Medical image ai processing method, workstation, medium, and apparatus

Abstract

Embodiments of the present application disclose a medical image AI processing method, a workstation, a medium and equipment. First, the workstation acquires MRI images to be processed to form a first image set. Then, the workstation performs similarity detection on the MRI images in the first image set and divides them into multiple image groups based on the structural or feature similarity between the images. This process can effectively identify images with repeated content or high similarity and construct a structured image distribution. Next, the workstation further screens images with significant differences between adjacent two groups, removes redundant information, and only retains images with strong representation and rich diagnostic information to form a processed image set. Finally, the image analysis process is performed on the processed image set to obtain a structured MRI image analysis result. Through the similarity grouping and screening method, the number of image processing is effectively reduced, the computational burden is reduced, and the image analysis efficiency and the response ability of the workstation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a medical image AI processing method, a workstation, a medium and equipment. BACKGROUND

[0002] Magnetic resonance imaging (MRI) is a non-invasive, non-radiation medical imaging technology widely used in imaging and diagnosis of internal structures of the human body. Due to its excellent imaging capability for soft tissues, MRI has important clinical value in brain, spinal cord, muscle, joint and internal organs.

[0003] In practical applications, the MRI image acquisition process usually involves multiple scanning sequences and section directions, and each examination can generate hundreds to thousands of images. Although this large-scale high-resolution data improves the imaging quality, it significantly increases the computational burden of the image processing process, resulting in a long overall processing period and low efficiency.

[0004] To alleviate the computational pressure, existing research and applications generally use image normalization, registration, denoising, region of interest (ROI) extraction and other preprocessing strategies to reduce the amount of redundant data and retain only the key regions for subsequent analysis. However, due to the large number of original images, even if only part of the region is processed, feature extraction and data conversion still need to be completed for each image, resulting in a time-consuming and inefficient processing process that cannot meet the actual needs.

[0005] In clinical applications, doctors often want to quickly load and view MRI images and adjust segmentation results and analysis parameters in real time. However, due to the time-consuming image preprocessing and feature extraction steps, the system cannot achieve efficient interactive operation, which seriously restricts the real-time and practicality of MRI images in clinical decision-making assistance.

[0006] In addition, in medical image artificial intelligence research, deep learning model training relies on a large amount of MRI image data. Before training, these data must complete complex data preprocessing procedures including format conversion, normalization, slicing, label alignment, etc. Limited by the existing data processing efficiency, the entire training period is significantly lengthened, and even under insufficient memory or resource bottlenecks, system interruptions, training failures and other problems occur, greatly affecting the iteration efficiency of the experiment and the usability of the model.

[0007] In summary, MRI image data processing in both clinical and scientific research scenarios faces the problems of complex processing procedures, long processing time, slow response and error-prone. SUMMARY

[0008] Therefore, it is necessary to propose a medical image AI processing method, a workstation, a computer device and a storage medium to solve the above problems.

[0009] The present application provides a medical image AI processing method, which comprises:

[0010] Obtaining an MRI image to be processed, and generating a first image set;

[0011] Detecting the similarity of the images in the first image set, grouping the MRI images in the first image set according to the similarity, and obtaining grouped images;

[0012] Selecting processing images from adjacent two grouped images according to a similarity threshold, and obtaining a processing image set;

[0013] Performing image analysis on the processing image set, and obtaining an MRI image analysis result.

[0014] In at least one embodiment of the present application, the specific steps of detecting the similarity of the images in the first image set, grouping the MRI images in the first image set according to the similarity, and obtaining grouped images comprise:

[0015] Calculating the similarity of all MRI images in the first image set to obtain first similarity data;

[0016] Obtaining a first similarity threshold, and grouping each MRI image in the first image set according to the first similarity threshold and the first similarity data to obtain grouped images.

[0017] In at least one embodiment of the present application, the specific steps of selecting processing images from adjacent two grouped images according to a similarity threshold, and obtaining a processing image set comprise:

[0018] Calculating the similarity of each MRI image in adjacent two grouped images to other MRI images to obtain second similarity data;

[0019] Selecting MRI images exceeding a second similarity threshold from the second similarity data to obtain a processing image set.

