Method for generating image detection model, image detection method, and computer device

Abstract

The present application relates to a method for generating an image detection model, an image detection method, and a computer device. The method includes: obtaining eye movement data of a first medical image, and determining an attention training heat map of the first medical image based on the eye movement data; training a first initial model through multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model; inputting a second medical image into the attention prediction model to obtain an attention heat map, and amplifying the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image; training a second initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and the labeled image of each second medical image to obtain an image detection model. By using this method, multiple amplified images including the attention area can be obtained, improving the accuracy of the image detection model.

Description

Technical Field

[0001] The present application relates to the technical field of medical image processing, and particularly to a method for generating an image detection model, an image detection method, and a computer device. Background Art

[0002] With the rapid development of computer technology, neural networks and medical imaging technology have greatly improved the modern medical level.

[0003] Neural networks require a large number of labeled medical images for full-supervised training to achieve better results. However, the number of professional radiologists is small, and the number of medical images labeled by professional radiologists is even smaller, resulting in low accuracy of the trained neural networks.

[0004] To solve the problem of the lack of labeled medical image data, it is necessary to expand the labeled medical images. Usually, image transformation, data augmentation, etc. are used to expand the labeled images. However, the above expansion methods are more suitable for training neural networks for natural images; compared with natural images, medical images have fewer pixels, and the regions of interest in medical images only occupy a very small part of the medical images. Using image transformation, data augmentation, etc. to expand medical images is likely to cause the loss of regions of interest. Therefore, expanding the medical image dataset by the above expansion methods will result in low accuracy of the neural network trained based on this dataset. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a method for generating an image detection model, an image detection method, and a computer device that can obtain multiple amplified images including regions of attention and improve the accuracy of the image detection model.

[0006] In a first aspect, the present application provides a method for generating an image detection model. The method for generating an image detection model includes:

[0007] Obtain the eye movement data of a first medical image, and determine the attention training heat map of the first medical image based on the eye movement data, where the eye movement data is collected when a person views the first medical image;

[0008] Train a first initial model with multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model;

[0009] Input a second medical image into the attention prediction model to obtain an attention heat map, and amplify the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image;

[0010] The second initial model is trained using multiple second medical images, multiple amplified images corresponding to each second medical image, and the annotation images of each second medical image to obtain an image detection model.

[0011] In one embodiment, the eye movement data includes the sampling times and sampling positions of multiple sampling points; determining the attention training heat map of the first medical image based on the eye movement data includes:

[0012] According to the sampling times of the multiple sampling points, the fixation points and saccade points are determined among the multiple sampling points, where the fixation points include a fixation duration greater than a preset duration, the saccade points include multiple discontinuous fixation durations, and the fixation duration includes multiple consecutive sampling times;

[0013] Based on the sampling positions of the fixation points and the saccade points, the fixation points and the saccade points are marked on the mask image to obtain an image to be processed, where the size of the mask image is the same as that of the first medical image;

[0014] The image to be processed is subjected to Gaussian smoothing processing to obtain the attention training heat map of the first medical image.

[0015] In one embodiment, amplifying the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image includes:

[0016] Determine the segmentation mask of the attention heat map, and determine the eye movement amplified image based on the segmentation mask and the second medical image, where the eye movement amplified image includes the attention area;

[0017] The second medical image is amplified in multiple preset ways to obtain multiple initial images, and based on the attention heat map and the multiple initial images, multiple target images are determined, where each target image includes the attention area, and the multiple amplified images include the eye movement amplified image and the multiple target images.

[0018] In one embodiment, the multiple ways include a random masking method and a random cropping method; amplifying the second medical image in multiple preset ways to obtain multiple initial images, and based on the attention heat map and the multiple initial images, determining multiple target images includes:

[0019] The second medical image is processed using the random masking method to obtain an initial masked image;

[0020] Compare the initial masked image with the attention heat map. If the initial masked image includes the attention area, use the initial masked image as the target image;

[0021] Process the second medical image using a random cropping method to obtain an initial cropped image;

[0022] Compare the initial cropped image with the attention heat map. If the initial cropped image includes the attention area, use the initial cropped image as the target image.

[0023] In one embodiment, after processing the second medical image using a random masking method to obtain an initial masked image, the method further includes:

[0024] Compare the initial masked image with the attention heat map. If the initial masked image does not include the attention area, remove the initial masked image and repeat the process of obtaining the initial masked image until the obtained initial masked image includes the attention area.

[0025] In one embodiment, after processing the second medical image using a random cropping method to obtain an initial cropped image, the method further includes:

[0026] Compare the initial cropped image with the attention heat map. If the initial cropped image does not include the attention area, remove the initial cropped image and repeat the process of obtaining the initial cropped image until the obtained initial cropped image includes the attention area.

[0027] In one embodiment, the method further includes:

[0028] Perform rotation processing and color perturbation processing on the second medical image to obtain a rotated image and a color-perturbed image, and use the rotated image and the color-perturbed image as augmented images of the second medical image.

[0029] In a second aspect, the present application further provides a device for generating an image detection model. The device includes:

[0030] An eye movement data processing module, configured to obtain eye movement data of a first medical image and determine an attention training heat map of the first medical image based on the eye movement data, where the eye movement data is collected when a person views the medical image;

[0031] An attention prediction model training model, configured to train a first initial model through multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model;

[0032] An amplification module, configured to input a second medical image into the attention prediction model to obtain an attention heat map, and amplify the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image;

[0033] An image detection model training module, configured to train a second initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and an annotated image of each second medical image to obtain an image detection model.

