A medical image auxiliary analysis method and system based on artificial intelligence

Abstract

The present application discloses a medical image auxiliary analysis method and system based on artificial intelligence, which involves medical image analysis and image processing technology, including: obtaining a specified number of medical images of the target part; selecting a reference medical image from the previous medical images; selecting and aligning the reference boundaries of the specified number of medical images in a unified coordinate system; performing segmentation, calculating the similarity between the two sub-segments for the segmentation at the same position; determining the possible lesion area according to the trend of the calculated similarity; determining the shadow range of the possible lesion area, and extracting the shadow features of multiple medical images in chronological order; adding similarity tags to the shadow features, and inputting a pre-trained LSTM model to use the LSTM model to output the predicted shadow range to complete the auxiliary analysis. The present application combines image processing and artificial intelligence technology to realize the analysis of the local shadow area of ​​medical images to improve the diagnostic efficiency of medical images.

Description

Technical Field

[0001] The present application relates to the field of medical image analysis and image processing technology, and in particular to an artificial intelligence-based medical image auxiliary analysis method and system. Background Art

[0002] Medical imaging refers to the technology and process of obtaining images of internal tissues of the human body or a part of the human body in a non-invasive manner for medical treatment or medical research. It includes the following two relatively independent research directions: medical imaging systems and medical image processing. The former refers to the process of image formation, including research on imaging mechanisms, imaging equipment, imaging system analysis and other issues. The latter refers to further processing of the images that have been obtained, with the purpose of either restoring the original unclear image, or highlighting certain feature information in the image, or performing pattern classification on the image.

[0003] In the existing technology, the application of medical images is still quite rough, and doctors observe medical images with their naked eyes and then make judgments based on the observation results, which is relatively subjective. In particular, it is difficult to analyze the small changes in the shadow area of ​​the local area of ​​the medical image, resulting in low efficiency of medical image diagnosis. Summary of the invention

[0004] The embodiments of the present application provide a medical image auxiliary analysis method and system based on artificial intelligence, which combines image processing and artificial intelligence technology to realize the analysis of local shadow areas of medical images to improve the diagnostic efficiency of medical images.

[0005] The present application embodiment proposes an artificial intelligence-based medical image assisted analysis method, including:

[0006] Acquire a specified number of medical images of the target part according to the time sequence of the medical images;

[0007] Selecting a reference medical image from previous medical images, and establishing a unified coordinate system based on the reference medical image;

[0008] Select reference boundaries and align a specified number of medical images in the unified coordinate system;

[0009] Segment the aligned medical images according to uniform specifications;

[0010] For the segmentation at the same position, the similarity between the two sub-segments is calculated in chronological order;

[0011] According to the trend of the calculated similarity, the possible lesion area is determined;

[0012] Determine the shadow range of the possible lesion area, and extract shadow features of multiple medical images in chronological order;

[0013] A similarity tag is added to the shadow feature and input into a pre-trained LSTM model to utilize the LSTM model to output a predicted shadow range to complete auxiliary analysis.

[0014] Optionally, selecting a reference medical image from previous medical images, and establishing a unified coordinate system based on the reference medical image includes:

[0015] Using an edge detection algorithm to identify a first boundary of a target area from a prior medical image;

[0016] Calculating the clarity within a first boundary of the identified target region;

[0017] Selecting a previous medical image with the highest definition within the first boundary as a reference medical image;

[0018] And, based on the first boundary of the target area, reference points are selected to establish a unified coordinate system.

[0019] Optionally, selecting reference boundaries and aligning a specified number of medical images in the unified coordinate system includes:

[0020] Using edge detection algorithm to identify the second boundary of the target area in any medical image;

[0021] Establishing an association layer of a reference medical image and any medical image based on the first boundary and the second boundary, dividing the first boundary and the second boundary into a plurality of boundary segments, and calculating a matching degree between each boundary segment;

[0022] According to the calculated matching degree, the boundary segments of the first boundary and the second boundary with the highest matching degree are used as reference boundaries, and are translated and / or rotated so that the overlap degree of the reference boundary segments is the highest;

[0023] And, synchronously according to the associated layer, any medical image is translated and / or rotated according to the translation distance and / or rotation angle of the reference boundary segment to complete the alignment.

