowledge base based on a knowledge graph for retrieving and matching similar cases, and use deep learning style transfer technology to convert image diagnosis cases with high similarity in historical experience into uniformly standardized image presentations.

[0040] Compared with the prior art, the technical solution of the present invention has the following advantages:

[0041] (1) Optimize the imaging conditions according to the fetal position and amniotic fluid volume, improve the imaging quality and provide real-time feedback, facilitate quick positioning of the best imaging position, and reduce the operation time.

[0042] (2) Apply 3D reconstruction technology to construct a 3D fetal model, make up for the deficiencies of 2D imaging, and use an image correction algorithm to eliminate image distortion and improve the diagnostic accuracy.

[0043] (3) Focus on identifying and enhancing tiny anatomical structures in the image, such as heart valves, facial details, etc., improve the clarity and recognition accuracy of the structures through high-definition rendering and edge enhancement technologies, and use a deep learning model for fine segmentation and high-definition rendering to achieve accurate recognition and clear presentation of tiny structures.

[0044] (4) Combine a personalized analysis model, consider various biometric data, and improve the adaptability and accuracy of diagnosis.

[0045] (5) Apply transfer learning technology to continuously optimize the model according to new case data, ensure that the model can continuously adapt to new diagnostic requirements and data changes, achieve multi-modal information fusion, combine data such as blood tests to improve the diagnostic accuracy, and at the same time use a Bayesian network to provide uncertainty analysis to enhance the reliability of diagnosis.

[0046] The additional aspects and advantages of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is the topology diagram of the medical image assisted diagnosis system based on artificial intelligence in Embodiment 1 of the present invention.

[0049] Figure 2 It is the workflow diagram of the medical image assisted diagnosis system based on artificial intelligence in Embodiment 2 of the present invention. Detailed implementation manners

[0050] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention, and should not be construed as limiting the present invention.

[0051] Embodiment 1

[0052] Please refer to Figure 1 、 Figure 2 As shown, the present invention provides a medical image assisted diagnosis system based on artificial intelligence, which is applicable to the diagnosis assistance of fetal ultrasonic images, and includes an image preprocessing module, a fetal recognition module, an imaging optimization module, a micro-structure enhancement module, a feature parameter extraction module, a diagnosis assistance module, and an image quality control module;

[0053] Among them, the image preprocessing module is used for the preliminary processing of the original ultrasonic image; the fetal recognition module is used to recognize the position and state of the fetus in the mother's body according to the processed ultrasonic image; the imaging optimization module, according to the fetal recognition result, optimizes the imaging conditions for micro-structure imaging by intelligently adjusting the probe position and angle, combined with three-dimensional reconstruction and image correction techniques; the micro-structure enhancement module is used to identify and enhance the micro-anatomical structures in the image; the feature parameter extraction module automatically identifies and extracts the key feature parameters in the ultrasonic image by using a deep learning model; the diagnosis assistance module gives preliminary diagnosis suggestions based on the extracted feature parameters and a personalized analysis model, and provides uncertainty analysis for the diagnosis suggestions through a Bayesian network; the image quality control module quantitatively scores the imaging effect through a deep learning-based image quality assessment algorithm, monitors the change of the image quality in real time, and automatically adjusts the imaging parameters or suggests re-scanning according to the evaluation result.

[0054] According to an example of the present invention, the aforementioned image preprocessing module includes an image denoising unit and an image enhancement unit; the image denoising unit performs image denoising through an adaptive median filtering algorithm to remove speckle noise and random noise in the ultrasonic image, improve the signal-to-noise ratio of the image, and further refine the denoising effect using a deep learning denoising network while preserving image details; the image enhancement unit applies a contrast stretching algorithm to enhance the image contrast to make the fetal structure clearer, and uses a deep learning-based super-resolution reconstruction technology to improve the image resolution, especially the clarity of the tiny structure part.