[0020] In at least one embodiment of the present application, the specific steps of performing image analysis on the processing image set, and obtaining an MRI image analysis result comprise:

[0021] Performing normalization processing on the processing image set to obtain a normalized image set;

[0022] Performing noise removal processing on the normalized image set to obtain an image set after noise removal processing;

[0023] Performing ROI region extraction on the image set after noise removal processing to obtain an ROI image set;

[0024] segmenting the ROI image set to obtain a segmented image set;

[0025] performing image analysis on the segmented image set to obtain an MRI image analysis result.

[0026] In at least one embodiment of the present application, the specific step of performing noise removal processing on the normalized image set to obtain a noise-removed image set comprises:

[0027] performing noise removal on the normalized image set to obtain a noise-removed image set;

[0028] performing artifact correction on the noise-removed image set to obtain a noise-removed image set.

[0029] In at least one embodiment of the present application, the medical image AI processing method further comprises:

[0030] establishing a standard image set;

[0031] generating a calibration image from the processed image set, the grouped image and the standard image set;

[0032] performing image analysis on the calibration image to obtain an MRI image calibration result.

[0033] In at least one embodiment of the present application, the specific step of generating a calibration image from the processed image set, the grouped image and the standard image set comprises:

[0034] selecting an MRI image with the largest difference from the standard image set from the processed image set to obtain a difference image;

[0035] selecting two adjacent frame images of the difference image from the grouped image;

[0036] generating a calibration image according to the two adjacent frame images and the difference image.

[0037] A workstation applied to the medical image AI processing method described in any one of the above embodiments, the workstation comprising:

[0038] an image acquisition module for acquiring an MRI image;

[0039] a similarity detection module for performing similarity detection on a first image set to generate a grouped image;

[0040] a selection module for selecting a processed image from two adjacent grouped images to generate a processed image set;

[0041] An image analysis module performs image analysis on the processing image set to generate an MRI image analysis result.

[0042] A computer device includes a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the following steps:

[0043] Obtain MRI images to be processed to generate a first image set;

[0044] Perform similarity detection on the images of the first image set, group multiple MRI images in the first image set according to similarity to obtain grouped images;

[0045] From the adjacent two grouped images, filter out processing images according to a similarity threshold to obtain a processing image set;

[0046] Perform image analysis on the processing image set to obtain an MRI image analysis result.

[0047] A computer-readable storage medium stores a computer program, the computer program being executed by a processor to cause the processor to perform the following steps:

[0048] Obtain MRI images to be processed to generate a first image set;

[0049] Perform similarity detection on the images of the first image set, group multiple MRI images in the first image set according to similarity to obtain grouped images;

[0050] From the adjacent two grouped images, filter out processing images according to a similarity threshold to obtain a processing image set;

[0051] Perform image analysis on the processing image set to obtain an MRI image analysis result.

[0052] The medical image AI processing method, workstation, medium and device of the embodiment have at least the following beneficial effects:

[0053] The medical image AI processing method, workstation, medium and device provided above first, the workstation obtains MRI images to be processed to form a first image set. Then, the workstation performs similarity detection on the MRI images in the first image set, and divides them into multiple image groups based on the structural or feature similarity between images. This process can effectively identify images with repeated or highly similar content and construct a structured image distribution.

[0054] Next, the workstation further screens images with significant differences between adjacent groups, eliminating redundant information and retaining only images with strong representativeness and rich diagnostic information to form a processed image set. Finally, the image analysis process is performed on the processed image set to obtain structured MRI image analysis results.

[0055] By using similarity grouping and screening, key images with diagnostic value are extracted from the original MRI image set, and a processed image set is constructed for analysis. This effectively reduces the number of images to be processed, lowers the computational burden, and improves image analysis efficiency and workstation responsiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] in:

[0058] Figure 1 Flowchart of a medical image AI processing method in one embodiment;

[0059] Figure 2 A flowchart of a medical image AI processing method in another embodiment;

[0060] Figure 3 This is a flowchart of a medical image AI processing method in yet another embodiment;

[0061] Figure 4 This is the application environment diagram of the workstation;

[0062] Figure 5 for Figure 4 The structural diagram of the workstation;

[0063] Figure 6 A block diagram of the structure of a computer device in one embodiment.

