AI automatic monitoring method and system for telemedicine images

Abstract

The invention relates to the field of data processing, in particular to an AI automatic monitoring method and system for remote medical images, and the method comprises the steps: extracting a focus suspected region in medical image data through the image features of the medical image data; then cutting the medical image data to obtain a target image block containing the focus suspected area and a non-target image block not containing the focus suspected area; during compression, compressing the non-target image blocks to obtain a compressed image; the compressed image and the target image block are packaged and sent to a remote execution end; and restoring the compressed image and the target image block at the remote execution end to obtain a restored image, and identifying the restored image by using an AI identification model of the remote execution end to obtain an identification result. When the image is compressed, the focus suspected area and the non-focus suspected area are distinguished, so that more disease information is reserved, the image is effectively compressed and transmitted, the transmission speed can be increased, and more useful information can be reserved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and specifically to an AI automatic monitoring method and system for remote medical images. Background Art

[0002] AI technology can analyze a large amount of medical image data through deep learning algorithms, identify subtle features that may be overlooked by the human eye, and thus help doctors diagnose diseases more accurately. For example, in the fields of early tumor detection, cardiovascular disease assessment, etc., AI can provide more accurate judgments.

[0003] However, in the prior art, when using AI technology to remotely (such as cloud technology) automatically identify medical images, in order to improve the transmission speed of medical images, the image is generally compressed and then transmitted. The compressed image will lose some information, resulting in a decrease in the correct rate of model recognition. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an AI automatic monitoring method and system for remote medical images to solve the problems in the background art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] An AI automatic monitoring method for remote medical images of the present invention includes the steps of:

[0007] Obtain the medical image data of the patient;

[0008] Extract the suspected lesion area in the medical image data based on the image features of the medical image data, where the image features are density features, edge contour features, morphological features or contrast features;

[0009] Crop the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area;

[0010] Compress the non-target tile to obtain a compressed image; and package and send the compressed image and the target tile to the remote execution end;

[0011] Restore the compressed image and the target tile at the remote execution end to obtain a restored image, and use the AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

[0012] In an embodiment of the present application, extracting the suspected lesion area in the medical image data based on the density feature of the medical image data includes:

[0013] Gray-scale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre;

[0014] Normalize the gray values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre);

[0015] Calculate the average value and standard deviation of the gray values of all pixel points in the normalized image nor(image_Pre) to obtain a global reference mean gray_A and a global reference standard deviation gray_σ;

[0016] Slide along the normalized image nor(image_Pre) based on a pre-constructed sliding window, and calculate the local mean gray_A' of the gray values of all pixel points within the sliding window during the sliding;

[0017] Calculate the difference gray_A - gray_A' between the global reference mean gray_A and the local mean gray_A', and when the difference gray_A - gray_A' satisfies: |gray_A - gray_A'| ≥ n×gray_σ, determine that the area within the corresponding sliding window is a suspected lesion area, where n is a magnification threshold.

[0018] In an embodiment of the present application, extracting a suspected lesion area in the medical image data based on the edge contour features of the medical image data includes:

[0019] Gray-scale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre;

[0020] Perform binaryzation on the preprocessed image image_Pre based on an adaptive binaryzation algorithm to obtain a binary image bin;

[0021] Extract all contour features in the binary image bin;

[0022] Extract the lengths of all contour features, and use the contour features within a preset length range as the first target contour features;

[0023] Perform a fast Fourier transform on the edge contour of the first target contour feature to obtain the frequency domain distribution information of the first target contour feature;

[0024] Extract high-frequency components with frequencies greater than a frequency threshold from the frequency domain distribution information of the first target contour feature, and calculate the proportion of the high-frequency components; when the proportion of the high-frequency components is greater than a preset proportion threshold, determine that the target contour feature is a suspected contour of a disease with burrs;

[0025] Calculate the bounding box of the suspected contour of the disease condition, and use the area within the bounding box as the suspected lesion area.

