Image target detection method based on sequence analysis
Through the analysis and scoring system of the possibility of multi-level target existence in CT image sequences, combined with the ResNet-50 model integrating spatial transformation network, the problem of time-consuming and insufficient accuracy of traditional CT image analysis methods is solved, achieving a more efficient and Accurate target detection.
Patent Information
- Application Number
- CN202411663077.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-20
- Publication Date
- 2025-01-21
AI Technical Summary
Traditional CT image analysis relies on the experience and subjective judgment of radiologists, which makes it time-consuming to process large numbers of images and affects the consistency and accuracy of diagnosis. Deep learning methods fail to make full use of key information in images.
By analyzing the possibility of multi-level targets in the CT image sequence, including a three-level attention priority judgment and scoring system, combined with the ResNet-50 model integrating the spatial transformation network for target detection, the local amplification degree of the cavity area is dynamically adjusted.
It significantly improves the judgment ability of the model, improves the accuracy and reliability of target detection, and optimizes the efficiency and accuracy of the medical diagnosis process.
Smart Images
Figure CN119338801A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of medical image analysis, and in particular relates to a computer tomography (CT) image target detection method based on sequence analysis, which is used to optimize the training and application of deep learning models through multi-level target existence possibility judgment. Background Art
[0002] Chest computed tomography (CT) is an important technology used in modern medicine to detect and diagnose diseases in the chest cavity. CT images can show the structures of the lungs, heart, and large blood vessels in the chest cavity in detail, and are particularly critical for the detection of lesions such as pneumonia and tumors. With the development of medical technology, the number and quality of CT images have been significantly improved, but this has also led to an increase in the difficulty of data processing.
[0003] Traditional CT image analysis relies on the experience and subjective judgment of radiologists. This method is time-consuming when processing a large number of images and may affect the consistency and accuracy of diagnosis due to individual differences. Therefore, automated image processing methods, especially those based on deep learning, have become a key technology to improve the speed and accuracy of image analysis. Although deep learning has made significant progress in the field of image recognition, most methods are still directly applied to the entire image and fail to fully utilize the key information in the image.
[0004] In order to solve these problems, a method is needed that can carefully analyze the CT image sequence and make judgments based on the possibility of the existence of targets in the image. The present invention can not only accurately evaluate the importance of each image area by analyzing the possibility of the existence of multiple levels of targets in the CT image sequence, but also perform more accurate target detection based on the accumulated information in the sequence. This method significantly improves the judgment ability of the model, and is particularly suitable for the analysis of long-sequence CT images, thereby optimizing the efficiency and accuracy of the entire medical diagnosis process. Summary of the invention
[0005] In view of the above-mentioned deficiencies of the prior art, the present invention proposes an image target detection method based on sequence analysis, comprising the following steps:
[0006] Step 1: Perform attention priority judgment on each CT image in the CT image sequence to obtain the cavity area attention priority attribute in each CT image in the CT image sequence;
[0007] Step 2: Establish a scoring system for the cavity area in each CT image in the CT image sequence;
[0008] Step 3: Selecting CT image samples from the CT image sequence;
[0009] Step 4: Determine the priority attribute of the entire CT image sequence based on the CT image samples;
[0010] Step 4.1: Calculate the mean score of cavity area A and cavity area B in each selected CT image sample;
[0011] Step 4.2: Setting a threshold thS, which is used to compare the relationship between the mean score of the cavity area and the threshold, and judging the priority attribute of the entire CT image sequence according to the comparison result between the mean score and the threshold;
[0012] Step 4.3: further analyzing the two cavity regions in the CT image samples in the sequence to obtain the cavity region of focus of the entire CT image sequence;
[0013] Step 5: Based on the attention priority attributes in the CT image sequence, the ResNet-50 model with integrated spatial transformer network is used to perform object detection in CT images.
