Target detection method based on target existence possibility

Abstract

The invention relates to a target detection method based on target existence possibility. According to the method, firstly, a CT image is preprocessed, a thoracic cavity region is calibrated, and a foreground region and a background region are distinguished through threshold segmentation; then, the foreground and background proportions of these regions are calculated in detail to assess similarity and set target presence likelihood attributes. Finally, according to the attributes, an optimized training strategy is adopted to train a ResNet-50 model, including applying different learning rates to images with different possibilities, and performing special labeling on key areas. According to the method, the precision and efficiency of target detection can be remarkably improved, and the method is particularly suitable for quickly and accurately identifying and positioning key targets in large-scale medical image data.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing, and in particular relates to a method based on computer vision and deep learning for identifying and evaluating the possibility of the existence of potential targets in the chest area from a CT image sequence. Background Art

[0002] Chest computed tomography (CT) is an important tool in modern medical diagnosis, especially in observing the internal structures of the chest cavity such as the lungs and large blood vessels. It not only helps to identify diseases such as pneumonia and tumors, but also with the advancement of technology, the significant increase in CT scanning speed and data volume provides more details for diagnosis. However, this also brings huge data processing challenges.

[0003] Traditional CT image analysis mainly relies on visual inspection by radiologists, which is time-consuming and susceptible to subjective factors, especially when dealing with large amounts of data. To improve analysis efficiency and accuracy, automated image processing technology, especially deep learning technology, has been introduced into CT image analysis.

[0004] Although deep learning, especially convolutional neural networks (CNNs), has achieved remarkable results in the field of image recognition, most existing methods are directly applied to the entire image when processing CT images, without fully utilizing the key area information in the image. This method not only increases the computational burden, but also limits the accuracy and specificity of diagnosis.

[0005] The present invention uses the existence possibility attribute of the chest area defined in the preprocessing process to adjust the training parameters of the ResNet-50 model, such as the learning rate and the labeling of specific areas, to improve the model's attention and recognition ability to the target area. In this way, the present invention not only optimizes the learning efficiency of the model, but also improves the feasibility and effect of automated analysis when processing large-scale CT image data. Summary of the invention

[0006] In view of the above-mentioned deficiencies of the prior art, the present invention proposes a target detection method based on the possibility of target existence, comprising the following steps:

[0007] Step 1: preprocessing the CT image to obtain the thoracic region and two cavity regions in the thoracic region in the CT image;

[0008] Step 2: Perform threshold segmentation on the chest area in the CT image to obtain the foreground area and background area in the two cavity areas in the CT image;

[0009] Step 3: Calculate the number of pixels Af and Bf in the foreground area of ​​the two cavity areas in the chest area of ​​the CT image, namely, cavity area A and cavity area B; calculate the average value Ab_avg and Bb_avg of the number of pixels in the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image;

[0010] Step 4: Calculate the ratios ARfb and BRfb of the foreground area and the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image respectively;

[0011] Step 5: Calculate the similarity S of the foreground area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image;

[0012] Step 6: Set the ratio threshold thr of the foreground area and the background area to determine the possible existence attribute of the CT image target;

[0013] Step 7: According to the target existence possibility attribute of the CT image, the deep learning model is trained using a training strategy based on target existence possibility.

[0014] Furthermore, in step 2, threshold segmentation is performed on the chest area in the CT image, specifically: all pixels in the chest area of ​​the CT image are segmented according to the segmentation threshold thdiv, and pixel points whose pixel values ​​are greater than the segmentation threshold thdiv are divided into foreground areas, and pixel points whose pixel values ​​are less than the segmentation threshold thdiv are divided into background areas.

[0015] Furthermore, in step 4, the ratios ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image are calculated respectively, specifically: ARfb=Af / Ab_avg, BRfb=Bf / Bb_avg.

[0016] Furthermore, in step 5, the similarity S of the foreground area of ​​cavity area A and cavity area B in the chest area of ​​the CT image is calculated, specifically: S = abs[Af / (Af+Ab_avg)-Bf / (Bf+Bb_avg)], where abs[*] is the absolute value.

[0017] Furthermore, in step 6, a threshold value thr of the proportion of the foreground area and the background area is set to judge the possibility attribute of the existence of the CT image target, specifically: when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both greater than the proportion threshold thr, a possibility one judgment is performed; when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both less than the proportion threshold thr, a possibility two judgment is performed; when the proportion ARfb of the foreground area and the background area of ​​the cavity area A is greater than the threshold thr, and the proportion BRfb of the foreground area and the background area of ​​the cavity area B is less than the threshold thr, or when the proportion BRfb of the foreground area and the background area of ​​the cavity area B is greater than the threshold thr, and the proportion ARfb of the foreground area and the background area of ​​the cavity area A is less than the threshold thr, a possibility three judgment is performed.