[0026] In one embodiment of the present application, extracting the suspected lesion area in the medical image data based on the morphological features of the medical image data includes:

[0027] Grayscale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre;

[0028] Perform binary processing on the preprocessed image image_Pre based on an adaptive binary algorithm to obtain a binary image bin;

[0029] Extract all contour features in the binary image bin;

[0030] Perform morphological processing on all contour features to obtain a closed contour, where the morphological processing includes one or a combination of corrosion, dilation, opening operation, and closing operation;

[0031] Calculate the area of the closed contour, and use the contour features within a preset area range as the second target contour features;

[0032] Calculate the roundness and aspect ratio of the second target contour features, and use the second target contour features that meet the preset roundness screening range and length ratio screening range as the suspected disease condition contours;

[0033] Calculate the bounding box of the suspected disease condition contour, and use the area within the bounding box as the suspected lesion area.

[0034] In one embodiment of the present application, extracting the suspected lesion area in the medical image data based on the contrast feature of the medical image data includes:

[0035] Grayscale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre;

[0036] Normalize the gray values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre);

[0037] Segment the normalized image nor(image_Pre) based on an adaptive threshold segmentation method to obtain a background area and a foreground area;

[0038] Extract the edge contour features of each foreground area, and perform morphological processing on each edge contour feature to obtain a foreground closed contour;

[0039] Calculate the grayscale mean in the background region to obtain the reference grayscale value gray_Ar; and calculate the grayscale mean of the pixel points within each of the foreground closed contours to obtain the candidate region grayscale mean gray_Ac;

[0040] Calculate the ratio gray_Ac / gray_Ar of the candidate region grayscale mean gray_Ac to the reference grayscale value gray_Ar, and when the ratio gray_Ac / gray_Ar is greater than a preset ratio threshold, determine that the foreground closed contour is a suspected lesion contour;

[0041] Calculate the bounding box of the suspected lesion contour, and use the region within the bounding box as the suspected lesion region.

[0042] In an embodiment of the present application, cropping the medical image data to obtain a target tile containing the suspected lesion region and a non-target tile not containing the suspected lesion region includes:

[0043] Crop the medical image data into multiple n×n tiles, and number each tile;

[0044] Calculate the proportion of the suspected lesion region in each tile, and use the tile with a proportion greater than a preset proportion threshold as the target tile, and other tiles as non-target tiles.

[0045] In an embodiment of the present application, restoring the compressed image and the target tile at the remote execution end to obtain a restored image includes:

[0046] Perform image enhancement on the compressed image to obtain an enhanced image;

[0047] Based on the numbering, splice all the target tiles and the enhanced image to obtain a restored image.

[0048] The present application also provides an AI automatic monitoring system for remote medical images, including:

[0049] An acquisition module, configured to acquire medical image data of a patient;

[0050] A lesion determination module, configured to extract a suspected lesion region in the medical image data based on the image features of the medical image data, where the image features are density features, edge contour features, morphological features, or contrast features;

[0051] A cropping module, configured to crop the medical image data to obtain a target tile containing the suspected lesion region and a non-target tile not containing the suspected lesion region;

[0052] A compression and transmission module, configured to compress the non-target tiles to obtain a compressed image; and package and send the compressed image and the target tiles to a remote execution end;

[0053] A restoration and recognition module, configured to restore the compressed image and the target tiles at the remote execution end to obtain a restored image, and use an AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

[0054] This application also provides an electronic device, including: a processor and a memory;

[0055] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the method described above.

[0056] This application also provides a computer-readable storage medium, on which a computer program is stored, characterized in that: when the computer program is executed by a processor, the method described above is implemented.

[0057] The beneficial effects of the present invention are: an AI automatic monitoring method and system for remote medical images of the present invention extract suspected lesion areas in medical image data through the image features of medical image data; then crop the medical image data to obtain target tiles containing suspected lesion areas and non-target tiles not containing suspected lesion areas; during compression, compress the non-target tiles to obtain a compressed image; and package and send the compressed image and the target tiles to a remote execution end; at the remote execution end, restore the compressed image and the target tiles to obtain a restored image, and use an AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result. This application distinguishes between suspected lesion areas and non-lesion areas when compressing images, thereby retaining more disease information and effectively compressing and transmitting images, which can not only improve the transmission speed but also retain more useful information, thus effectively improving the recognition accuracy of the remote AI model. Description of the Drawings

[0058] The present invention will be further described below with reference to the drawings and embodiments:

[0059] Figure 1 is a flowchart of an AI automatic monitoring method for remote medical images shown in an embodiment of this application;

[0060] Figure 2 is a structural diagram of an AI automatic monitoring system for remote medical images shown in an embodiment of this application. Detailed Embodiments

[0061] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0062] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show the layers related to the present invention rather than being drawn according to the number, shape and size of the layers in actual implementation. In actual implementation, the type, quantity and proportion of each layer may be changed arbitrarily, and the layer layout may also be more complicated.