[0055] According to an example of the present invention, the aforementioned fetal recognition module includes a posture recognition unit and an amniotic fluid volume assessment unit; the posture recognition unit classifies and recognizes the fetal posture through a convolutional neural network, and the recognition information includes at least the fetal facing direction and the positions of the fetal limbs, and uses a pose estimation algorithm to accurately locate each part of the fetus to provide a basis for subsequent imaging optimization; the amniotic fluid volume assessment unit uses image segmentation technology to separate the amniotic fluid area and calculate the amniotic fluid index, and combines a machine learning regression model to predict the impact of the amniotic fluid volume on the imaging quality.

[0056] According to an example of the present invention, the aforementioned imaging optimization module includes a dynamic imaging adjustment unit and an image reconstruction and correction unit; the dynamic imaging adjustment unit optimizes the imaging conditions by adjusting the probe position and angle according to the posture recognition result; the image reconstruction and correction unit applies three-dimensional reconstruction technology to construct a three-dimensional model of the fetus to make up for the deficiency of two-dimensional imaging, and uses an image correction algorithm to eliminate image distortion caused by the posture or amniotic fluid.

[0057] According to an example of the present invention, the aforementioned tiny structure enhancement module includes a micro-structure recognition unit and a detail enhancement and rendering unit; among them, the micro-structure recognition unit uses a fine segmentation network to identify and extract tiny anatomical structures, and the tiny anatomical structures include at least heart valve structures and facial detail structures, and uses an attention mechanism to enhance the attention to key structures and improve the recognition accuracy; the detail enhancement and rendering unit performs high-definition rendering on the identified tiny anatomical structures through a conditional generative adversarial network, and integrates edge enhancement technology to improve the clarity of the structure edges.

[0058] According to an example of the present invention, the aforementioned feature parameter extraction module includes a parameter extraction unit and a personalized analysis model; the parameter extraction unit automatically identifies and extracts key feature parameters in the ultrasonic image using a deep learning model, and the key feature parameters include at least heart size parameters and blood vessel diameter parameters, which are used to achieve multi-scale feature fusion and improve the comprehensiveness and accuracy of parameter extraction. The personalized analysis model is used to provide a personalized analysis model based on multiple factors such as race and nutritional level, uses an ensemble learning method to fuse various biometric data, applies transfer learning technology, and continuously optimizes the model according to new case data to improve the diagnostic adaptability.

[0059] According to an example of the present invention, the foregoing diagnostic assistance module further includes a preliminary diagnosis suggestion unit and a diagnosis evaluation unit; the preliminary diagnosis suggestion unit gives preliminary diagnosis suggestions based on the extracted feature parameters and the personalized analysis model, through decision tree or random forest algorithms, realizes multi-modal information fusion, and combines blood tests to improve the diagnostic accuracy. The diagnosis evaluation unit uses a Bayesian network to evaluate the confidence level of the diagnosis suggestions, provides uncertainty analysis, introduces a doctor feedback mechanism, and continuously optimizes the model according to the confirmed or corrected diagnosis results by the doctor.

[0060] According to an example of the present invention, the auxiliary diagnosis system further includes an image quality control module; the image quality control module includes a quality evaluation unit and an image optimization feedback unit; the quality evaluation unit uses a deep learning-based image quality evaluation algorithm to quantitatively score the imaging effect, implements dynamic quality control, and monitors the change of image quality in real time and reminds the doctor. The image optimization feedback unit adjusts the imaging parameters according to the quality evaluation results or suggests re-scanning to ensure that the image quality meets the standards, and establishes an image quality database for continuously optimizing the image processing algorithm.

[0061] According to an example of the present invention, the auxiliary diagnosis system further includes an image presentation module; the image presentation module is used to construct an experience knowledge base based on a knowledge graph for retrieving and matching similar cases; and has a consistent image presentation unit, which can use deep learning style transfer to convert the image diagnosis cases in historical experience into consistent standardized image presentations, provides an interactive interface to allow fine-tuning of the image presentation, and allows the doctor to fine-tune the image presentation according to personal preferences to ensure diagnostic consistency.