[0064] Wherein: 100, workstation; 110, image acquisition module; 120, similarity detection module; 130, screening module; 140, image analysis module. DETAILED DESCRIPTION

[0065] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0066] The present application provides a medical image AI processing method, comprising:

[0067] S101, acquiring an MRI image to be processed to generate a first image set;

[0068] S102, performing similarity detection on the images in the first image set, grouping the multiple MRI images in the first image set according to similarity, and obtaining grouped images;

[0069] S103, filtering out processing images from the adjacent two grouped images according to a similarity threshold to obtain a processing image set;

[0070] S104, performing image analysis on the processing image set to obtain an MRI image analysis result.

[0071] Please refer to Figures 1-4 In this embodiment, first, the workstation acquires the MRI image to be processed to form a first image set. Then, the workstation performs similarity detection on the MRI images in the first image set and divides them into multiple image groups based on the structural or feature similarity between images. This process can effectively identify images with repeated content or high similarity and construct a structured image distribution.

[0072] Next, the workstation further filters out images with significant differences between the adjacent two groups, removes redundant information, and only retains representative images with rich diagnostic information to form a processing image set. Finally, the image analysis process is performed on the processing image set to obtain a structured MRI image analysis result.

[0073] By similarity grouping and filtering, key images with diagnostic value are extracted from the original MRI image set to construct a processing image set for further analysis. This effectively reduces the number of image processing, reduces the computational burden, and improves the image analysis efficiency and workstation response capability.

[0074] It should be noted that the workstation is one of a host, a server, and a mobile terminal, and in this embodiment, it is a host.

[0075] In at least one embodiment of the present application, the specific steps of performing similarity detection on the images in the first image set, grouping the plurality of MRI images in the first image set by similarity, and obtaining grouped images include:

[0076] S201, performing similarity calculation on all MRI images in the first image set to obtain first similarity data;

[0077] S202, obtaining a first similarity threshold, and grouping each MRI image in the first image set by similarity according to the similarity threshold and the first similarity data to obtain grouped images.

[0078] Please refer to Figures 1-4 In this embodiment, similarity calculation is performed on all MRI images in the first image set to generate first similarity data for measuring the structural similarity between images; then, according to a set similarity threshold, images with similarity higher than the similarity threshold are classified into the same group, thereby obtaining a plurality of image groups. Each image group contains images with highly similar content, and the group can be used for subsequent image screening and analysis, improving the overall processing efficiency.

[0079] By performing similarity calculation on the MRI image set and combining the set similarity threshold, grouping of high-similarity images is realized, effectively reducing image redundancy and improving image processing efficiency and workstation response capability.

[0080] The similarity calculation can be based on traditional image similarity indicators, feature space comparison, deep learning encoding features, etc. The result of the similarity calculation forms a two-dimensional matrix or graph structure, which is the first similarity data. The first similarity data reflects the similarity between images in the image set.

[0081] The first similarity threshold is a manually set value. According to the first similarity threshold and the first similarity data, the images in the image set are grouped to ensure that the similarity between images in each group is higher than the similarity threshold.

[0082] For example, image clustering based on graph clustering algorithm, similarity graph modeling + connected component recognition, neighborhood optimization strategy based on time series or slice order, etc. Ultimately, a number of image groups are obtained, each containing MRI images with highly similar structure or content, which constitute the basic unit for subsequent image screening and processing. In this embodiment, image clustering based on graph clustering algorithm is used for processing.

[0083] By performing similarity detection and grouping on the first image set, images with highly consistent content are grouped together, which can avoid repeated calculation and analysis in subsequent processing, effectively reducing the amount of data to be processed.

[0084] The AI model is sensitive to highly repetitive images during training or inference, which can cause training bias or memory waste. By pre-screening by grouping to remove redundant images, the diversity and representativeness of the training data can be improved, while reducing the memory occupation and improving the running stability.

[0085] In at least one embodiment of the present application, the specific steps of selecting the processing images from the two adjacent grouping images according to the similarity threshold include:

[0086] S203, calculate the similarity of each MRI image in the two adjacent grouping images with other MRI images, and obtain second similarity data;

[0087] S204, screen out the MRI images in the second similarity data that exceed the second similarity threshold to obtain a processing image set.