[0018] Furthermore, the possibility judgment is performed as follows: a similarity threshold ths is set, and when the similarity S is greater than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be medium, and the possibility attributes of the target existence in the cavity area A and the cavity B are both medium-medium; when the similarity S is less than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be high, and then the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are compared. When ARfb is greater than BRfb, the possibility attribute of the target existence in the cavity area A is high-high, and the possibility attribute of the target existence in the cavity area B is high-medium; when ARfb is less than BRfb, the possibility attribute of the target existence in the cavity area B is high-high, and the possibility attribute of the target existence in the cavity area A is high-medium.

[0019] Furthermore, the second possibility judgment is specifically as follows: a similarity threshold ths is set, and when the similarity S is greater than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be low, and the possibility attribute of the target existence in the cavity area A and the cavity area B is low-low; when the similarity S is less than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be medium, and then the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are compared. When ARfb is greater than BRfb, the possibility attribute of the target existence in the cavity area A is medium-medium, and the possibility attribute of the target existence in the cavity area B is medium-low; when ARfb is less than BRfb, the possibility attribute of the target existence in the cavity area B is medium-medium, and the possibility attribute of the target existence in the cavity area A is medium-low.

[0020] Furthermore, the third possibility judgment is specifically: when the ratio ARfb of the foreground area and the background area of ​​cavity area A is greater than the threshold thr, and the ratio BRfb of the foreground area and the background area of ​​cavity area B is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-high, and the possibility attribute of the target existence in cavity area B is high-medium; when the ratio BRfb of the foreground area and the background area of ​​cavity area B is greater than the threshold thr, and the ratio ARfb of the foreground area and the background area of ​​cavity area A is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-medium, and the possibility attribute of the target existence in cavity area B is high-high.

[0021] Furthermore, in step 1, the CT image sequence is preprocessed, specifically: first, the contrast and brightness of the image are adjusted using an adaptive threshold technology; then, the processed image is segmented using a graph cut method, which optimizes the segmentation boundary by considering the relationship between pixels and effectively distinguishes different structures in the chest cavity; then, a region growing algorithm is used to identify and expand the connected areas in the segmented image, ensuring continuity and integrity while merging adjacent approximate areas by setting a similarity threshold; finally, feature matching is performed on the merged areas, and the SIFT feature matching technology is used to compare the similarities between the areas, so as to locate and mark the chest area and the two cavity areas inside it, and finally the chest area and the two cavity areas within the chest area in the CT image are obtained.

[0022] Furthermore, in step 7, according to the target existence possibility attribute of the CT image, a deep learning model is trained using a training strategy based on the target existence possibility, specifically: for CT images of the chest area with different target existence possibilities, different learning rates are applied to model training, and a high learning rate is used for CT images with high target existence possibility; a low learning rate is used for CT images with low target existence possibility; a medium learning rate is used for CT images with medium target existence possibility; the target existence possibility attributes of cavity area A and cavity area B are specially labeled to enhance the model's attention and recognition ability to these labeled areas, and the cavity area with high-high target existence possibility is labeled as priority-highest; the cavity area with high-medium target existence possibility is labeled as priority-high; the cavity area with medium-medium target existence possibility is labeled as priority-medium; and the cavity area with medium-low target existence possibility is labeled as priority-low.

[0023] Compared with the traditional target method, the present invention has the following advantages, thereby solving the corresponding technical problems:

[0024] 1. Through precise threshold segmentation and detailed analysis of foreground and background, the key cavity areas in CT images are effectively distinguished. This method makes the distinction between target and non-target areas in the image more obvious, thereby improving the accuracy of subsequent processing.

[0025] 2. By calculating and comparing the ratio of the foreground to background areas of cavity area A and cavity area B, this method can effectively identify and emphasize those areas with significant changes or anomalies. By calculating the similarity of these foreground areas, the consistency and reliability of the detection results can be further improved.

[0026] 3. The present invention allows the system to flexibly adjust the judgment criteria according to changes in the actual image content by finely adjusting the proportion thresholds of the foreground and background areas and combining similarity analysis, thereby more accurately identifying and classifying the probability of target existence. By combining the proportion threshold with the similarity threshold, this method not only improves the sensitivity of target detection, but also enhances the adaptability to changes in different scenarios, especially in the presence of obvious foreground and background differences, and can more effectively distinguish the probability levels of target existence, high, medium, and low. In addition, through detailed comparison and rating of different cavity areas, such as high-high and high-medium, the accuracy and practicality of target detection are further optimized, so that the model can not only determine the possibility of target existence, but also accurately evaluate the criticality of different areas, providing more accurate data support for subsequent diagnosis.