[0034] In a third aspect, the present application further provides an image detection method. The image detection method includes:

[0035] Obtain a medical image to be processed;

[0036] Input the medical image to be processed into an image detection model to determine a region of interest;

[0037] The image detection model is obtained by training an initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and an annotated image of each second medical image;

[0038] The multiple amplified images corresponding to the second medical image are obtained by amplifying the second medical image based on an attention heat map;

[0039] The attention heat map is related to simulated eye movement data.

[0040] In a fourth aspect, the present application further provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0041] Obtain a medical image to be processed;

[0042] Input the medical image to be processed into an image detection model to determine a region of interest;

[0043] The image detection model is obtained by training an initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and an annotated image of each second medical image;

[0044] The multiple amplified images corresponding to the second medical image are obtained by amplifying the second medical image based on an attention heat map;

[0045] The attention heat map is obtained through real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0046] Fifth aspect, the present application also provides a computer-readable storage medium. On the computer-readable storage medium, there is a computer program stored, and when the computer program is executed by a processor, the following steps are implemented:

[0047] Obtain the medical image to be processed;

[0048] Input the medical image to be processed into an image detection model to determine the region of interest;

[0049] The image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image;

[0050] The multiple augmented images corresponding to the second medical image are obtained by augmenting the second medical image based on the attention heat map;

[0051] The attention heat map is obtained by real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0052] Sixth aspect, the present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0053] Obtain the medical image to be processed;

[0054] Input the medical image to be processed into an image detection model to determine the region of interest;

[0055] The image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image;

[0056] The multiple augmented images corresponding to the second medical image are obtained by augmenting the second medical image based on the attention heat map;

[0057] The attention heat map is obtained by real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0058] The method for generating the above image detection model, the apparatus for generating the image detection model, the image detection method, the computer device, the storage medium, and the computer program product determine the attention training heat map of the first medical image according to the eye movement data of the first medical image, train the first initial model according to multiple first medical images and the attention training heat map of each first medical image to obtain the attention prediction model, predict the attention heat map of the second medical image through the attention prediction model, the attention heat map of the second medical image includes the attention area of the second medical image, and amplify the second medical image based on the attention heat map so that the amplified images of the second medical image all include the attention area, and train the second initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and the labeled image of each second medical image to obtain the image detection model. Since multiple second medical images and multiple amplified images corresponding to each second medical image all include the attention area, and the attention area can reflect the area of interest in the second medical image, the second initial model can learn the information of the attention area, thereby improving the accuracy of the trained image detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Schematic flowchart of the method for generating the image detection model in one embodiment;

[0060] Figure 2 Schematic diagram of collecting eye movement data in one embodiment;

[0061] Figure 3 Schematic diagram of determining the attention training heat map of the first medical image in one embodiment;

[0062] Figure 4 Schematic diagram of the second medical image in one embodiment;

[0063] Figure 5 For Figure 4 attention heat map;

[0064] Figure 6 For Figure 4 eye movement amplified image;

[0065] Figure 7 For Figure 4 initial occlusion image;

[0066] Figure 8 For Figure 4 initial cropped image;

[0067] Figure 9 For Figure 4 initial occlusion image including the attention area;

[0068] Figure 10 For Figure 4 the initial cropped image including the attention area;

[0069] Figure 11 is a schematic flowchart for determining multiple amplified images of a second medical image in one embodiment;

[0070] Figure 12 is a schematic flowchart for determining multiple amplified images of a second medical image in another embodiment;

[0071] Figure 13 is a schematic flowchart for a method of generating an image detection model in a specific embodiment;

[0072] Figure 14 is a schematic flowchart for an image detection method in one embodiment;

[0073] Figure 15 is a structural block diagram of an image detection model generation device in one embodiment;

[0074] Figure 16 is an internal structure diagram of a computer device in one embodiment. Detailed implementation manners

[0075] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0076] In one embodiment, as Figure 1 shown, a method for generating an image detection model is provided. In this embodiment, the method is illustrated by taking its application to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0077] S101, obtain the eye movement data of the first medical image, and determine the attention training heat map of the first medical image based on the eye movement data.

[0078] Among them, the eye movement data is collected when a person views the first medical image. The attention training heat map of the first medical image includes the attention area of the first medical image, and the attention area is the area that the person pays attention to when viewing the first medical image.

[0079] Specifically, the first medical image is obtained by a medical imaging device. The first medical image can be a single-modal medical image such as a Computed Tomography (CT) image, a Magnetic Resonance Imaging (MRI) image, a Direct Digit Radiography (DDR) image, an ultrasonic (US) image, a Positron Emission Computed Tomography (PET) image, a single photon emission computed tomography (SPECT) image, or a Digital subtraction angiography (DSA) image. It can also be a hybrid-modal medical image such as MR-PET, CT-PET, or MR-US.

[0080] The eye movement data generated when a person views the first medical image is collected by an eye tracker. The person can be a radiologist. The eye tracker sends the collected eye movement data of the first medical image to a terminal, enabling the terminal to obtain the eye movement data of the first medical image. The eye movement data includes the sampling times and sampling positions of multiple sampling points. Based on the sampling times and sampling positions of the multiple sampling points, the area that the person is interested in is determined in the first medical image, and then the attention training heat map of the first medical image is determined. The area that the person is interested in is the attention area of the first medical image. Usually, the area that the person is interested in is the area where the region of interest is located in the first medical image. Therefore, the attention area of the first medical image can reflect the region of interest in the first medical image, and the region of interest is usually the area where the lesion is located.