[0088] Please refer to Figures 1-4 In this embodiment, the similarity of each MRI image in the two adjacent image grouping is calculated to obtain second similarity data; then, according to the preset second similarity threshold, the MRI images with similarity exceeding the second similarity threshold are screened out and included in the processing image set. The processing image set is composed of images with high similarity in content across groups, which represents the key continuity content in the image sequence and helps to more accurately capture structural changes and anatomical features in the subsequent analysis stage.

[0089] Although the similarity grouping has been completed in the previous stage, there may still be redundancy of image content between groups. By cross-group similarity screening, this step can eliminate images with insignificant differences between groups, further compressing the data volume and ensuring that only the most valuable images for analysis are retained.

[0090] Based on the inter-graph difference, it is ensured that the selected images are more likely to contain these changes, improving the ability of the image analysis model to capture key features.

[0091] By focusing on the information change area to construct the processing image set, the number of input images can be significantly reduced, thereby reducing the model input overhead, improving the processing speed, and reducing the learning bias of the model due to redundant data, and improving the accuracy of segmentation, classification, detection and other tasks.

[0092] In at least one embodiment of the present application, the specific steps of performing image analysis on the processing image set to obtain the MRI image analysis result include:

[0093] S205, normalizing the processing image set to obtain a normalized image set;

[0094] S206, performing noise removal processing on the normalized image set to obtain a denoised image set;

[0095] S207, performing ROI region extraction on the denoised image set to obtain an ROI image set;

[0096] S208, performing tissue segmentation on the ROI image set to obtain a segmented image set;

[0097] S209, performing image analysis on the segmented image set to obtain an MRI image analysis result.

[0098] Please refer to Figures 1-4 In this embodiment, the workstation performs grayscale normalization processing on each MRI image in the processed image set, maps the image pixel values to the interval [0, 1] or [-1, 1], eliminates the grayscale distribution difference caused by different scanning conditions (such as TR, TE, magnetic field strength), and realizes data unification in a multi-center, multi-device data environment, so that the subsequent model input remains consistent.

[0099] Noise suppression algorithms are used on the normalized image set to reduce artifacts and background noise in the MRI image. Non-local mean filtering (NLM), bilateral filtering, or deep convolutional network-based denoising models are used in combination with artifact detection algorithms to identify abnormal regions such as stripes and ghosting in the image and correct or mask them.

[0100] The signal-to-noise ratio of the image is improved, and the loss of details is reduced, providing accurate edge information for subsequent ROI extraction and boundary segmentation.

[0101] From the denoised image, the target anatomical region or lesion region is identified, and an ROI image set is generated. Based on template matching, image-guided segmentation, or deep learning methods, the key regions are located, the non-target regions are reduced, the lesion or anatomical structure is precisely focused, and the model training and analysis efficiency is improved.

[0102] In the ROI region, different tissue types are segmented at the semantic or instance level using neural network models such as U-Net, ResUNet, and TransUNet to achieve fine tissue boundary recognition, output the label of each pixel, and identify the tissue category to which it belongs, such as tumor tissue, normal tissue, necrotic area, and blood vessels.

[0103] The workstation performs quantitative analysis on the segmented image to extract clinical or research-related indicators and outputs the MRI image analysis result.

[0104] Specific operations include: voxel counting method to calculate lesion volume, space-occupying proportion, morphological feature analysis (such as long diameter, short diameter, edge regularity), time sequence image contrast analysis (such as preoperative and postoperative changes, progression speed), output chart, heat map or structured report for assisting doctor decision or model training input.

[0105] In at least one embodiment of the present application, the specific steps of denoising the normalized image set to obtain the denoised image set include:

[0106] Denoising the normalized image set to obtain a denoised image set;

[0107] Correcting the artifacts of the denoised image set to obtain the denoised image set.

[0108] In this embodiment, the workstation uses a non-local mean filter denoising algorithm to process each normalized image to obtain a denoised image set, and corrects the motion artifacts of the denoised image set to obtain a denoised image set.