[0014] Furthermore, in the step 1, an attention priority is judged for each CT image in the CT image sequence to obtain an attention priority attribute of the cavity area in each CT image in the CT image sequence, specifically: for each CT image, firstly, the entropy of the entire CT image is calculated, and according to the set image information amount threshold, the CT image information amount level is divided into three levels: high, medium and low, as the first layer of attention priority; then, the image is binarized using the Otsu method, and then morphological operations are applied to the binarized image to locate two cavity areas in the chest area, the entropy value of the located cavity area is calculated, the cavity area information amount threshold is set, and the cavity area information amount level is divided into three levels: high, medium and low, as the second layer of attention priority; then, the gray level co-occurrence matrix is applied to the located cavity area to extract homogeneity features, the cavity area homogeneity threshold is set, and the cavity area homogeneity level is divided into three levels: high, medium and low, as the third layer of attention priority; finally, the three layers of attention priority are combined to form a multi-level attention priority attribute.
[0015] Furthermore, in step 2, a scoring system is established for the cavity area in each CT image in the CT image sequence, specifically: when the cavity area attention priority attribute is 3 highs, the cavity area obtains 10 points; when the cavity area attention priority attribute is 2 highs and 1 medium, the cavity area obtains 8 points; when the cavity area attention priority attribute is 1 high and 2 mediums, the cavity area obtains 6 points; when the cavity area attention priority attribute is 3 mediums, the cavity area obtains 5 points; when the cavity area attention priority attribute is 2 mediums and 1 low, the cavity area obtains 3 points; when the cavity area attention priority attribute is 1 medium and 2 lows, the cavity area obtains 2 points; when the cavity area attention priority attribute is 3 lows, the cavity area obtains 1 point.
[0016] Furthermore, in step 3, CT image samples in the CT image sequence are selected, specifically as follows: first, a central image in the CT image sequence is determined; then, based on the selected central image, one image before and after it is selected, for a total of three consecutive images, as CT image samples for analysis.
[0017] Further, the central image in the CT image sequence is determined, specifically: if the length of the CT image sequence is an odd number, the middle image is directly selected as the central image; if the length of the CT image sequence is an even number, the first one of the two middle images is selected as the central image.
[0018] Furthermore, in step 4.1, the mean scores of cavity area A and cavity area B in each selected CT image sample are calculated respectively, specifically: the scores of each cavity area in the sequence are summed and divided by the number of images to obtain the mean score AJa of cavity area A and the mean score BJa of cavity area B.
[0019] Furthermore, in the step 4.2, a threshold thS is set to compare the relationship between the mean score of the cavity area and the threshold. According to the comparison result of the mean score and the threshold, the attention priority attribute of the entire CT image sequence is judged, specifically: if AJa and BJa are both greater than the threshold thS, the attention priority attribute of the CT image sequence is judged to be high; if AJa and BJa are both less than the threshold thS, the attention priority attribute of the CT image sequence is judged to be low; if one of AJa and BJa is higher than the threshold thS and the other is lower than thS, the attention priority attribute of the CT image sequence is judged to be medium.
[0020] Furthermore, in step 4.3, the two cavity areas in the CT image samples in the sequence are further analyzed to obtain the cavity area that is the focus of the entire CT image sequence, specifically: if the average score AJa of cavity area A in the CT image samples is higher than the average score BJa of cavity area B, it is judged that the CT image sequence focuses on cavity area A; otherwise, it is judged that the CT image sequence focuses on cavity area B.
[0021] Furthermore, in step 5, based on the attention priority attribute in the CT image sequence, the ResNet-50 model of the integrated spatial transformation network is used to perform target detection on the CT image, specifically: first, the CT image on which the attention priority judgment has been performed is input into the ResNet-50 model of the integrated spatial transformation network; secondly, based on the attention priority attribute of the third layer and the attention priority attribute of the second layer of the CT image, the integrated spatial transformation network module is used to automatically adjust the local magnification degree of the identified cavity areas A and B in the image through local affine transformation, and then the adjusted image is transmitted to the subsequent layers of ResNet-50 for feature extraction, and finally, target detection is performed through the binary classification output layer of the model to determine whether there is a target.