[0081] Exemplarily, as Figure 2 shown, the first medical image is displayed on a display screen, and an eye tracker is set below the display screen so that the eye tracker can collect the eye movement data generated when a person views the first medical image. A 5-point calibration procedure is used to calibrate the eye tracking of the person. The person views a first medical image displayed on the display screen until the region of interest of the first medical image can be determined. Then, the next first medical image is replaced, and the eye movement data of this first medical image is collected by the eye tracker. The eye tracker sends the eye movement data of this first medical image to the terminal, enabling the terminal to obtain the eye movement data of the first medical image. In the case where the person needs to view multiple first medical images, it is set to prompt the person to rest after viewing a preset number of first medical images to reduce the fatigue of the person, thereby improving the attention level and enabling the collected eye movement data to accurately reflect the attention area in the first medical image.

[0082] S102. Train a first initial model using multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model.

[0083] Specifically, obtain the eye movement data of multiple first medical images, and determine the attention training heat map of each first medical image according to the eye movement data of each first medical image.

[0084] Input the first medical image into the first initial model, obtain the attention prediction image of the first medical image through the first initial model, calculate the first loss value according to the attention prediction image and the attention training heat map of the first medical image, adjust the model parameters of the first initial model according to the first loss value to complete one training of the first initial model, and perform multiple iterative trainings on the first initial model using multiple first medical images and the attention training heat map of each first medical image until the model parameters of the first initial model converge, and use the first initial model with converged model parameters as the attention prediction model.

[0085] The attention prediction model trained through the above process is equivalent to a virtual radiologist and can predict the attention area of medical images.

[0086] S103. Input the second medical image into the attention prediction model to obtain an attention heat map, and amplify the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image.

[0087] Among them, the attention heat map includes the attention area in the second medical image, and each amplified image includes the attention area in the second medical image.

[0088] Specifically, the second medical image is also obtained by a medical imaging device. The second medical image can be a single-modal medical image such as a Computed Tomography (CT) image, a Magnetic Resonance Imaging (MRI) image, a Direct Digit Radiography (DDR) image, an ultrasonic (US) image, a Positron Emission Computed Tomography (PET) image, a single photon emission computed tomography (SPECT) image, or a Digital subtraction angiography (DSA) image. It can also be a mixed-modal medical image such as MR-PET, CT-PET, or MR-US. The second medical image can be the same as the first medical image.

[0089] Input the second medical image into the attention prediction model to obtain the attention heat map of the second medical image; the attention heat map of the second medical image only includes the attention area of the second medical image.

[0090] Perform augmentation processing on the second medical image in a variety of preset ways to obtain multiple initial images, and determine multiple augmented images based on the attention heat map of the second medical image and the multiple initial images, so that each augmented image includes the attention area.

[0091] S104. Train the second initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image to obtain an image detection model.

[0092] Among them, the annotation image of the second medical image is obtained by a doctor using the gold standard to annotate the region of interest in the second medical image.

[0093] The terminal obtains multiple second medical images, multiple amplified images corresponding to each second medical image, and the annotation image of each second medical image; uses the multiple second medical images and the multiple amplified images corresponding to each second medical image as multiple training images, inputs the training images into the second initial model, obtains the predicted image of the region of interest of the training image through the second initial model, calculates the second loss value according to the annotation image and the predicted image of the region of interest of the training image, and adjusts the model parameters of the second initial model according to the second loss value to complete one training of the second initial model. The second initial model is iteratively trained multiple times through the multiple training images and the annotation image of each training image until the model parameters of the second initial model converge, and the second initial model with converged model parameters is used as the image detection model.

[0094] The image detection model obtained through the above training process can predict the region of interest in the medical image.

[0095] In the method for generating the above image detection model, the attention training heat map of the first medical image is determined according to the eye movement data of the first medical image, the first initial model is trained according to the multiple first medical images and the attention training heat map of each first medical image to obtain the attention prediction model, the attention heat map of the second medical image is predicted through the attention prediction model, the attention heat map of the second medical image includes the attention region of the second medical image, and the second medical image is amplified based on the attention heat map so that the amplified images of the second medical image all include the attention region. The second initial model is trained through the multiple second medical images, the multiple amplified images corresponding to each second medical image, and the annotation image of each second medical image to obtain the image detection model. Since the multiple second medical images and the multiple amplified images corresponding to each second medical image all include the attention region, and the attention region can reflect the region of interest in the second medical image, the second initial model can learn the information of the attention region, thereby improving the accuracy of the trained image detection model.

[0096] In one embodiment, the eye movement data includes the sampling time and sampling position of multiple sampling points; the sampling point is the landing point of the line of sight, each sampling point has its corresponding sampling time and sampling position, the sampling time is the time when the person's line of sight lands on the sampling point, and the sampling position is the coordinate of the sampling point in the first medical image. Determining the attention training heat map of the first medical image based on the eye movement data includes the following process:

[0097] S111, determine the fixation point and saccade point among the multiple sampling points according to the sampling time of the multiple sampling points.

[0098] Among them, the fixation point includes a fixation duration greater than a preset duration, the review point includes multiple discontinuous fixation durations, and the fixation duration includes multiple consecutive sampling moments.

[0099] Specifically, during the process of a person viewing the first medical image, eye movement data is collected, and multiple sampling points of the eye movement data are determined. For any sampling point, the multiple sampling moments of this sampling point are counted to determine the fixation duration of this sampling point. If the fixation duration of this sampling point is greater than the preset duration, then this sampling point is a fixation point. If there are multiple discontinuous fixation durations at this sampling point, then this sampling point is a review point.