[0062] Embodiment 2

[0063] The present invention also provides a workflow of a medical image auxiliary diagnosis system based on artificial intelligence, which is applicable to the diagnostic assistance of fetal ultrasound images. The workflow includes the following steps:

[0064] S1. Image denoising, removing speckle noise and random noise of the ultrasound image through an adaptive median filtering algorithm, and further using a deep learning denoising network to refine the image; by training and learning the detailed features in the image, improving the signal-to-noise ratio; applying a contrast stretching algorithm to enhance the image contrast to make the fetal structure clearer.

[0065] Specifically, the size of the filtering window is dynamically adjusted according to the statistical characteristics of local noise, and an edge-preserving mechanism is introduced to ensure that the edge details in the image are not damaged while removing the noise. For each pixel, calculate the noise level within its neighborhood;

[0066] Let the original image be I(x,y), and the denoised image be I denoised(x, y), the adaptive median filtering algorithm adjusts the filter window size according to the local noise situation, effectively removing noise while preserving image details;

[0067] Furthermore, a denoising network based on a convolutional neural network (CNN) is trained with a large number of noisy and noise-free image pairs to learn the mapping from the noisy image to the noise-free image. The network structure contains multiple convolutional layers, and residual learning is used to accelerate convergence and improve denoising performance. The network input is I denoised (x, y), and the output is the denoised image I refined (x, y). By training, the detailed features in the image are learned to improve the signal-to-noise ratio;

[0068] The contrast stretching algorithm is applied to enhance the image contrast, making the fetal structure clearer, as shown in Equation (1):

[0069]

[0070] In Equation (1), I in (x, y) is the pixel value of the input image, and I out (x, y) is the pixel value of the output image. I min and I max are the minimum and maximum pixel values in the input image respectively, and L is the number of gray levels of the output image. Let the enhanced image be I enhanced (x, y). The pixel value distribution is adjusted through linear or non-linear transformation to enhance the image contrast; at the same time, the image resolution is improved by SRGAN, especially the clarity of the tiny structure part, to obtain the high-resolution image I high-res (x, y).

[0071] S2. Use a convolutional neural network to classify and identify the fetal position, including the fetal facing direction, the positions of the fetal limbs, the flexed or extended posture, and the head position in the supine or prone position; learn the fetal features in the image through training, and use the pose estimation algorithm to accurately locate each part of the fetus to obtain the position recognition result;

[0072] Let the input image be I high-res (x, y), and the network output is the position classification result C position ;

[0073] Accurate classification is achieved by learning the fetal features in the image through training;

[0074] Pose estimation. Use the OpenPose pose estimation algorithm to accurately locate each part of the fetus, providing a basis for subsequent imaging optimization;

[0075] Let the positions of each part of the fetus be P parts , and the accurate positions of the fetal head, limbs, and torso are calculated through the algorithm.

[0076] Amniotic fluid volume assessment: Using image segmentation technology to separate the amniotic fluid area in the image and calculate the amniotic fluid index;

[0077] Let the amniotic fluid area be R amniotic , obtain the area of the amniotic fluid area through the segmentation algorithm, and then calculate the amniotic fluid index AFI;

[0078] Combine the machine learning regression model to predict the influence degree of amniotic fluid volume on the imaging quality, and obtain the influence coefficient k AFI;

[0079] S3. According to the body position recognition result and amniotic fluid volume assessment result in step S2, optimize the imaging conditions by manually adjusting the probe position and angle, apply three-dimensional reconstruction technology to construct a three-dimensional model of the fetus, and use an image correction algorithm to eliminate the image distortion caused by the fetus position or amniotic fluid to obtain a corrected image;

[0080] Let the three-dimensional model be M 3D , obtain the three-dimensional model of the fetus through the three-dimensional reconstruction algorithm;

[0081] Use an image correction algorithm to eliminate the image distortion caused by the body position or amniotic fluid to obtain a corrected image I corrected (x, y);

[0082] S4. Identify and extract the minute anatomical structures in the image, such as heart valves, facial details, etc. through a fine segmentation network (U-Net, Mask R-CNN or DeepLab);

[0083] Let the input image be I corrected (x, y), and the network output be the minute structure area R micro ;