[0022] Furthermore, the attention priority attribute of the third layer and the attention priority attribute of the second layer based on the CT image automatically adjusts the local magnification degree of the identified cavity areas A and B in the image through local affine transformation using an integrated spatial transformation network module, specifically: firstly, different degrees of local magnification multiples are selected for the cavity area according to the attention priority attribute of the second layer, and then the local magnification multiples selected for the cavity area are corrected according to the attention priority attribute of the third layer, specifically: firstly, different degrees of local magnification multiples are selected for the cavity area according to the attention priority attribute of the second layer, including: when the attention priority attribute of the cavity area A or B contains 3 highs, a local magnification of 1.5 times is performed; when it contains 2 highs and 1 medium, a local magnification of 1.4 times is performed; when it contains 1 high and 2 medium, a local magnification of 1.3 times is performed; when it contains 3 mediums, a local magnification of 1.2 times is performed; when it contains 2 mediums and 1 low , perform a local magnification of 1.15 times; when it includes 1 medium and 2 low, perform a local magnification of 1.1 times; when it includes 3 low, keep the original size without magnification; then modify the local magnification factor selected for the cavity area according to the judgment result of the attention priority attribute of the third layer, including: when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is high, the local magnification factors of the two cavity areas in the CT image are additionally magnified by 1.2 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is medium, the local magnification factors of the two cavity areas in the CT image are additionally magnified by 1.1 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is low, the local magnification factors of the two cavity areas in the CT image are maintained at the original size without magnification.
[0023] Compared with the traditional method, the present invention has the following advantages, thereby solving the corresponding technical problems:
[0024] 1. By implementing a three-level target existence possibility evaluation system, this multi-level judgment method allows for a more in-depth analysis of CT images. In particular, the third-level judgment is performed on the entire image sequence, which allows the model to not only consider the features in a single image, but also conduct a comprehensive evaluation based on the change trend of the entire sequence, thereby greatly improving the accuracy and reliability of target detection.
[0025] 2. By constructing a scoring system based on the possibility of the existence of the second-level target, the present invention can assign a specific score to each cavity area, and judge the possibility of the existence of the target in the entire sequence based on the average score of each image in the sequence. This method not only increases the precise control of target detection, but also clarifies the cavity area that should be focused on in the sequence, thereby providing more targeted information for subsequent target detection.
[0026] 3. Application of the ResNet-50 model with integrated spatial transformer network, which dynamically adjusts the local magnification for cavity areas A and B in the CT image sequence. This adjustment is based on the evaluation results of the three-level target existence possibility, which enables the model to dynamically adjust the magnification according to the target existence possibility of the cavity area. First, according to the second-level target existence possibility attribute, the cavity area is assigned different degrees of preliminary local magnification. Then, according to the third-level sequence-level target existence possibility judgment results, these local magnifications are further adjusted. If the target existence possibility of the entire sequence is high, the local magnification is additionally enhanced. The purpose of this is to increase the focus on high-risk areas to improve the sensitivity and accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 It is a flow chart of an image target detection method based on sequence analysis; DETAILED DESCRIPTION
[0028] The following will clearly and completely describe the technical solution in the embodiment of the present invention in conjunction with the accompanying drawings in the embodiment of the present invention. The method includes the following steps:
[0029] Step 1: Perform attention priority judgment on each CT image in the CT image sequence to obtain the cavity area attention priority attribute in each CT image in the CT image sequence;
[0030] Step 2: Establish a scoring system for the cavity area in each CT image in the CT image sequence;
[0031] Step 3: Selecting CT image samples from the CT image sequence;
[0032] Step 4: Determine the priority attribute of the entire CT image sequence based on the CT image samples;
[0033] Step 4.1: Calculate the mean score of cavity area A and cavity area B in each selected CT image sample;
[0034] Step 4.2: Setting a threshold thS, which is used to compare the relationship between the mean score of the cavity area and the threshold, and judging the priority attribute of the entire CT image sequence according to the comparison result between the mean score and the threshold;
[0035] Step 4.3: further analyzing the two cavity regions in the CT image samples in the sequence to obtain the cavity region of focus of the entire CT image sequence;
[0036] Step 5: Based on the attention priority attributes in the CT image sequence, the ResNet-50 model with integrated spatial transformer network is used to perform object detection in CT images.