[0100] For any sampling point, if there are multiple consecutive sampling moments among the multiple sampling moments of this sampling point, and the number of consecutive sampling moments exceeds the preset fixation number, then it is determined that the multiple consecutive sampling moments constitute the fixation duration of this sampling point. If the fixation duration of this sampling point (the duration corresponding to the multiple consecutive sampling moments) is greater than the preset duration, then this sampling point is a fixation point. If this sampling point includes multiple discontinuous fixation durations, then this sampling point is a review point.

[0101] For example, the sampling moments of sampling point f1 include: t1, t2, t3, t4, t5, t6, t7, t11, t21, t30, t31, t32, t33, t34; assuming that the fixation duration includes 4 consecutive sampling moments, and the preset duration is the duration corresponding to 6 consecutive sampling moments; the consecutive t1, t2, t3, t4, t5, t6, t7 constitute the first fixation duration of f1, and the consecutive t30, t31, t32, t33, t34 constitute the second fixation duration of f2; the first fixation duration of f1 includes 7 consecutive sampling moments, and the first fixation duration is greater than the preset duration, so f1 is a fixation point; the first fixation duration and the second fixation duration of f1 are discontinuous, so f1 is a review point.

[0102] For example, the sampling moments of sampling point f2 include: t1, t2, t3, t21, t22, t23, t24, t25, t26, t27; assuming that the fixation duration includes 4 consecutive sampling moments, and the preset duration is the duration corresponding to 6 consecutive sampling moments; the consecutive t21, t22, t23, t24, t25, t26, t27 constitute the fixation duration of f2. Since the fixation duration of f2 is greater than the preset duration, f2 is a fixation point; f2 does not include multiple discontinuous fixation durations, so f2 is not a review point.

[0103] S112, based on the sampling positions of the fixation points and the review points, mark the fixation points and the review points on the mask image to obtain an image to be processed.

[0104] Among them, the mask image has the same size as the first medical image.

[0105] Specifically, a mask image having the same size as the first medical image is obtained, and a fixation point and a saccade point are marked on the mask image to obtain an image to be processed. The fixation point is a sampling point where the person has a relatively long fixation time, and the saccade point is a sampling point where the person fixates multiple times. Therefore, the fixation point and the saccade point are sampling points that the person pays attention to. The image to be processed only includes the fixation point and the saccade point in the first medical image. By this step, the sampling points that are not concerned by the person are removed.

[0106] S113. Perform Gaussian smoothing on the image to be processed to obtain an attention training heat map of the first medical image.

[0107] Specifically, the image to be processed is used to reflect the fixation point and the saccade point that the person pays attention to. In fact, when the person views the first medical image, what the person should pay attention to are some regions of the first medical image. Therefore, Gaussian smoothing is performed on the image to be processed to obtain an attention training heat map of the first medical image. The attention training heat map includes the attention regions in the first medical image.

[0108] Exemplarily, as Figure 3 shown, the first medical image P1 is obtained, and eye movement data generated during the process of the person viewing the first medical image is collected. The eye movement data includes multiple sampling points. Multiple sampling points are marked on the first medical image to obtain image P2. Fixation points and saccade points are determined among the multiple sampling points. Sampling points other than the fixation points and non-saccade points among the multiple sampling points in P2 are filtered out to obtain image P3. The fixation points and saccade points in P3 are marked on a mask image having the same size as the first medical image to obtain an image to be processed. Gaussian smoothing is performed on the image to be processed to obtain an attention training heat map P4 of the first medical image.

[0109] In one embodiment, amplifying the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image includes the following process:

[0110] S311. Determine a segmentation mask of the attention heat map, and determine an eye movement amplified image based on the segmentation mask and the second medical image.

[0111] Wherein, the eye movement amplified image includes the attention region.

[0112] Specifically, the attention heat map only includes the attention area in the second medical image. In the segmentation mask of the attention heat map, the pixel value of the pixel points within the attention area is 1, and the pixel value of the pixel points not within the attention area is 0. Multiply the segmentation mask and the second medical image to obtain the eye movement augmented image. The multiplication of the segmentation mask and the second medical image is pixel-by-pixel multiplication. Since the pixel value of the pixel points within the attention area in the segmentation mask is 1 and the pixel value of the pixel points not within the attention area is 0, therefore, the eye movement augmented image only includes the attention area of the second medical image. For example, the second medical image is as Figure 4 shown, the attention heat map of the second medical image is as Figure 5 shown, and the eye movement augmented image of the second medical image is as Figure 6 shown.

[0113] S312. Perform augmentation processing on the second medical image in multiple preset ways to obtain multiple initial images, and determine multiple target images based on the attention heat map and the multiple initial images.

[0114] Wherein, each target image includes the attention area, and the multiple augmented images include the eye movement augmented image and the multiple target images.

[0115] In one embodiment, the multiple preset ways include: random masking method and random cropping method. The process of obtaining multiple target images includes:

[0116] Process the second medical image using the random masking method to obtain an initial masked image; compare the initial masked image with the attention heat map. If the initial masked image includes the attention area, then use the initial masked image as the target image; process the second medical image using the random cropping method to obtain an initial cropped image; compare the initial cropped image with the attention heat map. If the initial cropped image includes the attention area, then use the initial cropped image as the target image.

[0117] In one implementation, processing the second medical image using the random masking method to obtain an initial masked image includes: randomly selecting a pixel point in the second medical image, taking this pixel point as the center point, determining a masked area in the second medical image, and setting the pixel values of all pixel points within the masked area to 0 to obtain the initial masked image; the shape and size of the masked area are preset. For example, the masked area is a rectangular area, and the size of the masked area is 40*40. For example, the second medical image is as Figure 4 shown, and the initial masked image is as Figure 7 shown.