[0084] Learn the minute structure features in the image through training to achieve accurate recognition; then perform high-definition rendering on the identified minute anatomical structures through a conditional generative adversarial network to obtain a high-definition rendered image I HD (x, y); Enhance the clarity of the structure edges through edge enhancement to obtain the final enhanced image I enhanced-micro (x, y);

[0085] S5. Use a deep learning model to automatically identify and extract the key feature parameters in the ultrasonic image, such as heart size, blood vessel diameter, fetal weight, biparietal diameter of the fetal head, and femur length. Let the input image be I enhanced-micro (x, y), and the model output be the feature parameter set F parameters ;

[0086] Learn the feature parameters in the image through training to achieve accurate extraction;

[0087] Personalized analysis, integrating biometric data through an integrated learning method that combines multiple factors such as race, nutritional level, maternal age, gestational week, and medical history, to construct a personalized analysis model;

[0088] Let the model be M personalized , with the input being the set of feature parameters F parameters , and the output being the personalized analysis result A personalized ;

[0089] Apply transfer learning technology to continuously optimize the model based on new case data to improve diagnostic adaptability.

[0090] S6. Preliminary diagnosis suggestion: Based on the feature parameters extracted in step S5 and the personalized analysis model, give a preliminary diagnosis suggestion through a decision tree or random forest algorithm;

[0091] Let the algorithm be A diagnosis , with the input being the set of feature parameters F parameters and the personalized analysis result A personalized , and the output being the preliminary diagnosis suggestion D suggestion ;

[0092] Combine blood test, blood sugar level, blood pressure, and blood routine index data of the pregnant woman to improve diagnostic accuracy.

[0093] Diagnostic evaluation: Use a Bayesian network to evaluate the confidence level of the diagnostic suggestion and provide uncertainty analysis;

[0094] Let the Bayesian network be B network , with the input being the preliminary diagnosis suggestion D suggestion , and the output being the confidence level L confidence and the uncertainty analysis U analysis ;

[0095] Introduce a doctor feedback mechanism to continuously optimize the model according to the confirmed or corrected diagnostic results by doctors;

[0096] S7. Quality assessment: Through an image quality assessment algorithm based on deep learning, quantitatively score the imaging effect;

[0097] Let the assessment algorithm be A quality , with the input being the finally enhanced image I enhanced-micro (x, y), and the output being the quality score S quality ;

[0098] Implement dynamic quality control to monitor the change of image quality in real time and remind the doctor;

[0099] Image optimization feedback: According to the quality assessment result, automatically adjust the imaging parameters or suggest re-scanning;

[0100] Let the adjusted imaging parameters be P aramsadjusted , and calculate the optimal imaging parameters through the image quality evaluation algorithm;

[0101] Establish an image quality database for continuously optimizing the image processing algorithm.

[0102] S8. Empirical knowledge base, construct an empirical knowledge base based on a knowledge graph for retrieving and matching similar cases;

[0103] Let the knowledge base be K knowledge , and retrieve the set of cases C similar to the current case in the knowledge base through an algorithm similar ;

[0104] Consistent image presentation, utilize deep learning style transfer technology to transform image diagnosis cases with high similarity in historical experience into consistent and standardized image presentations;

[0105] Let the style transfer algorithm be A style-transfer , the input is the current case image I enhanced-micro (x, y) and the set of similar case images C similar , and the output is the consistent and standardized image I standardized (x, y);

[0106] Provide an interactive interface to allow doctors to fine-tune the image presentation according to their personal preferences to ensure diagnostic consistency.

[0107] Through the above steps, the artificial intelligence-based medical image assisted diagnosis system of the present invention effectively solves the problems of traditional solutions in aspects such as the imaging effect being affected by the fetal position and amniotic fluid volume, the difficulty in clearly presenting small or complex anatomical structures, and the lack of a perfect analysis method for characteristic parameters, etc., by means of intelligent adjustment of imaging conditions, enhancement of the clarity of small structures, provision of personalized analysis, and multi-modal information fusion, etc., improving the accuracy and efficiency of diagnosis.

[0108] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means more than two unless otherwise specifically defined.