[0024] Optionally, segmenting the aligned medical images according to uniform specifications includes:

[0025] Determining a pixel area of ​​the target site in the reference medical image;

[0026] Determining a segmentation specification according to the pixel area;

[0027] The aligned medical image is segmented according to the determined segmentation specification, wherein the determined segmentation specification enables the target part to contain a plurality of sub-segments with a specified pixel area after segmentation.

[0028] Optionally, based on the trend of the calculated similarity, possible lesion areas are determined to include:

[0029] For the segmentation at the same position, the similarities of the two adjacent segmentations calculated are sorted according to the time order;

[0030] If the deviation of the similarity between two adjacent segmentations exceeds the set deviation range, or the deviation between the calculated similarity between any two segmentations at a later time and the calculated similarity between the initial two segmentations exceeds a preset threshold, it is determined to be a possible lesion segmentation;

[0031] The possible lesion area is determined based on the combined area segmented from each possible lesion.

[0032] Optionally, determining the shadow range of the possible lesion area includes:

[0033] For each possible lesion segmentation combined area, the transformation map is obtained by transformation according to the following formula:

[0034] ;

[0035] in, is the transformed pixel value, , , are the pixel values ​​of the red, green and blue channels of any pixel in the combined area respectively;

[0036] Perform secondary segmentation in the transformation graph;

[0037] Calculate the color mean of each secondary segmented sub-region on three channels respectively;

[0038] Let the area darker than the corresponding color mean on the three channels be the absolute shadow area, and find the absolute color mean of the absolute shadow area;

[0039] Based on the obtained absolute color mean, the two channels with the largest difference are determined;

[0040] The corresponding pixels in the two channels with the largest difference Subtract the color value of the pixel Features ;

[0041] Combined Area The pixels of are all regarded as shadow range, where for The average value of .

[0042] Optionally, extracting shadow features of multiple medical images in chronological order is achieved using CNN.

[0043] Optionally, the LSTM model is used to output the predicted shadow range to complete auxiliary analysis including:

[0044] When the specifications of the shadow range predicted by the LSTM model output change compared to the previous specifications, the warning level is determined based on the size of the changed pixel range and the warning level is output.

[0045] An embodiment of the present application also proposes an artificial intelligence-based medical image assisted analysis system, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the artificial intelligence-based medical image assisted analysis method as described above are implemented.

[0046] The medical image assisted analysis method of the present application combines image processing and artificial intelligence technology to realize the analysis of local shadow areas of medical images to improve the diagnostic efficiency of medical images.

[0047] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0049] Figure 1 The following is a schematic diagram of the basic process of the medical image assisted analysis method of this embodiment. DETAILED DESCRIPTION

[0050] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0051] The present application embodiment proposes a medical image assisted analysis method based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0052] In step S101, a specified number of medical images of the target part are acquired in a time sequence of the medical images. In some embodiments, for example, in the process of tracking the patient's condition, the medical images taken of the patient are recorded in a time sequence of the shooting, and a certain number of medical images are acquired according to the recording time. In some examples, the time sequence can be a plurality of different shooting times or a continuous shooting time. For the continuous shooting time, a plurality of medical images are captured according to the medical images.

[0053] In step S102, a reference medical image is selected from the previous medical images, and a unified coordinate system is established based on the reference medical image. The embodiment of the present application tracks the patient's condition at different shooting times, and by selecting a reference medical image and establishing a unified coordinate system, it is convenient to track and warn the patient's target department in the later stage.

[0054] In step S103, a reference boundary is selected and aligned for a specified number of medical images in the unified coordinate system. Aligning in the same coordinate system can facilitate subsequent segmentation with unified specifications.