[0027] 4. The present invention significantly improves the accuracy and efficiency of target detection by optimizing the model training strategy for the target existence possibility attribute of each image in the CT image sequence. The method strengthens the learning and recognition of key areas, especially the high-probability areas, by adjusting the learning rate and performing special annotations. This personalized training strategy not only improves the model's ability to recognize targets in complex backgrounds, but also optimizes the use of computing resources and accelerates the automated image analysis process. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of a target detection method based on the possibility of target existence; DETAILED DESCRIPTION

[0029] 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:

[0030] Step 1: preprocessing the CT image to obtain the thoracic region and two cavity regions in the thoracic region in the CT image;

[0031] Step 2: Perform threshold segmentation on the chest area in the CT image to obtain the foreground area and background area in the two cavity areas in the CT image;

[0032] Step 3: Calculate the number of pixels Af and Bf in the foreground area of ​​the two cavity areas in the chest area of ​​the CT image, namely, cavity area A and cavity area B; calculate the average value Ab_avg and Bb_avg of the number of pixels in the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image;

[0033] Step 4: Calculate the ratios ARfb and BRfb of the foreground area and the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image respectively;

[0034] Step 5: Calculate the similarity S of the foreground area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image;

[0035] Step 6: Set the ratio threshold thr of the foreground area and the background area to determine the possible existence attribute of the CT image target;

[0036] Step 7: According to the target existence possibility attribute of the CT image, the deep learning model is trained using a training strategy based on target existence possibility.

[0037] Furthermore, in step 2, threshold segmentation is performed on the chest area in the CT image, specifically: all pixels in the chest area of ​​the CT image are segmented according to the segmentation threshold thdiv, and pixel points whose pixel values ​​are greater than the segmentation threshold thdiv are divided into foreground areas, and pixel points whose pixel values ​​are less than the segmentation threshold thdiv are divided into background areas.

[0038] Furthermore, in step 4, the ratios ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image are calculated respectively, specifically: ARfb=Af / Ab_avg, BRfb=Bf / Bb_avg.

[0039] Furthermore, in step 5, the similarity S of the foreground area of ​​cavity area A and cavity area B in the chest area of ​​the CT image is calculated, specifically: S = abs[Af / (Af+Ab_avg)-Bf / (Bf+Bb_avg)], where abs[*] is the absolute value.

[0040] Furthermore, in step 6, a threshold value thr of the proportion of the foreground area and the background area is set to judge the possibility attribute of the existence of the CT image target, specifically: when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both greater than the proportion threshold thr, a possibility one judgment is performed; when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both less than the proportion threshold thr, a possibility two judgment is performed; when the proportion ARfb of the foreground area and the background area of ​​the cavity area A is greater than the threshold thr, and the proportion BRfb of the foreground area and the background area of ​​the cavity area B is less than the threshold thr, or when the proportion BRfb of the foreground area and the background area of ​​the cavity area B is greater than the threshold thr, and the proportion ARfb of the foreground area and the background area of ​​the cavity area A is less than the threshold thr, a possibility three judgment is performed.

[0041] Furthermore, the possibility judgment is performed as follows: a similarity threshold ths is set, and when the similarity S is greater than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be medium, and the possibility attributes of the target existence in the cavity area A and the cavity B are both medium-medium; when the similarity S is less than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be high, and then the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are compared. When ARfb is greater than BRfb, the possibility attribute of the target existence in the cavity area A is high-high, and the possibility attribute of the target existence in the cavity area B is high-medium; when ARfb is less than BRfb, the possibility attribute of the target existence in the cavity area B is high-high, and the possibility attribute of the target existence in the cavity area A is high-medium.

[0042] Furthermore, the second possibility judgment is specifically as follows: a similarity threshold ths is set, and when the similarity S is greater than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be low, and the possibility attribute of the target existence in the cavity area A and the cavity area B is low-low; when the similarity S is less than the similarity threshold ths, the possibility attribute of the target existence in the chest area of ​​the CT image is judged to be medium, and then the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are compared. When ARfb is greater than BRfb, the possibility attribute of the target existence in the cavity area A is medium-medium, and the possibility attribute of the target existence in the cavity area B is medium-low; when ARfb is less than BRfb, the possibility attribute of the target existence in the cavity area B is medium-medium, and the possibility attribute of the target existence in the cavity area A is medium-low.