[0118] In one implementation, the second medical image is processed by means of random cropping to obtain an initial cropped image, which includes: randomly selecting a ratio from a preset set of ratios, randomly cropping a partial image from the second medical image according to this ratio, and enlarging the cropped partial image to obtain an initial cropped image with the same area as the second medical image; the ratio of the area of the partial image to the area of the second medical image is this ratio. For example, the second medical image is as Figure 4 shown, and the initial cropped image is as Figure 8 shown.

[0119] When the second medical image is processed by means of random masking, it may mask the attention area in the second medical image, resulting in the initial masked image not including the attention area. Therefore, the initial masked image is compared with the attention heat map. That the initial masked image includes the attention area means that the initial masked image includes the complete attention area. If the initial masked image includes a partial attention area, then the initial masked image does not include the attention area; if the initial masked image includes the attention area, then this initial masked image can be used as the target image for training the first initial model.

[0120] When the second medical image is processed by means of random cropping, it may crop and remove the attention area in the second medical image, resulting in the initial cropped image not including the attention area. Therefore, the initial cropped image is compared with the attention heat map. Similarly, it means that the initial cropped image includes the complete attention area. If the initial cropped image includes the attention area, then this initial cropped image can be used as the target image for training the first initial model.

[0121] In one embodiment, in order to ensure that the second medical image has its corresponding initial masked image and initial cropped image including the attention area, the process of obtaining multiple target images further includes:

[0122] Compare the initial masked image with the attention heat map. If the initial masked image does not include the attention area, then reject the initial masked image and repeat the above process of obtaining the initial masked image until the obtained initial masked image includes the attention area. For example, the second medical image is as Figure 4 shown, and the initial masked image including the attention area is as Figure 9 shown.

[0123] Compare the initial cropped image with the attention heat map. If the initial cropped image does not include the attention area, then reject the initial cropped image and repeat the above process of obtaining the initial cropped image until the obtained initial cropped image includes the attention area. For example, the second medical image is as Figure 4 shown, and the initial cropped image including the attention area is as Figure 10as shown

[0124] Specifically, the cases where the initial masked image does not include the attention area include: the case where the initial masked image includes a partial attention area, and the case where the initial masked image does not include the attention area at all. Similarly, the cases where the initial cropped image does not include the attention area include: the case where the initial cropped image includes a partial attention area, and the case where the initial cropped image does not include the attention area at all.

[0125] If the initial masked image does not include the attention area, the second medical image is processed again using the random masking method until the obtained initial masked image includes the attention area, and the initial masked image including the attention area is used as the target image.

[0126] If the initial cropped image does not include the attention area, the second medical image is processed again using the random cropping method until the obtained initial cropped image includes the attention area, and the initial cropped image including the attention area is used as the target image.

[0127] Through the above process, by processing the second medical image in a variety of preset ways, two target images including the attention area can be obtained. It can be imagined that the variety of ways is not limited to the random masking method and the random cropping method. Other image processing can also be performed on the second medical image to obtain a processed image. The processed image is compared with the attention heat map. If the processed image includes the attention area, the processed image is used as the target image. If the processed image does not include the attention area, the process of obtaining the processed image is repeated until a processed image including the attention area is obtained.

[0128] In one embodiment, in order to increase the number of augmented images of the second medical image, the method for generating the image detection model further includes:

[0129] The second medical image is subjected to rotation processing and color perturbation processing to obtain a rotated image and a color-perturbed image, and the rotated image and the color-perturbed image are used as the augmented images of the second medical image.

[0130] Specifically, performing rotation processing on the second medical image includes performing rotation processing on the second medical image in a random direction and at a random angle, performing horizontal flipping processing on the second medical image, and performing vertical flipping processing on the second medical image.

[0131] Since neither the rotation processing nor the color perturbation processing of the second medical image will cause the loss of the attention area, the rotated image and the color-perturbed image are directly used as the augmented images of the second medical image.

[0132] In one embodiment, as Figure 11As shown, the process of determining multiple amplified images of the second medical image includes:

[0133] A1. Obtain the attention heat map of the second medical image, where the attention heat map of the second medical image is related to the simulated eye movement data;

[0134] A2. Determine the segmentation mask of the attention heat map, and determine the eye movement amplified image based on the segmentation mask and the second medical image, where the eye movement amplified image includes the attention area;

[0135] A3. Process the second medical image in a random masking manner to obtain an initial masked image;

[0136] A4. Process the second medical image in a random cropping manner to obtain an initial cropped image;

[0137] A5. Compare the initial masked image with the attention heat map to determine whether the initial masked image includes the attention area. If not, go to A6; if so, go to A7;

[0138] A6. Reject the initial masked image and go to A3;

[0139] A7. Take the initial masked image as the target image;

[0140] A8. Compare the initial cropped image with the attention heat map to determine whether the initial cropped image includes the attention area. If not, go to A9; if so, go to A10;

[0141] A9. Reject the initial cropped image and go to A4;

[0142] A10. Take the initial cropped image as the target image.

[0143] The multiple amplified images of the second medical image include the eye movement amplified image and multiple target images.

[0144] Specifically, obtain the attention heat map of the second medical image. The attention heat map of the second medical image is related to the eye movement data of the second medical image, and the attention heat map of the second medical image includes the attention area of the second medical image. Amplify the second medical image according to the attention heat map of the second medical image to obtain multiple amplified images, so that each amplified image includes the attention area of the second medical image.