[0055] In step S104, the aligned medical image is segmented according to uniform specifications. Specifically, the segmentation specifications can be set according to the size of the target part. The uniform segmentation after alignment can improve the efficiency of segmentation. The same segmentation includes multiple sub-image areas arranged in time sequence, and the shadow area can be quickly determined by processing and judging the sub-areas of the same segmentation in subsequent examples.

[0056] In step S105, for the segmentation at the same position, the similarity between the two sub-segments is calculated in chronological order. In some embodiments, the calculation can be completed by MSE cosine similarity or the like.

[0057] In step S106, the possible lesion area is determined according to the trend of the calculated similarity. In some examples, this example does not adopt the method of performing similarity calculation with the initial reference image. By calculating the similarity successively, the possible similarity change trend for the development of the lesion can be intuitively reflected. For example, the calculated values ​​of adjacent similarities gradually decrease, which should be paid special attention to in subsequent calculations.

[0058] In step S107, the shadow range of the possible lesion area is determined, and the shadow features of multiple medical images are extracted in chronological order. In the embodiment of the present application, the image features of the shadow area are focused on, and the LSTM model is further used for identification by extracting the features.

[0059] In step S108, a similarity tag is added to the shadow feature and input into a pre-trained LSTM model so as to utilize the LSTM model to output a predicted shadow range to complete auxiliary analysis.

[0060] The medical image assisted analysis method of the present application combines image processing and artificial intelligence technology to realize the analysis of local shadow areas of medical images to improve the diagnostic efficiency of medical images.

[0061] In some embodiments, selecting a reference medical image from previous medical images and establishing a unified coordinate system based on the reference medical image includes:

[0062] The first boundary of the target part region is identified from the previous medical image using an edge detection algorithm. In a specific example, the approximate boundary of the target part is identified from the previous medical image using an edge detection algorithm.

[0063] The clarity within the first boundary of the identified target part area is calculated, and by calculating the clarity, a reference image that meets the requirements can be selected.

[0064] A previous medical image with the highest definition within the first boundary is selected as the reference medical image.

[0065] Based on the first boundary of the target region, reference points are selected to establish a unified coordinate system. In some examples, the unified coordinate system can be established based on the line relationship of the first boundary of the target region, such as selecting the intersection of the lines as reference points.

[0066] In some embodiments, selecting reference boundaries and aligning a specified number of medical images in the unified coordinate system includes:

[0067] Using edge detection algorithm to identify the second boundary of the target area in any medical image;

[0068] Based on the first boundary and the second boundary, an association layer between the reference medical image and any medical image is established, the first boundary and the second boundary are divided into a plurality of boundary segments, and the matching degree between each boundary segment is calculated. For example, in some examples, a transparent layer can be set and the first boundary and the second boundary can be mapped on the transparent layer, and an association between the transparent layer and the corresponding reference image and any subsequent medical image is further established, so that the first boundary and the second boundary are respectively divided into a plurality of boundary segments on the transparent layer, and finally the matching degree of the boundary is calculated.

[0069] According to the calculated matching degree, the boundary segments of the first boundary and the second boundary with the highest matching degree are used as reference boundaries and translated and / or rotated so that the overlap degree of the reference boundary segments is the highest, that is, the boundary segments with the highest overlap degree are determined segment by segment.

[0070] According to the associated layer, any medical image is translated and / or rotated according to the translation distance and / or rotation angle of the reference boundary segment to complete the alignment. That is, during the translation and / or rotation process of the reference boundary, the medical image is translated and / or rotated according to the associated layer to complete the alignment.

[0071] In some embodiments, segmenting the aligned medical images according to a uniform specification includes:

[0072] The pixel area of ​​the target part in the reference medical image is determined. In a specific example, the pixel area within the identified approximate boundary is counted, and the segmentation specification is further determined based on the pixel area.