[0043] Furthermore, the third possibility judgment is specifically: when the ratio ARfb of the foreground area and the background area of ​​cavity area A is greater than the threshold thr, and the ratio BRfb of the foreground area and the background area of ​​cavity area B is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-high, and the possibility attribute of the target existence in cavity area B is high-medium; when the ratio BRfb of the foreground area and the background area of ​​cavity area B is greater than the threshold thr, and the ratio ARfb of the foreground area and the background area of ​​cavity area A is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-medium, and the possibility attribute of the target existence in cavity area B is high-high.

[0044] Furthermore, in step 1, the CT image sequence is preprocessed, specifically: first, the contrast and brightness of the image are adjusted using an adaptive threshold technology; then, the processed image is segmented using a graph cut method, which optimizes the segmentation boundary by considering the relationship between pixels and effectively distinguishes different structures in the chest cavity; then, a region growing algorithm is used to identify and expand the connected areas in the segmented image, ensuring continuity and integrity while merging adjacent approximate areas by setting a similarity threshold; finally, feature matching is performed on the merged areas, and the SIFT feature matching technology is used to compare the similarities between the areas, so as to locate and mark the chest area and the two cavity areas inside it, and finally the chest area and the two cavity areas within the chest area in the CT image are obtained.

[0045] Furthermore, in step 7, according to the target existence possibility attribute of the CT image, a deep learning model is trained using a training strategy based on the target existence possibility, specifically: for CT images of the chest area with different target existence possibilities, different learning rates are applied to model training, and a high learning rate is used for CT images with high target existence possibility; a low learning rate is used for CT images with low target existence possibility; a medium learning rate is used for CT images with medium target existence possibility; the target existence possibility attributes of cavity area A and cavity area B are specially labeled to enhance the model's attention and recognition ability to these labeled areas, and the cavity area with high-high target existence possibility is labeled as priority-highest; the cavity area with high-medium target existence possibility is labeled as priority-high; the cavity area with medium-medium target existence possibility is labeled as priority-medium; and the cavity area with medium-low target existence possibility is labeled as priority-low.

[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 target detection method based on target existence possibility, comprising the following steps: Step 1: preprocessing the CT image to obtain the thoracic region and two cavity regions in the thoracic region in the CT image; Step 2: Perform threshold segmentation on the chest area in the CT image to obtain the foreground area and background area in the two cavity areas in the CT image; Step 3: Calculate the number of pixels Af and Bf in the foreground area of ​​the two cavity areas in the chest area of ​​the CT image, namely, cavity area A and cavity area B; calculate the average value Ab_avg and Bb_avg of the number of pixels in the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image; Step 4: Calculate the ratios ARfb and BRfb of the foreground area and the background area of ​​cavity area A and cavity area B in the chest area of ​​the CT image respectively; Step 5: Calculate the similarity S of the foreground area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image; Step 6: Set the ratio threshold thr of the foreground area and the background area to determine the possible existence attribute of the CT image target; Step 7: According to the target existence possibility attribute of the CT image, the deep learning model is trained using a training strategy based on target existence possibility.

2. A target detection method based on target existence possibility as claimed in claim 1, characterized in that: In step 2, threshold segmentation is performed on the chest area in the CT image, specifically: all pixels in the chest area of ​​the CT image are segmented according to the segmentation threshold thdiv, and pixel points whose pixel values ​​are greater than the segmentation threshold thdiv are divided into foreground areas, and pixel points whose pixel values ​​are less than the segmentation threshold thdiv are divided into background areas.

3. A target detection method based on target existence possibility as claimed in claim 1, characterized in that: In step 4, the ratios ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B in the chest area of ​​the CT image are calculated respectively, specifically: ARfb=Af / Ab_avg, BRfb=Bf / Bb_avg.

4. A target detection method based on target existence possibility as claimed in claim 1, characterized in that: In step 5, the similarity S of the foreground area of ​​cavity area A and cavity area B in the chest area of ​​the CT image is calculated, specifically: S=abs[Af / (Af+Ab_avg)-Bf / (Bf+Bb_avg)], where abs[*] is the absolute value.

5. The target detection method based on target existence possibility as claimed in claim 1, characterized in that: In the step 6, a threshold value thr of the proportion of the foreground area and the background area is set to judge the possibility attribute of the existence of the CT image target, specifically: when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both greater than the proportion threshold thr, a possibility one judgment is performed; when the proportions ARfb and BRfb of the foreground area and the background area of ​​the cavity area A and the cavity area B are both less than the proportion threshold thr, a possibility two judgment is performed; when the proportion ARfb of the foreground area and the background area of ​​the cavity area A is greater than the threshold thr, and the proportion BRfb of the foreground area and the background area of ​​the cavity area B is less than the threshold thr, or when the proportion BRfb of the foreground area and the background area of ​​the cavity area B is greater than the threshold thr, and the proportion ARfb of the foreground area and the background area of ​​the cavity area A is less than the threshold thr, a possibility three judgment is performed.