[0063] In the following description, numerous details are discussed to provide a more thorough explanation of embodiments of the present invention; however, it is apparent to one skilled in the art that embodiments of the present invention may be practiced without these specific details.

[0064] Figure 1 is a flow chart of an AI automatic monitoring method for remote medical imaging shown in one embodiment of the present application, such as Figure 1 As shown, an AI automatic monitoring method for remote medical images in this embodiment may include the following steps:

[0065] S110, obtaining medical imaging data of the patient;

[0066] Among them, the medical imaging data in this application can be medical images such as X-ray imaging (Radiography), computed tomography (CT, Computed Tomography), magnetic resonance imaging (MRI, Magnetic Resonance Imaging), and ultrasound imaging (Ultrasound).

[0067] S120, extracting a suspected lesion region in the medical image data based on an image feature of the medical image data, wherein the image feature is a density feature, an edge contour feature, a morphological feature, or a contrast feature;

[0068] In this application, in order to retain more information about the suspected lesion area, the idea of ​​block compression is adopted, that is, the original image information of the suspected lesion area is retained, and other less important areas are compressed. In this way, the original image information of the lesion is retained as much as possible, the size of the image data is reduced, and the transmission speed is improved.

[0069] This application uses image processing technology to extract suspected lesion regions, mainly extracting suspected lesion regions through density features, edge contour features, morphological features, and contrast features. This is because:

[0070] In CT scans, lesions may appear as higher or lower density than surrounding tissues. In MRI images, different tissue types will exhibit different signal intensities, and the lesion area may show abnormal high or low signals.

[0071] The edges of the lesion area may be irregular, blurred, or have "burr-like" protrusions, which is usually a sign of a malignant tumor. In contrast, benign lesions tend to have clearer and smoother boundaries.

[0072] The size, shape, and growth pattern (such as uniformity) of the lesion may also provide diagnostic clues. For example, certain types of cancer tend to form masses of specific shapes.

[0073] After using a contrast agent, the enhancement pattern of the lesion area in the image can provide additional information. For example, rapid and strong enhancement may be a sign of a highly vascular tumor.

[0074] In an embodiment of this application, extracting the suspected lesion region in the medical image data based on the density feature of the medical image data includes:

[0075] S1201, grayscale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre;

[0076] If the image is color, first convert it to a grayscale image. This can be done by simply averaging or weighted averaging the RGB values. In addition, use a Gaussian filter to smooth the image and reduce the influence of noise.

[0077] S1202, perform normalization processing on the grayscale values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre);

[0078] In this embodiment, the minimum-maximum normalization method is used to convert the grayscale values of the similarity to the range of [0,1] to facilitate subsequent processing.

[0079] S1203, calculate the average value and standard deviation of the grayscale values of all pixel points in the normalized image nor(image_Pre) to obtain a global reference mean gray_A and a global reference standard deviation gray_σ;

[0080] The gray value of each pixel point in the CT image actually reflects the absorption degree of the corresponding voxel (volume element) to X-rays. The higher the absorption degree, the greater the density of the substance in this area, and correspondingly, it will be displayed with a higher gray value (brighter area) on the CT image; on the contrary, the area with a low absorption degree corresponds to a lower gray value (darker area). For example, in a CT image, bones appear as white or bright areas due to their high density, fat and soft tissues show different shades of gray according to their different densities, while the lungs appear relatively dark because they contain a large amount of air, and lesions have a higher reflection density and are reflected as higher gray values in the image. Therefore, this application uses the difference between the global and local gray levels to extract the suspected lesion area.

[0081] S1204, slide along the normalized image nor(image_Pre) based on a pre-constructed sliding window, and calculate the local mean gray_A' of the gray values of all pixel points within the sliding window during the sliding;

[0082] S1205, calculate the difference gray_A - gray_A' between the global reference mean gray_A and the local mean gray_A', and when the difference gray_A - gray_A' satisfies: |gray_A - gray_A'| ≥ n × gray_σ, determine that the area within the corresponding sliding window is a suspected lesion area, where n is the magnification threshold.