[0109] Denoising significantly reduces background interference, making image details clearer, providing reliable input for subsequent segmentation and recognition tasks, correcting image distortion, restoring tissue contours, preventing models from misjudging boundary positions during segmentation, and making the denoised image more stable, avoiding models from learning false features during training / inference, improving analysis accuracy, cleaning image data, reducing processing complexity in subsequent steps, and speeding up ROI extraction and segmentation.

[0110] In at least one embodiment of the present application, the medical image AI processing method further comprises:

[0111] S301, establishing a standard image set;

[0112] Generating a calibration image from the processed image set, the grouped images and the standard image set;

[0113] S303, performing image analysis on the calibration image to obtain an MRI image calibration result.

[0114] Please refer to Figures 1-4 In this embodiment, the AI workstation pre-processes standard reference images, and the standard image set comes from normal samples and healthy people. The standard reference images cover standardized MRI image templates of different parts, sequences and tissue structures, with annotation or structural boundary information.

[0115] The current processing image set and the grouped images are compared with the standard image set, the difference score between the images is calculated, and the image with the highest difference score is selected as the "calibration image", which represents the most significant change area (such as lesion mutation, morphological abnormalities, etc.) in this examination.

[0116] Finally, the workstation performs enhanced image analysis on the generated calibration image to obtain a calibration result, which includes the most significant change area in this examination and the morphological data of the lesion.

[0117] After introducing the standard image set, images under different scanning conditions can be pulled back to a unified reference, improving the objectivity and reproducibility of the analysis. The calibration image can be used to directly show the difference with the standard, making it easy for doctors to intuitively understand the judgment basis of the AI model. The calibration process can quantify the anatomical differences between individuals and standards, which helps disease staging, typing or prediction modeling.

[0118] If it is found that a certain image deviates from the standard too much, the workstation can automatically adjust the segmentation model or analysis parameters to enhance the adaptive ability of the workstation.

[0119] In at least one embodiment of the present application, the specific steps of generating a calibration image from the processing image set, the grouped images and the standard image set include:

[0120] S3021, selecting an MRI image with the largest difference from the standard image set from the processing image set to obtain a difference image;

[0121] S3022, selecting two adjacent frame images of the difference image from the grouped images;

[0122] S3023, generating a calibration image according to the two adjacent frame images and the difference image.

[0123] Please refer to Figures 1-4 In this embodiment, image comparison (such as structural similarity SSIM, feature distribution distance, etc.) is performed on each MRI image in the processing image set and the corresponding reference image in the standard image set, the difference score between the images is calculated, and the image with the highest difference score is selected as the difference image. The image represents the most significant change area (such as lesion mutation, morphological abnormalities, etc.) in this examination.

[0124] Automatic recognition of key images, without manual intervention, precise positioning of the most abnormal area, and enhanced AI perception of key structural changes.

[0125] In the image group where the difference image is located, find its previous frame and next frame images (which can be indexed by image number, position or spatial coordinates), and introduce these two images as supplementary information into the subsequent calibration process.

[0126] A calibration image is generated based on two adjacent frame images and a difference image. The three images are aligned with a standard image template. A unified structure is generated through a deformation field or affine transformation. The texture or edge information of the upper and lower frames is fused to compensate for the difference image, retain the difference area, and fuse the contextual structural information to improve comparability and anatomical integrity. A deep model (such as an image reconstruction network) is used to learn the transition relationship between the difference image and the context to generate a clear and aligned calibration image.

[0127] The differential image recognition algorithm enables the workstation to automatically lock on to the most clinically significant images without the need for manual labeling intervention, thereby improving the intelligence of the workstation.

[0128] The introduction of upper and lower frame images makes up for the lack of information in a single frame, making the calibration image smoother and more structurally continuous, and enhancing the clinical readability of the image.

[0129] Use standard image sets and difference images to generate calibrated images in a unified space, improving the feasibility of image comparison across devices and centers.

[0130] The calibrated image has good boundaries and clear organizational structure, and can serve as a more stable and accurate input for the AI ​​model, improving the robustness of subsequent processing links.

[0131] Doctors or systems can make intuitive comparisons between standard images and calibrated images to accurately identify features such as structural abnormalities and lesion development trends.