[0145] In one embodiment, as Figure 12 shown, the process of determining multiple amplified images of the second medical image includes:

[0146] B1. Obtain the attention heatmap of the second medical image, where the attention heatmap is obtained through real-time acquisition or simulation, and the attention heatmap is related to the eye movement data when a person views the second medical image;

[0147] B2. Determine the segmentation mask of the attention heatmap, and determine the eye movement augmented image based on the segmentation mask and the second medical image, where the eye movement augmented image includes the attention area;

[0148] B3. Process the second medical image by using a random masking method to obtain an initial masked image;

[0149] B4. Process the second medical image by using a random cropping method to obtain an initial cropped image;

[0150] B5. Compare the initial masked image with the attention heatmap to determine whether the initial masked image includes the attention area. If not, go to B6; if so, go to B7;

[0151] B6. Reject the initial masked image and go to B3;

[0152] B7. Take the initial masked image as the target image;

[0153] B8. Compare the initial cropped image with the attention heatmap to determine whether the initial cropped image includes the attention area. If not, go to B9; if so, go to B10;

[0154] B9. Reject the initial cropped image and go to B4;

[0155] B10. Take the initial cropped image as the target image.

[0156] Multiple augmented images of the second medical image include the eye movement augmented image and multiple target images.

[0157] Specifically, obtain the eye movement data generated when a person views the second medical image, or simulate a person viewing the second medical image, obtain the eye movement data through simulation, determine the attention heatmap of the second medical image according to the eye movement data, and the attention heatmap of the second medical image includes the attention area of the second medical image. Augment the second medical image according to the attention heatmap of the second medical image to obtain multiple augmented images, so that each augmented image includes the attention area of the second medical image.

[0158] In a specific embodiment, as Figure 13 shown, the method for generating an image detection model includes:

[0159] The first process:

[0160] The eye movement data of the first medical image Y1 is determined using an eye tracker, and the attention training heat map Z1 of the first medical image Y1 is determined based on the eye movement data of the first medical image Y1;

[0161] The first initial model is trained with multiple first medical images Y1 and the attention training heat map Z1 of each first medical image to obtain an attention prediction model;

[0162] The second process:

[0163] The second medical image Y2 is input into the attention prediction model to obtain the attention heat map Z2 of the second medical image Y2;

[0164] The eye movement amplified image K1 is determined based on the attention heat map Z2 and the second medical image Y2;

[0165] The second medical image Y2 is processed by means of random masking and random cropping to obtain an initial masked image C1 and an initial cropped image C2;

[0166] In this embodiment, if C1 does not include the attention area, the second medical image Y2 is re - processed by means of random masking to obtain the target image K2;

[0167] In this embodiment, if C2 does not include the attention area, the second medical image Y2 is re - processed by means of random cropping to obtain the target image K3; The amplified images of the second medical image include: the eye movement amplified image K1, the target image K2, and the target image K3;

[0168] The second initial model is trained with multiple second medical images, the annotation images of each second medical image, and multiple amplified images to obtain an image detection model.

[0169] In one embodiment, the image detection model is used to detect the region of interest in a medical image. The medical image to be detected is input into the image detection model, and the detection result is obtained through the image detection model. The detection result includes the position and type of the region of interest in the medical image to be detected, and the region of interest can be outlined in the medical image to be detected to display the region of interest in the medical image to be detected.

[0170] In this embodiment, an attention training heat map of a first medical image is determined according to the eye movement data of the first medical image, a first initial model is trained according to a plurality of first medical images and the attention training heat map of each first medical image to obtain an attention prediction model, an attention heat map of a second medical image is predicted by the attention prediction model, the attention heat map of the second medical image includes an attention area of the second medical image, the second medical image is amplified based on the attention heat map so that the amplified images of the second medical image all include the attention area, and a second initial model is trained by using a plurality of second medical images, a plurality of amplified images corresponding to each second medical image, and the annotation image of each second medical image to obtain an image detection model. Since the plurality of second medical images and the plurality of amplified images corresponding to each second medical image all include the attention area, and the attention area can reflect the area of interest in the second medical image, the second initial model can learn the information of the attention area, thereby improving the accuracy of the trained image detection model.

[0171] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0172] The medical image to be processed is input into an image detection model to determine the area of interest, wherein the image detection model is obtained by training an initial model with a plurality of second medical images, a plurality of amplified images corresponding to each second medical image, and the annotation image of each second medical image, and the plurality of amplified images corresponding to the second medical image are obtained by amplifying the second medical image based on the attention heat map, and the attention heat map is related to simulated eye movement data.

[0173] In one embodiment, as Figure 14 shown, an image detection method is provided, including:

[0174] M100. Obtain a medical image to be processed;

[0175] M200. Input the medical image to be processed into an image detection model to determine the area of interest.

[0176] Among them, the image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image. The multiple augmented images corresponding to the second medical image are obtained by augmenting the second medical image based on an attention heat map, and the attention heat map is related to simulated eye movement data.

[0177] Specifically, the medical image to be processed is obtained by a medical imaging device photographing a body part of a human body. The image detection model is used to detect the region of interest in the medical image to be processed, that is, the region where the lesion is located. The medical image to be processed is input into the image detection model, and a detection result is obtained through the image detection model. The detection result includes the position and type of the region of interest in the medical image to be processed, and the region of interest can be outlined in the medical image to be processed.

[0178] The attention heat map of the second medical image is related to the eye movement data of the second medical image. The eye movement data generated when a person views the second medical image is obtained, or a person is simulated to view the second medical image, and the eye movement data is obtained through a simulation method. The attention heat map of the second medical image is determined according to the eye movement data. The attention heat map of the second medical image includes the attention region of the second medical image. Multiple augmented images are obtained by augmenting the second medical image according to the attention heat map of the second medical image, so that each augmented image includes the attention region of the second medical image.