[0083] In this embodiment, by calculating the difference gray_A - gray_A' between the global reference mean gray_A and the local mean gray_A', if the difference is large, the area within the sliding window may be a lesion area.

[0084] In addition, if the areas corresponding to adjacent sliding windows are both suspected lesion areas, merge them.

[0085] In an embodiment of this application, extracting the suspected lesion area from the medical image data based on the edge contour features of the medical image data includes:

[0086] S1211, perform graying and Gaussian filtering on the medical image data to obtain a preprocessed image image_Pre;

[0087] Convert the color image to a gray image to reduce the computational complexity and focus on the brightness information. And smooth the image through a Gaussian filter to remove noise while trying to retain the edge information.

[0088] S1212, perform binarization processing on the preprocessed image image_Pre based on an adaptive binarization algorithm to obtain a binarized image bin;

[0089] S1213, extract all contour features in the binary image bin;

[0090] Specifically, use an edge detection algorithm (such as the Canny or Sobel operator) to find the edges in the image, and then use a contour detection algorithm (such as the findContours function in OpenCV) to extract all contours.

[0091] S1214, extract the lengths of all contour features, and use the contour features within a preset length range as the first target contour features;

[0092] Filter all the extracted contours, and only retain those contours whose lengths fall within the preset range as the first target contour features. The purpose is to exclude too small or too large contours and retain the areas that conform to the lesion size range.

[0093] S1215, perform a fast Fourier transform on the edge contours of the first target contour features to obtain the frequency domain distribution information of the first target contour features;

[0094] Some lesion contours have a phenomenon of burrs. The edges of normal organs or tissues are smooth. Using this phenomenon, the edges with burrs can be determined as the edges of lesions.

[0095] Apply a fast Fourier transform to the edge contours of the first target contour features, convert it from the spatial domain to the frequency domain, and obtain the frequency domain distribution information. This helps to analyze the frequency components of the contour and thus find the high-frequency burr contours.

[0096] S1216, extract the high-frequency components with frequencies greater than the frequency threshold from the frequency domain distribution information of the first target contour features, and calculate the proportion of the high-frequency components; when the proportion of the high-frequency components is greater than the preset proportion threshold, determine that the target contour feature is a suspected contour of a disease with burrs;

[0097] Normal tissue contours are usually relatively smooth, while the lesion area may have more complex boundaries, resulting in a higher proportion of high-frequency components. Therefore, by calculating the proportion of high-frequency components and determining the contours with a higher proportion of high-frequency parts as the contours of the lesion area with lesions.

[0098] S1217, calculate the bounding box of the suspected disease contour, and use the area within the bounding box as the suspected lesion area.

[0099] The bounding box can help to more quickly locate the disease area in the subsequent analysis process.

[0100] In an embodiment of the present application, extracting the suspected lesion area in the medical image data based on the morphological features of the medical image data includes:

[0101] S1221, perform grayscale conversion and Gaussian filtering on the medical image data to obtain a preprocessed image image_Pre;

[0102] S1222, perform binaryzation on the preprocessed image image_Pre based on an adaptive binarization algorithm to obtain a binary image bin;

[0103] S1223, extract all contour features in the binary image bin;

[0104] S1224, perform morphological processing on all the contour features to obtain closed contours, where the morphological processing includes one or a combination of erosion, dilation, opening operation, and closing operation;

[0105] The purpose of morphological processing is to close the contours to facilitate the extraction of morphological features of the closed contours.

[0106] S1225, calculate the areas of the closed contours, and use the contour features within a preset area range as the second target contour features;

[0107] The purpose of area screening is to exclude contours that are too small or too large and retain regions within the size range of the lesions. Remove regions that are too small or too large because they are less likely to be true lesions.

[0108] S1226, calculate the roundness and aspect ratio of the second target contour features, and use the second target contour features that meet the preset roundness screening range and length ratio screening range as the disease suspected contours;

[0109] The contours of some lesions may exhibit a high roundness and a specific length ratio, such as a benign tumor. Therefore, calculate its roundness and aspect ratio (the ratio of width to length). Screen out contour features that are more likely to be diseases according to the preset roundness and aspect ratio ranges.