[0132] A workstation 100, applied to any one of the above-described medical image AI processing methods, includes:

[0133] An image acquisition module 110 is used to acquire MRI images;

[0134] A similarity detection module 120 performs similarity detection on the first image set to generate grouped images;

[0135] A screening module 130 is configured to screen out processing images from two adjacent grouped images and generate a processing image set;

[0136] The image analysis module 140 performs image analysis on the processed image set to generate MRI image analysis results.

[0137] Please refer to Figures 4-5In the embodiment, the examination data of the patient is imported from a PACS system, a medical image archiving platform or directly from an MRI scanning device by an image acquisition module to obtain an MRI image, a first image set is generated according to the MRI image; similarity detection is performed on each two images in the first image set by a similarity detection module (such as structural similarity SSIM, feature space cosine distance, convolution feature comparison, etc.), and images with high similarity are classified into the same group according to a similarity threshold, and a plurality of grouped image sets are output.

[0138] Any two adjacent grouped images are compared by a screening module, the similarity between the two grouped images is further calculated, and images with a similarity exceeding a threshold are selected as a processing image set. The processing image set is subjected to a standardization processing procedure, including normalization, denoising, ROI extraction, tissue segmentation, calling of a pre-trained AI model (such as U-Net, ResNet, etc.) to extract image features, judgment of a lesion area, measurement of a volume, generation of a structured report, and output of image analysis results, including an anatomical structure segmentation map, a lesion positioning map, volume statistics, and a pathological score.

[0139] The data acquisition efficiency is improved, multi-source and multi-modal image unified input is supported, image content is structured and classified, redundant data is compressed, key images are selected for analysis, the amount of calculation is reduced, the response speed is improved, and automatic and accurate image intelligent analysis is realized.

[0140] A computer device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the following steps:

[0141] Obtaining an MRI image to be processed to generate a first image set;

[0142] Performing similarity detection on the images of the first image set, grouping a plurality of MRI images in the first image set according to similarity, and obtaining grouped images;

[0143] Selecting processing images from the two adjacent grouped images according to a similarity threshold to obtain a processing image set;

[0144] Performing image analysis on the processing image set to obtain an MRI image analysis result.

[0145] Please refer to Figure 6In the embodiment, the examination data of the patient is imported from a PACS system, a medical image archiving platform, or directly from an MRI scanning device by an image acquisition module, an MRI image is obtained, a first image set is generated according to the MRI image, and similarity calculation (such as structural similarity SSIM, feature space cosine distance, convolution feature comparison, etc.) is performed on each two images in the first image set by a similarity detection module. According to a similarity threshold, images with high similarity are classified into the same group, and a plurality of grouped image sets are output.

[0146] By screening the two adjacent grouped images, the similarity between the two grouped images is further calculated, and the images with similarity exceeding the threshold are selected as a processing image set. The processing image set is subjected to a standardization processing procedure, including normalization, denoising, ROI extraction, tissue segmentation, calling of a pre-trained AI model (such as U-Net, ResNet, etc.) to extract image features, judgment of a lesion region, measurement of a volume, generation of a structured report, and output of image analysis results, including an anatomical structure segmentation map, a lesion positioning map, volume statistics, and pathological scoring.

[0147] The data acquisition efficiency is improved, multi-source and multi-modal image unified input is supported, image content is structured and classified, redundant data is compressed, key images are selected for analysis, the amount of calculation is reduced, the response speed is improved, and automatic and accurate image intelligent analysis is realized.

[0148] Figure 6 An internal structure diagram of a computer device in an embodiment is shown. The computer device can be a terminal or a server. As shown in Figure 6 The computer device includes a processor, a memory, and a network interface connected by a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and can also store a computer program. When the computer program is executed by the processor, the processor can implement the medical image AI processing method. The internal memory can also store a computer program. When the computer program is executed by the processor, the processor can execute the medical image AI processing method. Those skilled in the art can understand that Figure 6 The structure shown in the embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0149] In one embodiment, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the processor performs the following steps:

[0150] Acquiring an MRI image to be processed to generate a first image set;

[0151] performing similarity detection on images in the first image set, grouping the plurality of MRI images in the first image set according to similarity, and obtaining grouped images;

[0152] Selecting a processing image from two adjacent grouped images according to a similarity threshold to obtain a processing image set;

[0153] Perform image analysis on the processed image set to obtain MRI image analysis results.