[0073] The aligned medical image is segmented according to the determined segmentation specification, wherein the determined segmentation specification enables the target part to contain a plurality of sub-segments with a specified pixel area after segmentation.

[0074] In some embodiments, determining the possible lesion area according to the calculated similarity trend includes:

[0075] For the segmentation at the same position, the calculated similarities of two adjacent segmentations are sorted according to the time sequence, that is, the similarities calculated two by two are arranged according to the time sequence, and the sorting order is not changed according to the calculated size of the similarity.

[0076] If the deviation of the similarity between two adjacent segments exceeds the set deviation range, or the deviation between the calculated similarity between any two segmentations at a later time and the calculated similarity between the two initial segmentations exceeds the preset threshold, it is determined to be a possible lesion segmentation. For example, in a specific example, if the similarity of multiple sets of medical images taken previously is 99%, but the similarity of the images taken subsequently gradually decreases, the corresponding area is considered to be a possible lesion segmentation.

[0077] The possible lesion area is determined based on the combined area segmented from each possible lesion.

[0078] After determining the combined area of ​​possible lesion segmentation, further fine shadow area identification is performed. In some embodiments, determining the shadow range of the possible lesion area includes:

[0079] For each possible lesion segmentation combined area, the transformation map is obtained by transformation according to the following formula:

[0080] ;

[0081] in, is the transformed pixel value, , , They are the pixel values ​​of the red, green and blue channels of any pixel in the combined area.

[0082] A secondary segmentation is performed in the transformation graph, for example, a secondary fine segmentation can be performed using methods such as Unet.

[0083] The color mean of each secondary segmented sub-region is calculated on three channels respectively, that is, the channel color mean of the sub-region is calculated on the red, green and blue channels respectively.

[0084] The area darker than the corresponding color mean in all three channels is defined as the absolute shadow area, and the absolute color mean of the absolute shadow area is calculated. In order to determine a clear shadow area, the present application first selects an area lower than the corresponding color mean in all three channels, which must be the absolute shadow area. In subsequent examples, the shadow range is further determined based on the absolute shadow area.

[0085] Based on the absolute color means obtained, the two channels with the largest difference are determined.

[0086] The corresponding pixels in the two channels with the largest difference Subtract the color value of the pixel Features , thus further highlighting the difference between shadow and non-shadow areas.

[0087] Combined Area The pixels of are all regarded as shadow range, where for The average value of the shadow area is finally combined with the absolute shadow area and the new pixels found to determine the shadow range. In this way, the shadow of the suspicious lesion area of ​​the target part in the medical image can be accurately determined, thereby providing accurate image input for subsequent prediction generation and improving the prediction effect of LSTM.

[0088] In some embodiments, extracting shadow features of multiple medical images in chronological order is achieved using CNN.

[0089] In some embodiments, the LSTM model is used to output a predicted shadow range to complete auxiliary analysis, including: when the specifications of the shadow range predicted by the LSTM model output change compared to the previous specifications, the alarm level is judged according to the size of the changed pixel range, and the alarm level is output. In some examples, multiple levels can be set in advance according to the size of the changed pixel range. After the LSTM model outputs the predicted shadow range, it is compared with the shadow range at the most recent moment to determine the alarm level.

[0090] The method of the present application analyzes multiple medical images of the patient and determines the precise shadow area for specification judgment, thereby achieving continuous monitoring of the target area.

[0091] An embodiment of the present application also proposes an artificial intelligence-based medical image assisted analysis system, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the artificial intelligence-based medical image assisted analysis method as described above are implemented.

[0092] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present disclosure with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. It is not limited to the examples described in this specification or during the implementation of this application, and its examples will be interpreted as non-exclusive.

[0093] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art may use other embodiments when reading the above description.