[0109] In the present invention, unless otherwise clearly specified or limited, terms such as "install", "connect", "link", "fix", etc. shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0110] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0111] For those skilled in the art, various changes and corrections will undoubtedly be obvious after reading the above description. Therefore, the appended claims should be regarded as covering all changes and corrections that embrace the true intent and scope of the present invention. Any and all equivalent scopes and contents within the scope of the claims should be considered to still fall within the intent and scope of the present invention.

Claims

1. An artificial intelligence-based medical image assisted diagnosis system, applicable to the diagnosis assistance of fetal ultrasound images, characterized in that, It includes an image preprocessing module, a fetal recognition module, an imaging optimization module, a micro-structure enhancement module, a feature parameter extraction module, a diagnostic assistance module, and an image quality control module; The image preprocessing module is used to perform preliminary processing on the original ultrasonic image; The fetal recognition module is used to identify the position and state of the fetus in the mother's body based on the processed ultrasonic image; According to the fetal recognition result, the imaging optimization module optimizes the imaging conditions for micro-structure imaging by intelligently adjusting the probe position and angle, combined with three-dimensional reconstruction and image correction technologies; The micro-structure enhancement module is used to identify and enhance the micro-anatomical structures in the image; The feature parameter extraction module automatically identifies and extracts the key feature parameters in the ultrasonic image using a deep learning model; The diagnostic assistance module gives preliminary diagnostic suggestions based on the extracted feature parameters and a personalized analysis model, and provides uncertainty analysis for the diagnostic suggestions through a Bayesian network; The image quality control module quantitatively scores the imaging effect through a deep learning-based image quality assessment algorithm, monitors the change of image quality in real time, and automatically adjusts the imaging parameters or suggests rescan according to the evaluation result.

2. The artificial intelligence-based medical image assisted diagnosis system according to claim 1, wherein The image preprocessing module includes an image denoising unit and an image enhancement unit; The image denoising unit performs image denoising through an adaptive median filtering algorithm to remove speckle noise and random noise in the ultrasonic image and improve the signal-to-noise ratio of the image; The image enhancement unit applies a contrast stretching algorithm to enhance the image contrast.

3. The artificial intelligence-based medical image assisted diagnosis system according to claim 1 or 2, wherein The fetal recognition module includes a posture recognition unit and an amniotic fluid volume assessment unit; The posture recognition unit classifies and recognizes the fetal posture through a convolutional neural network, and the recognition information includes at least the facing direction of the fetus and the positions of the fetal limbs; The amniotic fluid volume assessment unit uses image segmentation technology to separate the amniotic fluid area to calculate the amniotic fluid index, and combines a machine learning regression model to predict the influence degree of the amniotic fluid volume on the imaging quality.

4. The artificial intelligence-based medical image assisted diagnosis system according to claim 3, wherein The imaging optimization module includes a dynamic imaging adjustment unit and an image reconstruction and correction unit; The dynamic imaging adjustment unit optimizes the imaging conditions by adjusting the probe position and angle according to the posture recognition result; The image reconstruction and correction unit applies three-dimensional reconstruction technology to construct a three-dimensional model of the fetus.

5. The artificial intelligence-based medical image assisted diagnosis system according to claim 4, wherein The micro-structure enhancement module includes a micro-structure recognition unit and a detail enhancement and rendering unit; The micro-structure recognition unit uses a fine segmentation network to identify and extract micro-anatomical structures, and the micro-anatomical structures include at least cardiac valve structures and facial detail structures; The detail enhancement and rendering unit performs high-definition rendering on the identified micro-anatomical structures through a conditional generative adversarial network.

6. The medical image assisted diagnosis system based on artificial intelligence according to claim 5, characterized in that the feature parameter extraction module includes a parameter extraction unit and a personalized analysis model; the parameter extraction unit automatically identifies and extracts key feature parameters in the ultrasonic image by using a deep learning model, and the key feature parameters at least include a heart size parameter and a blood vessel diameter parameter for realizing multi-scale feature fusion; the personalized analysis model is used to provide a personalized analysis model based on multiple factors such as race and nutritional level, and adopts an ensemble learning method to fuse various biometric data.