[0110] S1227, calculate the bounding box of the disease suspected contours, and use the region within the bounding box as the disease suspected region.

[0111] In an embodiment of the present application, extracting the disease suspected region in the medical image data based on the contrast feature of the medical image data includes:

[0112] S1231, perform grayscale conversion and Gaussian filtering on the medical image data to obtain a preprocessed image image_Pre;

[0113] S1232, perform normalization processing on the gray values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre);

[0114] S1233, Segment the normalized image nor(image_Pre) based on an adaptive threshold segmentation method to obtain a background region and a foreground region;

[0115] For some medical images, the gray-scale contrast between the lesion region and the background region is large. Therefore, an adaptive threshold segmentation method (such as Adaptive Thresholding) is used to process the normalized image. This method dynamically adjusts the threshold according to the average intensity of each pixel neighborhood and is especially suitable for images with uneven illumination. This step segments the image into a background region and a foreground region.

[0116] S1234, Extract the edge contour features of each foreground region and perform morphological processing on each edge contour feature to obtain a foreground closed contour;

[0117] Perform morphological processing (such as erosion, dilation, opening operation, and closing operation) on all the extracted edge contour features to obtain a foreground closed contour. These operations help clean up the contour, fill small holes, disconnect the adhered parts, and make the contour smoother and more complete.

[0118] S1235, Calculate the gray-scale mean value in the background region to obtain a reference gray-scale value gray_Ar; and calculate the gray-scale mean value of the pixel points within each foreground closed contour to obtain a candidate region gray-scale mean value gray_Ac;

[0119] S1236, Calculate the ratio gray_Ac / gray_Ar of the candidate region gray-scale mean value gray_Ac to the reference gray-scale value gray_Ar, and when the ratio gray_Ac / gray_Ar is greater than a preset ratio threshold, determine that the foreground closed contour is a suspected lesion contour;

[0120] Calculate the ratio of the gray-scale mean value of the pixel points within each foreground closed contour to the gray-scale mean value of the background region. If this ratio is greater than a preset ratio threshold, then determine that the corresponding foreground closed contour is a suspected lesion contour. This is because the lesion region usually has gray-scale characteristics different from those of the surrounding normal tissues, and this difference can be quantified by the gray-scale mean value ratio.

[0121] S1237, Calculate the bounding box of the suspected lesion contour and use the region within the bounding box as the suspected lesion region.

[0122] S130, Crop the medical image data to obtain a target tile containing the suspected lesion region and a non-target tile not containing the suspected lesion region;

[0123] For the suspected lesion regions extracted in the above process, the present application crops the medical image data, thereby dividing the image. Specifically, it includes:

[0124] S131, crop the medical image data into multiple n×n tiles, and number each tile;

[0125] Specifically, the size of the tile can be fixed (such as 64x64 pixels, 128x128 pixels, etc.), or it can be dynamically adjusted according to actual needs. If the image boundary cannot be completely divided evenly, the edge part can be filled (padding).

[0126] In addition, assign a unique number to each tile for subsequent splicing and restoration. The numbering method can be in row-column order (such as numbering from left to right and from top to bottom in sequence), or other logical rules.

[0127] S132, calculate the proportion of the suspected lesion region in each tile, and use the tiles with a proportion greater than the preset proportion threshold as target tiles, and other tiles as non-target tiles.

[0128] The proportion is determined by calculating the number of pixel points, so that some regions with fewer suspected lesion regions are also compressed as background regions, and this process will not affect subsequent restoration and recognition.

[0129] S140, compress the non-target tiles to obtain a compressed image; and package and send the compressed image and the target tiles to the remote execution end;

[0130] S150, restore the compressed image and the target tiles at the remote execution end to obtain a restored image, and use the AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

[0131] During restoration, first perform image enhancement on the compressed image to obtain an enhanced image; the enhanced image has more AI-generated details, which helps to restore the image. Then, based on the numbering, splice all the target tiles and the enhanced image to obtain a restored image.

[0132] For the uncompressed suspected disease regions, the present application retains the image information of the original image, which has richer and more accurate input information for the subsequent AI recognition model.