[0179] Based on the same inventive concept, an embodiment of the present application further provides an image detection model generation device for implementing the above-mentioned image detection model generation method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the following image detection model generation device can refer to the limitations on the image detection model generation method in the above text, and will not be repeated here.

[0180] In one embodiment, as Figure 15 shown, an image detection model generation device is provided, including:

[0181] An eye movement data processing module 100, configured to obtain the eye movement data of the first medical image and determine the attention training heat map of the first medical image based on the eye movement data, where the eye movement data is collected when a person views the medical image;

[0182] An attention prediction model training model 200, configured to train a first initial model through multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model;

[0183] The amplification module 300 is configured to input the second medical image into the attention prediction model to obtain an attention heat map, and amplify the second medical image based on the attention heat map to obtain multiple amplified images of the second medical image;

[0184] The image detection model training module 400 is configured to train a second initial model through multiple second medical images, multiple amplified images corresponding to each second medical image, and the annotated image of each second medical image to obtain an image detection model.

[0185] In one embodiment, the eye movement data includes the sampling times and sampling positions of multiple sampling points; the eye movement data processing module includes:

[0186] The sampling point processing unit is configured to determine a fixation point and a saccade point among the multiple sampling points according to the sampling times of the multiple sampling points, wherein the fixation point includes a fixation duration greater than a preset duration, the saccade point includes multiple discontinuous fixation durations, and the fixation duration includes multiple consecutive sampling times;

[0187] The image to be processed determination unit is configured to mark the fixation point and the saccade point on the mask image based on the sampling positions of the fixation point and the saccade point to obtain an image to be processed, wherein the mask image has the same size as the first medical image;

[0188] The Gaussian smoothing processing unit is configured to perform Gaussian smoothing processing on the image to be processed to obtain the attention training heat map of the first medical image.

[0189] In one embodiment, the amplification module includes:

[0190] The eye movement amplified image determination unit is configured to determine a segmentation mask of the attention heat map, and determine an eye movement amplified image based on the segmentation mask and the second medical image, wherein the eye movement amplified image includes the attention area;

[0191] The target image determination unit is configured to perform amplification processing on the second medical image in multiple preset manners to obtain multiple initial images, and determine multiple target images based on the attention heat map and the multiple initial images, wherein each target image includes the attention area, and the multiple amplified images include the eye movement amplified image and the multiple target images.

[0192] In one embodiment, the target image determination unit includes:

[0193] The random masking unit is configured to process the second medical image in a random masking manner to obtain an initial masked image;

[0194] A first comparison unit, configured to compare the initial masked image and the attention heat map, and if the initial masked image includes the attention area, use the initial masked image as the target image;

[0195] A random cropping unit, configured to process the second medical image by using a random cropping method to obtain an initial cropped image;

[0196] A second comparison unit, configured to compare the initial cropped image and the attention heat map, and if the initial cropped image includes the attention area, use the initial cropped image as the target image.

[0197] In one embodiment, the target image determination unit further includes:

[0198] A third comparison unit, configured to compare the initial masked image and the attention heat map, and if the initial masked image does not include the attention area, eliminate the initial masked image, and repeat the process of obtaining the initial masked image until the obtained initial masked image includes the attention area

[0199] A fourth comparison unit, configured to compare the initial cropped image and the attention heat map, and if the initial cropped image does not include the attention area, eliminate the initial cropped image, and repeat the process of obtaining the initial cropped image until the obtained initial cropped image includes the attention area.

[0200] In one embodiment, the apparatus further includes:

[0201] A rotation processing and color perturbation processing unit, configured to perform rotation processing and color perturbation processing on the second medical image to obtain a rotated image and a color perturbed image, and use the rotated image and the color perturbed image as augmented images of the second medical image.

[0202] Each module in the above apparatus for generating an image detection model can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.

[0203] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structural diagram may be as Figure 16As shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for generating an image detection model device. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0204] Those skilled in the art can understand that Figure 16 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0205] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the following steps are implemented:

[0206] Obtain the medical image to be processed;

[0207] Input the medical image to be processed into the image detection model to determine the region of interest;

[0208] The image detection model is obtained by training an initial model with multiple second medical images, multiple amplified images corresponding to each second medical image, and the annotation image of each second medical image;

[0209] The multiple amplified images corresponding to the second medical image are obtained by amplifying the second medical image based on the attention heat map;

[0210] The attention heat map is obtained through real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0211] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the following steps are implemented:

[0212] Obtain a medical image to be processed;

[0213] Input the medical image to be processed into an image detection model to determine the region of interest;

[0214] The image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation images of each second medical image;

[0215] The multiple augmented images corresponding to the second medical image are obtained by augmenting the second medical image based on an attention heat map;

[0216] The attention heat map is obtained by real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0217] In one embodiment, a computer program product is provided, including a computer program, which when executed by a processor implements the following steps:

[0218] Obtain a medical image to be processed;

[0219] Input the medical image to be processed into an image detection model to determine the region of interest;

[0220] The image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation images of each second medical image;

[0221] The multiple augmented images corresponding to the second medical image are obtained by augmenting the second medical image based on an attention heat map;

[0222] The attention heat map is obtained by real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

[0223] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0224] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0225] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0226] The above-described embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for generating an image detection model, characterized in that, the method comprises: obtaining eye movement data of a first medical image, and determining an attention training heat map of the first medical image based on the eye movement data, wherein the eye movement data is collected when a person views the first medical image; training a first initial model through multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model; inputting a second medical image into the attention prediction model to obtain an attention heat map, determining a segmentation mask of the attention heat map, and determining an eye movement augmented image based on the segmentation mask and the second medical image, wherein the eye movement augmented image includes an attention area; performing augmentation processing on the second medical image in multiple preset manners to obtain multiple initial images, and determining multiple target images based on the attention heat map and the multiple initial images, wherein each target image includes the attention area; determining the eye movement augmented image and the multiple target images as multiple extended images of the second medical image; training a second initial model through multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image to obtain an image detection model.