[0094] The above embodiments are merely exemplary embodiments of the present disclosure. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present disclosure, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A medical image assisted analysis method based on artificial intelligence, characterized in that: include: Acquire a specified number of medical images of the target part according to the time sequence of the medical images; Selecting a reference medical image from previous medical images, and establishing a unified coordinate system based on the reference medical image; Select reference boundaries and align a specified number of medical images in the unified coordinate system; Segment the aligned medical images according to uniform specifications; For the segmentation at the same position, the similarity between the two sub-segments is calculated in chronological order; According to the trend of the calculated similarity, the possible lesion area is determined; Determine the shadow range of the possible lesion area, and extract shadow features of multiple medical images in chronological order; Add similarity tags to the shadow features and input them into a pre-trained LSTM model to output a predicted shadow range using the LSTM model to complete auxiliary analysis; Selecting a reference medical image from prior medical images and establishing a unified coordinate system based on the reference medical image includes: Using an edge detection algorithm to identify a first boundary of a target area from a prior medical image; Calculating the clarity within a first boundary of the identified target region; Selecting a previous medical image with the highest definition within the first boundary as a reference medical image; and, based on the first boundary of the target area, selecting reference points to establish a unified coordinate system; Selecting reference boundaries and aligning a specified number of medical images in the unified coordinate system includes: Using edge detection algorithm to identify the second boundary of the target area in any medical image; Establishing an association layer of a reference medical image and any medical image based on the first boundary and the second boundary, dividing the first boundary and the second boundary into a plurality of boundary segments, and calculating a matching degree between each boundary segment; According to the calculated matching degree, the boundary segments of the first boundary and the second boundary with the highest matching degree are used as reference boundaries, and are translated and / or rotated so that the overlap degree of the reference boundary segments is the highest; And, synchronously, according to the associated layer, translating and / or rotating any medical image according to the translation distance and / or rotation angle of the reference boundary segment to complete the alignment; Segmentation of aligned medical images according to uniform specifications includes: Determining a pixel area of ​​the target site in the reference medical image; Determining a segmentation specification according to the pixel area; Segmenting the aligned medical image according to the determined segmentation specification, wherein the determined segmentation specification enables the target part to contain a plurality of sub-segments of a specified pixel area after segmentation; Based on the trend of the calculated similarity, possible lesion areas are determined to include: For the segmentation at the same position, the similarities of the two adjacent segmentations calculated are sorted according to the time order; If the deviation of the similarity between two adjacent segmentations exceeds the set deviation range, or the deviation between the calculated similarity between any two segmentations at a later time and the calculated similarity between the initial two segmentations exceeds a preset threshold, it is determined to be a possible lesion segmentation; Determine the possible lesion area according to the combined area segmented from each possible lesion; Determining the shadow range of the possible lesion area includes: For each possible lesion segmentation combined area, the transformation map is obtained by transformation according to the following formula: ； in, is the transformed pixel value, , , are the pixel values ​​of the red, green and blue channels of any pixel in the combined area respectively; Perform secondary segmentation in the transformation graph; Calculate the color mean of each secondary segmented sub-region on three channels respectively; Let the area darker than the corresponding color mean on the three channels be the absolute shadow area, and find the absolute color mean of the absolute shadow area; Based on the obtained absolute color mean, the two channels with the largest difference are determined; The corresponding pixels in the two channels with the largest difference Subtract the color value of the pixel Features ; Combined Area The pixels of are all regarded as shadow range, where for The average value of .

2. The artificial intelligence-based medical image assisted analysis method according to claim 1, characterized in that: Extracting shadow features from multiple medical images in chronological order is achieved using CNN.

3. The artificial intelligence-based medical image assisted analysis method according to claim 1, characterized in that: The LSTM model is used to output the predicted shadow range to complete auxiliary analysis including: When the specifications of the shadow range predicted by the LSTM model output change compared to the previous specifications, the warning level is determined based on the size of the changed pixel range and the warning level is output.

4. A medical image assisted analysis system based on artificial intelligence, characterized in that: It comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the medical image assisted analysis method based on artificial intelligence as described in any one of claims 1 to 3 are implemented.

Images