[0133] An AI automatic monitoring method for remote medical images of the present invention extracts a suspected lesion area in the medical image data through image features of the medical image data; then crops the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area; during compression, compresses the non-target tile to obtain a compressed image; and packages and sends the compressed image and the target tile to a remote execution end; at the remote execution end, restores the compressed image and the target tile to obtain a restored image, and uses the AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result. In this application, the suspected lesion area and the non-lesion area are distinguished during image compression, so as to retain more disease information and effectively compress and transmit the image, which can not only improve the transmission speed but also retain more useful information, thereby effectively improving the recognition accuracy of the remote AI model.

[0134] As Figure 2 shown, this application also provides an AI automatic monitoring system for remote medical images, including:

[0135] An acquisition module, configured to acquire medical image data of a patient;

[0136] A lesion determination module, configured to extract a suspected lesion area in the medical image data based on image features of the medical image data, where the image features are density features, edge contour features, morphological features, or contrast features;

[0137] A cropping module, configured to crop the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area;

[0138] A compression and transmission module, configured to compress the non-target tile to obtain a compressed image; and package and send the compressed image and the target tile to a remote execution end;

[0139] A restoration and recognition module, configured to restore the compressed image and the target tile at the remote execution end to obtain a restored image, and use the AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

[0140] An AI automatic monitoring system for remote medical images of the present invention extracts suspected lesion areas in medical image data through image features of the medical image data; then crops the medical image data to obtain target tiles containing suspected lesion areas and non-target tiles not containing suspected lesion areas; during compression, compresses the non-target tiles to obtain a compressed image; and packages and sends the compressed image and the target tiles to a remote execution end; at the remote execution end, restores the compressed image and the target tiles to obtain a restored image, and uses an AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result. In this application, when compressing an image, the suspected lesion area and the non-lesion area are distinguished, so as to retain more disease information and effectively compress and transmit the image, which can not only improve the transmission speed but also retain more useful information, thereby effectively improving the recognition accuracy of the remote AI model.

[0141] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0142] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory so that the terminal executes any method in this embodiment.

[0143] For the computer-readable storage medium in this embodiment, those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to a computer program. The foregoing computer program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk, or optical disk that can store program codes.

[0144] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store a computer program, the communication interface is used for communication, and the processor and the transceiver are used to run the computer program so that the electronic terminal executes each step of the above method.

[0145] In this embodiment, the memory may include a random access memory (Random Access Memory, abbreviated as RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0146] The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0147] In the above embodiments, although the present invention has been described in conjunction with specific embodiments of the present invention, many substitutions, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the foregoing description. The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims.

[0148] The above embodiments are only illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. An AI automatic monitoring method for telemedicine images, characterized in that, Including steps: Obtain the medical image data of the patient; Extract the suspected lesion area in the medical image data based on the image features of the medical image data, where the image features are density features, edge contour features, morphological features or contrast features; Crop the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area; Compress the non-target tile to obtain a compressed image; and package and send the compressed image and the target tile to the remote execution end; Restore the compressed image and the target tile at the remote execution end to obtain a restored image, and use the AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

2. The AI automatic monitoring method for remote medical images according to claim 1, characterized in that, Extracting the suspected lesion area in the medical image data based on the density feature of the medical image data includes: Grayscale and Gaussian filter the medical image data to obtain a preprocessed image image_Pre; Normalize the grayscale values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre); Calculate the average value and standard deviation of the grayscale values of all pixel points in the normalized image nor(image_Pre) to obtain a global reference mean gray_A and a global reference standard deviation gray_σ; Slide along the normalized image nor(image_Pre) based on a pre-constructed sliding window, and calculate the local mean gray_A' of the grayscale values of all pixel points in the sliding window during sliding; Calculate the difference gray_A - gray_A' between the global reference mean gray_A and the local mean gray_A', and when the difference gray_A - gray_A' satisfies: |gray_A - gray_A'| ≥ n × gray_σ, determine that the area within the corresponding sliding window is a suspected lesion area, where n is a magnification threshold.