2. The method according to claim 1, characterized in that, the eye movement data includes the sampling time and sampling position of multiple sampling points; the determining the attention training heat map of the first medical image based on the eye movement data includes: determining a fixation point and a saccade point among the multiple sampling points according to the sampling time of the multiple sampling points, wherein the fixation point includes a fixation duration greater than a preset duration, the saccade point includes multiple discontinuous fixation durations, and the fixation duration includes multiple consecutive sampling times; marking the fixation point and the saccade point on a mask image based on the sampling positions of the fixation point and the saccade point to obtain an image to be processed, wherein the mask image has the same size as the first medical image; performing Gaussian smoothing processing on the image to be processed to obtain the attention training heat map of the first medical image.

3. The method according to claim 1, characterized in that, the multiple manners include a random masking manner and a random cropping manner; the performing augmentation processing on the second medical image in multiple preset manners to obtain multiple initial images, and determining multiple target images based on the attention heat map and the multiple initial images includes: processing the second medical image by using the random masking manner to obtain an initial masked image; comparing the initial masked image with the attention heat map, and if the initial masked image includes the attention area, using the initial masked image as a target image; processing the second medical image by using the random cropping manner to obtain an initial cropped image; comparing the initial cropped image with the attention heat map, and if the initial cropped image includes the attention area, using the initial cropped image as a target image.

4. The method according to claim 3, wherein, after processing the second medical image by using a random masking method to obtain an initial masked image, the method further includes: comparing the initial masked image with the attention heat map, if the initial masked image does not include the attention area, removing the initial masked image, and repeating the process of obtaining the initial masked image until the obtained initial masked image includes the attention area.

5. The method according to claim 3, wherein, after processing the second medical image by using a random cropping method to obtain an initial cropped image, the method further includes: comparing the initial cropped image with the attention heat map, if the initial cropped image does not include the attention area, removing the initial cropped image, and repeating the process of obtaining the initial cropped image until the obtained initial cropped image includes the attention area.

6. The method according to claim 3, wherein, the method further includes: performing rotation processing and color perturbation processing on the second medical image to obtain a rotated image and a color-perturbed image, and using the rotated image and the color-perturbed image as augmented images of the second medical image.

7. The method according to claim 3, wherein, processing the second medical image by using a random masking method to obtain an initial masked image, includes: randomly selecting a pixel point in the second medical image, taking the pixel point as a center point, determining a masked area in the second medical image, and setting the pixel values of all pixel points in the masked area to 0 to obtain the initial masked image.

8. An image detection method, wherein, the method includes: acquiring a medical image to be processed; inputting the medical image to be processed into an image detection model to determine an area of interest; the image detection model is obtained by training an initial model with multiple second medical images, multiple augmented images corresponding to each second medical image, and an annotation image of each second medical image; the process of determining multiple augmented images corresponding to a second medical image includes: determining a segmentation mask of an attention heat map, and determining an eye movement augmented image based on the segmentation mask and the second medical image, wherein the eye movement augmented image includes an attention area; performing augmentation processing on the second medical image in a plurality of preset manners to obtain a plurality of initial images, and determining a plurality of target images based on the attention heat map and the plurality of initial images, wherein each target image includes the attention area; determining the eye movement augmented image and the plurality of target images as a plurality of extended images of the second medical image; the attention heat map is related to simulated eye movement data.

9. An apparatus for generating an image detection model, wherein, the apparatus includes: an eye movement data processing module, configured to acquire eye movement data of a first medical image, and determine an attention training heat map of the first medical image based on the eye movement data, wherein the eye movement data is collected when a person views the medical image; An attention prediction model training model is used to train a first initial model through multiple first medical images and the attention training heat map of each first medical image to obtain an attention prediction model; An eye movement augmented image determination unit is configured to input a second medical image into the attention prediction model to obtain an attention heat map, determine a segmentation mask of the attention heat map, and determine an eye movement augmented image based on the segmentation mask and the second medical image, where the eye movement augmented image includes the attention area; A target image determination unit is configured to perform augmentation processing on the second medical image in multiple preset manners to obtain multiple initial images, and determine multiple target images based on the attention heat map and the multiple initial images, where each target image includes the attention area; determine the eye movement augmented image and the multiple target images as multiple extended images of the second medical image; An image detection model training module is used to train a second initial model through multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image to obtain an image detection model.

10. A computer device includes a memory and a processor, and the memory stores a computer program, wherein, when the processor executes the computer program, it realizes: acquiring a medical image to be processed; inputting the medical image to be processed into an image detection model to determine an area of interest; the image detection model is obtained by training an initial model through multiple second medical images, multiple augmented images corresponding to each second medical image, and the annotation image of each second medical image; the determination process of the multiple augmented images corresponding to the second medical image includes: determining a segmentation mask of the attention heat map, and determining an eye movement augmented image based on the segmentation mask and the second medical image, where the eye movement augmented image includes the attention area; performing augmentation processing on the second medical image in multiple preset manners to obtain multiple initial images, and determining multiple target images based on the attention heat map and the multiple initial images, where each target image includes the attention area; determining the eye movement augmented image and the multiple target images as multiple extended images of the second medical image; the attention heat map is obtained through real-time acquisition or simulation, and the attention heat map is related to the eye movement data when a person views the second medical image.

Images