7. The medical image assisted diagnosis system based on artificial intelligence according to claim 6, characterized in that the diagnosis assistance module further includes a preliminary diagnosis suggestion unit and a diagnosis evaluation unit; the preliminary diagnosis suggestion unit gives a preliminary diagnosis suggestion based on the extracted feature parameters and the personalized analysis model through a decision tree or a random forest algorithm; the diagnosis evaluation unit uses a Bayesian network to evaluate the confidence level of the diagnosis suggestion and provides uncertainty analysis.

8. The medical image assisted diagnosis system based on artificial intelligence according to claim 7, characterized in that, It further includes an image quality control module; the image quality control module includes a quality evaluation unit and an image optimization feedback unit; the quality evaluation unit quantitatively scores the imaging effect by using an image quality evaluation algorithm based on deep learning; the image optimization feedback unit adjusts the imaging parameters according to the quality evaluation result or suggests re-scanning to ensure that the image quality meets the standard, and establishes an image quality database for continuously optimizing the image processing algorithm.

9. The medical image assisted diagnosis system based on artificial intelligence according to claim 8, wherein, It further includes an image presentation module; the image presentation module constructs an experience knowledge base based on a knowledge graph for retrieving and matching similar cases, and uses deep learning style transfer to convert the image diagnosis cases in historical experience into a consistent and standardized image presentation, and provides an interactive interface to allow fine-tuning of the image presentation.

10. The workflow of a medical image assisted diagnosis system based on artificial intelligence, applicable to the diagnostic assistance of fetal ultrasound images, is characterized in that, The working process includes the following steps: S1. Remove the speckle noise and random noise of the ultrasonic image through an adaptive median filtering algorithm, and further refine the image by using a deep learning denoising network; learn the detailed features in the image through training to improve the signal-to-noise ratio; apply a contrast stretching algorithm to enhance the image contrast to make the fetal structure clearer; S2. Use a convolutional neural network to classify and identify the fetal position, including the fetal facing direction and the positions of the fetal limbs; learn the fetal features in the image through training, and use a pose estimation algorithm to accurately locate each part of the fetus to obtain the position recognition result; use image segmentation technology to separate the amniotic fluid area in the image, calculate the amniotic fluid index, and combine a machine learning regression model to predict the influence degree of the amniotic fluid volume on the imaging quality to obtain the amniotic fluid volume evaluation result; S3. According to the position recognition result and the amniotic fluid volume evaluation result in step S2, optimize the imaging conditions by adjusting the probe position and angle, apply three-dimensional reconstruction technology to construct a fetal three-dimensional model, and use an image correction algorithm to eliminate the image distortion caused by the fetal position or amniotic fluid to obtain a corrected image; S4. Identify and extract the tiny anatomical structures in the image through a fine segmentation network. Learn the tiny structure features in the image through training to achieve accurate identification. Then, perform high-definition rendering on the identified tiny anatomical structures through a conditional generative adversarial network, and enhance the clarity of the structure edges through edge enhancement to obtain the final enhanced image. S5. Automatically identify and extract the key feature parameters in the ultrasonic image using a deep learning model, and the model output is a set of feature parameters. Achieve accurate extraction by learning the feature parameters in the image through training. Combine multi-factor integrated learning methods such as race and nutritional level to fuse various biometric data, construct a personalized analysis model, and apply transfer learning technology to continuously optimize the model according to new case data. S6. Based on the feature parameters and personalized analysis model extracted in step S5, give preliminary diagnostic suggestions through a decision tree or random forest algorithm. S7. Through an image quality assessment algorithm based on deep learning, quantitatively score the imaging effect, implement dynamic quality control, monitor the change of image quality in real time, adjust the imaging parameters according to the quality assessment results or suggest rescanning, calculate the optimal imaging parameters through the image quality assessment algorithm, and then establish an image quality database for continuously optimizing the image processing algorithm. S8. Construct an experience knowledge base based on a knowledge graph for retrieving and matching similar cases, and use deep learning style transfer technology to transform the image diagnosis cases with high similarity in historical experience into a consistent and standardized image presentation.

Images