3. The AI automatic monitoring method for remote medical images according to claim 1, characterized in that, Extracting the suspected lesion area in the medical image data based on the edge contour feature of the medical image data includes: Grayscale and Gaussian filter the medical image data to obtain a preprocessed image image_Pre; Perform binary processing on the preprocessed image image_Pre based on an adaptive binary algorithm to obtain a binary image bin; Extract all contour features in the binary image bin; Extract the lengths of all contour features, and use the contour features within a preset length range as the first target contour features; Perform a fast Fourier transform on the edge contour of the first target contour feature to obtain the frequency domain distribution information of the first target contour feature; Extract the high-frequency components with frequencies greater than the frequency threshold from the frequency domain distribution information of the first target contour feature, and calculate the proportion of the high-frequency components; when the proportion of the high-frequency components is greater than a preset proportion threshold, determine that the target contour feature is a suspected contour of a disease with burrs; Calculate the bounding box of the suspected contour of the disease condition, and use the area within the bounding box as the suspected lesion area.

4. An AI automatic monitoring method for remote medical images according to claim 1, characterized in that, Extract the suspected lesion area from the medical image data based on the morphological features of the medical image data, including: Grayscale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre; Perform binary processing on the preprocessed image image_Pre based on an adaptive binary algorithm to obtain a binary image bin; Extract all contour features in the binary image bin; Perform morphological processing on all contour features to obtain closed contours, where the morphological processing includes one or a combination of erosion, dilation, opening operation, and closing operation; Calculate the area of the closed contour, and use the contour features within a preset area range as the second target contour features; Calculate the roundness and aspect ratio of the second target contour features, and use the second target contour features that meet the preset roundness screening range and length ratio screening range as the suspected disease condition contours; Calculate the bounding box of the suspected disease condition contour, and use the area within the bounding box as the suspected lesion area.

5. The AI automatic monitoring method for remote medical images according to claim 1, characterized in that, Extract the suspected lesion area from the medical image data based on the contrast feature of the medical image data, including: Grayscale the medical image data and perform Gaussian filtering to obtain a preprocessed image image_Pre; Normalize the gray values of the pixel points in the preprocessed image image_Pre to obtain a normalized image nor(image_Pre); Segment the normalized image nor(image_Pre) based on an adaptive threshold segmentation method to obtain a background area and a foreground area; Extract the edge contour features of each foreground area, and perform morphological processing on each edge contour feature to obtain a foreground closed contour; Calculate the average gray value in the background area to obtain a reference gray value gray_Ar; and calculate the average gray value of the pixel points within each foreground closed contour to obtain a candidate area average gray value gray_Ac; Calculate the ratio gray_Ac / gray_Ar of the candidate area average gray value gray_Ac to the reference gray value gray_Ar, and when the ratio gray_Ac / gray_Ar is greater than a preset ratio threshold, determine that the foreground closed contour is a suspected disease condition contour; Calculate the bounding box of the suspected disease condition contour, and use the area within the bounding box as the suspected lesion area.

6. The AI automatic monitoring method for remote medical images according to claim 1, characterized in that Crop the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area, including: Crop the medical image data into multiple n×n tiles, and number each tile; Calculate the proportion of the suspected lesion area in each tile, and use the tiles with a proportion greater than a preset proportion threshold as target tiles, and other tiles as non-target tiles.

7. The AI automatic monitoring method for remote medical images according to claim 6, characterized in that, Restore the compressed image and the target tile at the remote execution end to obtain a restored image, including: Perform image enhancement on the compressed image to obtain an enhanced image; Stitch all the target tiles and the enhanced image based on the numbers to obtain a restored image.

8. An AI automatic monitoring system for remote medical images, characterized in that, Including: An acquisition module, configured to acquire medical image data of a patient; A lesion determination module, configured to extract a suspected lesion area in the medical image data based on the image features of the medical image data, where the image features are density features, edge contour features, morphological features, or contrast features; A cropping module, configured to crop the medical image data to obtain a target tile containing the suspected lesion area and a non-target tile not containing the suspected lesion area; A compression and transmission module, configured to compress the non-target tile to obtain a compressed image; and package and send the compressed image and the target tile to a remote execution end; A restoration and recognition module, configured to restore the compressed image and the target tile at the remote execution end to obtain a restored image, and use an AI recognition model at the remote execution end to recognize the restored image to obtain a recognition result.

9. An electronic device, comprising: A processor and a memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the method as claimed in claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the method as claimed in claims 1-7 is implemented.

Images