6. A target detection method based on target existence possibility as claimed in claim 5, characterized in that: The possibility judgment is specifically as follows: setting a similarity threshold ths, when the similarity S is greater than the similarity threshold ths, judging that the target existence possibility attribute of the chest region of the CT image is medium, and the target existence possibility attributes of the cavity region A and the cavity region B are both medium-medium; When the similarity S is less than the similarity threshold ths, the target existence possibility attribute of the chest area of ​​the CT image is judged to be high, and then the proportions of the foreground area and the background area of ​​cavity area A and cavity area B, ARfb and BRfb, are compared. When ARfb is greater than BRfb, the target existence possibility attribute of cavity area A is high-high, and the target existence possibility attribute of cavity area B is high-medium; When ARfb is less than BRfb, the target existence possibility attribute in cavity area B is high-high, and the target existence possibility attribute in cavity area A is high-medium.

7. A target detection method based on target existence possibility as claimed in claim 5, characterized in that: The second possibility judgment is specifically as follows: setting a similarity threshold ths, when the similarity S is greater than the similarity threshold ths, judging that the target existence possibility attribute of the chest region of the CT image is low, and the target existence possibility attribute of the cavity region A and the cavity region B is low-low; When the similarity S is less than the similarity threshold ths, the target existence possibility attribute of the chest area of ​​the CT image is judged to be medium, and then the proportions of the foreground area and the background area of ​​cavity area A and cavity area B, ARfb and BRfb, are compared. When ARfb is greater than BRfb, the target existence possibility attribute of cavity area A is medium-medium, and the target existence possibility attribute of cavity area B is medium-low; when ARfb is less than BRfb, the target existence possibility attribute of cavity area B is medium-medium, and the target existence possibility attribute of cavity area A is medium-low.

8. A target detection method based on target existence possibility as claimed in claim 5, characterized in that: The third possibility judgment is specifically: when the ratio ARfb of the foreground area to the background area of ​​cavity area A is greater than the threshold thr, and the ratio BRfb of the foreground area to the background area of ​​cavity area B is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-high, and the possibility attribute of the target existence in cavity area B is high-medium; when the ratio BRfb of the foreground area to the background area of ​​cavity area B is greater than the threshold thr, and the ratio ARfb of the foreground area to the background area of ​​cavity area A is less than the threshold thr, then it is judged that the possibility attribute of the target existence in the chest area of ​​the CT image is high, and the possibility attribute of the target existence in cavity area A is high-medium, and the possibility attribute of the target existence in cavity area B is high-high.

9. A target detection method based on target existence possibility as claimed in claim 1, characterized in that: In the step 1, the CT image sequence is preprocessed, specifically: first, the contrast and brightness of the image are adjusted using an adaptive threshold technology; then, the processed image is segmented using a graph cut method, which optimizes the segmentation boundary by considering the relationship between pixels and effectively distinguishes different structures in the chest cavity; then, a region growing algorithm is used to identify and expand the connected areas in the segmented image, while ensuring continuity and integrity, and merging adjacent similar areas by setting a similarity threshold; finally, feature matching is performed on the merged areas, and the SIFT feature matching technology is used to compare the similarities between the areas, so as to locate and mark the chest area and the two cavity areas inside it, and finally the chest area and the two cavity areas within the chest area in the CT image are obtained.

10. The target detection method based on target existence possibility as claimed in claim 1, characterized in that: In step 7, according to the target existence possibility attribute of the CT image, a training strategy based on the target existence possibility is used to train the deep learning model, specifically: for CT images of the chest area with different target existence possibilities, different learning rates are applied to perform model training respectively, and a high learning rate is used for CT images with high target existence possibility; For CT images with low probability of target existence, a low learning rate is used; For CT images with medium probability of target existence, a medium learning rate is used; the target existence probability attributes of cavity area A and cavity area B are specially labeled to enhance the model's attention and recognition ability to these labeled areas; for cavity areas with high-high probability of target existence, the label is labeled as priority-highest; For cavity areas where the probability of the target existing is high-medium, the label is marked as priority-high; For the cavity area where the target existence possibility is medium-medium, the label is marked as priority-medium; For cavity areas where the probability of the target existing is medium-low, the label is marked as priority-low.

Images