[0037] Furthermore, in the step 1, an attention priority is judged for each CT image in the CT image sequence to obtain an attention priority attribute of the cavity area in each CT image in the CT image sequence, specifically: for each CT image, firstly, the entropy of the entire CT image is calculated, and according to the set image information amount threshold, the CT image information amount level is divided into three levels: high, medium and low, as the first layer of attention priority; then, the image is binarized using the Otsu method, and then morphological operations are applied to the binarized image to locate two cavity areas in the chest area, the entropy value of the located cavity area is calculated, the cavity area information amount threshold is set, and the cavity area information amount level is divided into three levels: high, medium and low, as the second layer of attention priority; then, the gray level co-occurrence matrix is applied to the located cavity area to extract homogeneity features, the cavity area homogeneity threshold is set, and the cavity area homogeneity level is divided into three levels: high, medium and low, as the third layer of attention priority; finally, the three layers of attention priority are combined to form a multi-level attention priority attribute.
[0038] Furthermore, in step 2, a scoring system is established for the cavity area in each CT image in the CT image sequence, specifically: when the cavity area attention priority attribute is 3 highs, the cavity area obtains 10 points; when the cavity area attention priority attribute is 2 highs and 1 medium, the cavity area obtains 8 points; when the cavity area attention priority attribute is 1 high and 2 mediums, the cavity area obtains 6 points; when the cavity area attention priority attribute is 3 mediums, the cavity area obtains 5 points; when the cavity area attention priority attribute is 2 mediums and 1 low, the cavity area obtains 3 points; when the cavity area attention priority attribute is 1 medium and 2 lows, the cavity area obtains 2 points; when the cavity area attention priority attribute is 3 lows, the cavity area obtains 1 point.
[0039] Furthermore, in step 3, CT image samples in the CT image sequence are selected, specifically as follows: first, a central image in the CT image sequence is determined; then, based on the selected central image, one image before and after it is selected, for a total of three consecutive images, as CT image samples for analysis.
[0040] Further, the central image in the CT image sequence is determined, specifically: if the length of the CT image sequence is an odd number, the middle image is directly selected as the central image; if the length of the CT image sequence is an even number, the first one of the two middle images is selected as the central image.
[0041] Furthermore, in step 4.1, the mean scores of cavity area A and cavity area B in each selected CT image sample are calculated respectively, specifically: the scores of each cavity area in the sequence are summed and divided by the number of images to obtain the mean score AJa of cavity area A and the mean score BJa of cavity area B.
[0042] Furthermore, in the step 4.2, a threshold thS is set to compare the relationship between the mean score of the cavity area and the threshold. According to the comparison result of the mean score and the threshold, the attention priority attribute of the entire CT image sequence is judged, specifically: if AJa and BJa are both greater than the threshold thS, the attention priority attribute of the CT image sequence is judged to be high; if AJa and BJa are both less than the threshold thS, the attention priority attribute of the CT image sequence is judged to be low; if one of AJa and BJa is higher than the threshold thS and the other is lower than thS, the attention priority attribute of the CT image sequence is judged to be medium.
[0043] Furthermore, in step 4.3, the two cavity areas in the CT image samples in the sequence are further analyzed to obtain the cavity area that is the focus of the entire CT image sequence, specifically: if the average score AJa of cavity area A in the CT image samples is higher than the average score BJa of cavity area B, it is judged that the CT image sequence focuses on cavity area A; otherwise, it is judged that the CT image sequence focuses on cavity area B.
[0044] Furthermore, in step 5, based on the attention priority attribute in the CT image sequence, the ResNet-50 model of the integrated spatial transformation network is used to perform target detection on the CT image, specifically: first, the CT image on which the attention priority judgment has been performed is input into the ResNet-50 model of the integrated spatial transformation network; secondly, based on the attention priority attribute of the third layer and the attention priority attribute of the second layer of the CT image, the integrated spatial transformation network module is used to automatically adjust the local magnification degree of the identified cavity areas A and B in the image through local affine transformation, and then the adjusted image is transmitted to the subsequent layers of ResNet-50 for feature extraction, and finally, target detection is performed through the binary classification output layer of the model to determine whether there is a target.