[0154] Please refer to Figure 6 In this embodiment, the image acquisition module imports the patient's examination data from the PACS system, the medical image archiving platform or directly from the MRI scanning device to obtain MRI images, and generates a first image set based on the MRI images; the similarity detection module calculates the similarity of each image in the first image set (such as structural similarity SSIM, feature space cosine distance, convolution feature comparison, etc.), and according to the similarity threshold, the images with higher similarity are classified into the same group, and multiple grouped image sets are output.

[0155] The screening module is used to compare any two adjacent grouped images. The image similarity between the two grouped images is further calculated, and images with a similarity exceeding a threshold are screened out as the processing image set. The processed image set is subjected to a standardized processing flow, including normalization, denoising, ROI extraction, and tissue segmentation. Pre-trained AI models (such as U-Net, ResNet, etc.) can be called to extract image features, determine the lesion area, measure the volume, generate a structured report, and output image analysis results, including anatomical structure segmentation maps, lesion localization maps, volume statistics, pathology scores, etc.

[0156] Improve data acquisition efficiency, support unified input of multi-source and multi-modal images, realize structured classification of image content, compress redundant data, select and analyze key images, reduce the amount of calculation, improve response speed, and realize automatic and accurate image intelligent analysis.

[0157] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0158] The technical features of the above embodiments can be combined in any way. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0159] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A medical image AI processing method, characterized in that: The medical image AI processing method includes: Acquiring an MRI image to be processed to generate a first image set; performing similarity detection on images in the first image set, grouping the plurality of MRI images in the first image set according to similarity, and obtaining grouped images; performing similarity calculation on all MRI images in the first image set to obtain first similarity data; Obtaining a first similarity threshold, and performing similarity grouping on each MRI image in the first image set based on the first similarity threshold and the first similarity data to obtain a grouped image; Selecting a processing image from two adjacent grouped images according to a second similarity threshold to obtain a processing image set; calculating the similarity between each MRI image and other MRI images in two adjacent grouped images to obtain second similarity data; Filtering out MRI images exceeding a second similarity threshold from the second similarity data to obtain a processed image set; Perform image analysis on the processed image set to obtain MRI image analysis results.

2. The medical image AI processing method according to claim 1, characterized in that: The specific steps of performing image analysis on the processed image set to obtain MRI image analysis results include: performing normalization processing on the processed image set to obtain a normalized image set; Performing denoising on the normalized image set to obtain a denoised image set; Performing ROI region extraction on the image set after the noise removal process to obtain an ROI image set; performing tissue segmentation on the ROI image set to obtain a segmented image set; Image analysis is performed on the segmented image set to obtain MRI image analysis results.

3. The medical image AI processing method according to claim 2, characterized in that: The specific steps of performing denoising on the normalized image set to obtain a denoised image set include: Denoising the normalized image set to obtain a denoised image set; Performing artifact correction on the denoised image set to obtain a denoised image set.

4. The medical image AI processing method according to claim 1, characterized in that: The medical image AI processing method further includes: Establish a standard image set; generating a calibration image from the processed image set, the grouped images, and the standard image set; Perform image analysis on the calibration image to obtain an MRI image calibration result.

5. The medical image AI processing method according to claim 4, characterized in that: The specific steps of generating a calibration image from the processed image set, the grouped images and the standard image set include: Screening out the MRI image with the greatest difference from the standard image set from the processed image set to obtain a difference image; Filtering out two adjacent frames of the difference image from the grouped images; A calibration image is generated according to the two adjacent frame images and the difference image.

6. A workstation, characterized in that: Applied to the medical image AI processing method according to any one of claims 1 to 5, the workstation comprises: An image acquisition module, used for acquiring MRI images; a similarity detection module, performing similarity detection on the first image set to generate grouped images; A screening module, screening out a processing image from two adjacent grouped images and generating a processing image set; The image analysis module performs image analysis on the processed image set and generates MRI image analysis results.

7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

Images