[0045] Furthermore, the attention priority attribute of the third layer and the attention priority attribute of the second layer based on the CT image automatically adjusts the local magnification degree of the identified cavity areas A and B in the image through local affine transformation using an integrated spatial transformation network module, specifically: firstly, different degrees of local magnification multiples are selected for the cavity area according to the attention priority attribute of the second layer, and then the local magnification multiples selected for the cavity area are corrected according to the attention priority attribute of the third layer, specifically: firstly, different degrees of local magnification multiples are selected for the cavity area according to the attention priority attribute of the second layer, including: when the attention priority attribute of the cavity area A or B contains 3 highs, a local magnification of 1.5 times is performed; when it contains 2 highs and 1 medium, a local magnification of 1.4 times is performed; when it contains 1 high and 2 medium, a local magnification of 1.3 times is performed; when it contains 3 mediums, a local magnification of 1.2 times is performed; when it contains 2 mediums and 1 low , perform a local magnification of 1.15 times; when it includes 1 medium and 2 low, perform a local magnification of 1.1 times; when it includes 3 low, keep the original size without magnification; then modify the local magnification factor selected for the cavity area according to the judgment result of the attention priority attribute of the third layer, including: when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is high, the local magnification factors of the two cavity areas in the CT image are additionally magnified by 1.2 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is medium, the local magnification factors of the two cavity areas in the CT image are additionally magnified by 1.1 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is low, the local magnification factors of the two cavity areas in the CT image are maintained at the original size without magnification.
[0046] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.
Claims
1. A method for image target detection based on sequence analysis, comprising the following steps: Step 1: Perform attention priority judgment on each CT image in the CT image sequence to obtain the cavity area attention priority attribute in each CT image in the CT image sequence; Step 2: Establish a scoring system for the cavity area in each CT image in the CT image sequence; Step 3: Selecting CT image samples from the CT image sequence; Step 4: Determine the priority attribute of the entire CT image sequence based on the CT image samples; Step 4.1: Calculate the mean score of cavity area A and cavity area B in each selected CT image sample; Step 4.2: Setting a threshold thS, which is used to compare the relationship between the mean score of the cavity area and the threshold, and judging the priority attribute of the entire CT image sequence according to the comparison result between the mean score and the threshold; Step 4.3: further analyzing the two cavity regions in the CT image samples in the sequence to obtain the cavity region of focus of the entire CT image sequence; Step 5: Based on the attention priority attributes in the CT image sequence, the ResNet-50 model with integrated spatial transformer network is used to perform object detection in CT images.
2. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In the step 1, an attention priority is judged for each CT image in the CT image sequence to obtain an attention priority attribute of the cavity area in each CT image in the CT image sequence, specifically: for each CT image, firstly, the entropy of the entire CT image is calculated, and according to the set image information amount threshold, the CT image information amount level is divided into three levels: high, medium and low, as the first layer of attention priority; then, the image is binarized using the Otsu method, and then a morphological operation is applied to the binarized image to locate two cavity areas in the chest area, the entropy value of the located cavity area is calculated, and the cavity area information amount threshold is set, and the cavity area information amount level is divided into three levels: high, medium and low, as the second layer of attention priority; Then, the gray level co-occurrence matrix is applied to the located cavity area to extract homogeneity features, and the cavity area homogeneity threshold is set. The cavity area homogeneity level is divided into three levels: high, medium and low, as the third level of attention priority; finally, the three levels of attention priority are combined to form a multi-level attention priority attribute.
3. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In the step 2, a scoring system is established for the cavity area in each CT image in the CT image sequence, specifically: when the cavity area attention priority attribute is 3 highs, the cavity area obtains 10 points; when the cavity area attention priority attribute is 2 highs and 1 medium, the cavity area obtains 8 points; when the cavity area attention priority attribute is 1 high and 2 mediums, the cavity area obtains 6 points; when the cavity area attention priority attribute is 3 mediums, the cavity area obtains 5 points; when the cavity area attention priority attribute is 2 mediums and 1 low, the cavity area obtains 3 points; when the cavity area attention priority attribute is 1 medium and 2 lows, the cavity area obtains 2 points; when the cavity area attention priority attribute is 3 lows, the cavity area obtains 1 point.
4. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In step 3, CT image samples in the CT image sequence are selected, specifically: first, the central image in the CT image sequence is determined; then, based on the selected central image, one image before and after it is selected, a total of three consecutive images are selected as CT image samples for analysis.
5. The image target detection method based on sequence analysis as claimed in claim 3, characterized in that: The central image in the CT image sequence is determined, specifically: if the length of the CT image sequence is an odd number, the middle image is directly selected as the central image; if the length of the CT image sequence is an even number, the first one of the two middle images is selected as the central image.
6. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In step 4.1, the mean scores of cavity area A and cavity area B in each selected CT image sample are calculated respectively, specifically: the scores of each cavity area in the sequence are summed and divided by the number of images to obtain the mean score AJa of cavity area A and the mean score BJa of cavity area B.
7. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In the step 4.2, a threshold thS is set to compare the relationship between the mean score of the cavity area and the threshold. According to the comparison result of the mean score and the threshold, the attention priority attribute of the entire CT image sequence is judged, specifically: if AJa and BJa are both greater than the threshold thS, the attention priority attribute of the CT image sequence is judged to be high; if AJa and BJa are both less than the threshold thS, the attention priority attribute of the CT image sequence is judged to be low; if one of AJa and BJa is higher than the threshold thS and the other is lower than thS, the attention priority attribute of the CT image sequence is judged to be medium.
8. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In step 4.3, the two cavity areas in the CT image samples in the sequence are further analyzed to obtain the cavity area that is the focus of the entire CT image sequence. Specifically, if the average score AJa of cavity area A in the CT image samples is higher than the average score BJa of cavity area B, it is judged that the CT image sequence focuses on cavity area A; otherwise, it is judged that the CT image sequence focuses on cavity area B.
9. The image target detection method based on sequence analysis as claimed in claim 1, characterized in that: In the step 5, based on the attention priority attribute in the CT image sequence, the ResNet-50 model of the integrated space transform network is used to perform target detection on the CT image, specifically: first, the CT image on which the attention priority judgment has been performed is input into the ResNet-50 model of the integrated space transform network; secondly, based on the attention priority attribute of the third layer and the attention priority attribute of the second layer of the CT image, the integrated space transform network module is used to automatically adjust the local magnification degree of the identified cavity areas A and B in the image through local affine transformation, and then the adjusted image is transmitted to the subsequent layer of ResNet-50 for feature extraction, and finally, target detection is performed through the binary classification output layer of the model to determine whether there is a target.
10. The image target detection method based on sequence analysis according to claim 9, characterized in that: The method uses an integrated space transformation network module to automatically adjust the local magnification degree of the identified cavity areas A and B in the image through local affine transformation based on the third-layer attention priority attribute and the second-layer attention priority attribute of the CT image. Specifically, the method comprises: firstly selecting different degrees of local magnification multiples for the cavity area according to the attention priority attribute of the second layer, including: when the attention priority attribute of the cavity area A or B includes 3 highs, performing a local magnification of 1.5 times; when it includes 2 highs and 1 medium, performing a local magnification of 1.4 times; when it includes 1 high and 2 medium, performing a local magnification of 1.3 times; when it includes 3 medium, performing a local magnification of 1.2 times; when it includes 2 medium and 1 low, performing a local magnification of 1.15 times; when it includes 1 medium and 2 low, performing a local magnification of 1.1 times; when it includes 2 highs and 1 medium, performing a local magnification of 1.2 times; when it includes 2 mediums and 1 low, performing a local magnification of 1.15 times; when it includes 1 medium and 2 low, performing a local magnification of 1.2 times; when it includes 2 mediums and 1 ... When it contains 3 low, the original size is maintained without enlarging; then the local magnification factor of the cavity area selection is corrected according to the judgment result of the attention priority attribute of the third layer, including: when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is high, the local magnification factors of the two cavity areas in the CT image are additionally enlarged by 1.2 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is medium, the local magnification factors of the two cavity areas in the CT image are additionally enlarged by 1.1 times; when the attention priority attribute of the entire sequence in the third layer attention priority attribute of the sequence to which the CT image belongs is low, the local magnification factors of the two cavity areas in the CT image are maintained at the original size without enlarging.