Target detection method and device, equipment and storage medium

Abstract

The invention provides a target detection method and device, equipment and a storage medium, which are used for solving the problem of low target detection accuracy. The method at least comprises the following steps: based on biological tissue characteristics extracted from a to-be-detected image, carrying out region division on a region where abnormal tissues are located in the to-be-detected image to obtain at least one lesion reference region; based on the visual pathological features extracted from the to-be-detected image, predicting lesion positions in the to-be-detected image, and obtaining a plurality of candidate lesion positions; aiming at the at least one lesion reference area, respectively executing the following steps: selecting a target lesion position of which the position relationship meets a position association condition from the plurality of candidate lesion positions based on the position relationship between each of the plurality of candidate lesion positions and one lesion reference area; and obtaining a target detection result based on the obtained at least one target lesion position. And at least one target lesion position with the most suspected lesion and the most representative is screened out through the position relationship, so that the target detection accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a target detection method, apparatus, device, and storage medium. Background Technology

[0002] With the continuous development of technology, more and more client or server devices can perform target detection on medical images and obtain target detection results to assist doctors in medical diagnosis.

[0003] Taking cervical precancerous lesions as an example, the client can acquire colposcopy images and send them to the server. Upon receiving the colposcopy images, the server performs target detection to identify all possible lesion locations within the images, obtaining the target detection results. Based on these results, doctors can accurately diagnose the nature of cervical precancerous lesions and provide precise treatment plans.

[0004] In related technologies, target detection methods typically involve, after obtaining the image to be detected, using a trained liveness location prediction model to predict multiple liveness locations in the image, thus obtaining the target detection result. Based on the target detection result, a liveness test can then be performed at each liveness location to obtain the corresponding liveness result.

[0005] However, since many biopsy locations can be predicted from the image to be detected, and many of these biopsy locations may come from the same continuous lesion, when biopsies are performed on the biopsy locations obtained by the target detection method provided by the relevant technology, it is easy to repeatedly detect the same continuous lesion, resulting in the consumption of a lot of unnecessary computing resources and reducing biopsy efficiency. This also makes doctors need to combine many biopsy results to perform diagnostic analysis, resulting in a waste of medical resources.

[0006] Therefore, a more accurate target detection method is needed to overcome the above-mentioned shortcomings. Summary of the Invention

[0007] This application provides a target detection method, apparatus, device, and storage medium to address the problem of low detection accuracy in target detection.

[0008] Firstly, a target detection method is provided, including:

[0009] Based on the biological tissue features extracted from the image to be detected, the region where the abnormal tissue is located in the image to be detected is divided into regions to obtain at least one lesion reference region.

[0010] Based on the visual pathological features extracted from the image to be detected, the location of the lesion in the image to be detected is predicted, and multiple candidate lesion locations are obtained.

[0011] For the at least one lesion reference region, the following steps are performed: based on the positional relationship between each of the plurality of candidate lesion locations and a lesion reference region, a target lesion location whose positional relationship satisfies the positional association condition is selected from the plurality of candidate lesion locations;

[0012] Based on the location of at least one target lesion, the target detection result is obtained.

[0013] Secondly, a target detection device is provided, comprising:

[0014] First processing module: used to divide the region where the abnormal tissue is located in the image to be detected based on the biological tissue features extracted from the image to be detected, and obtain at least one lesion reference region;

[0015] The first processing module is further configured to: predict the location of lesions in the image to be detected based on the visual pathological features extracted from the image to be detected, and obtain multiple candidate lesion locations;

[0016] The second processing module is used to perform the following for the at least one lesion reference region: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, select the target lesion location whose positional relationship satisfies the positional association condition from the multiple candidate lesion locations;

[0017] The second processing module is also used to: obtain target detection results based on the obtained at least one target lesion location.

[0018] Optionally, the second processing module is specifically used for:

[0019] Based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, determine the number of candidate lesion locations located within the lesion reference region among the multiple candidate lesion locations;

[0020] When the number of locations is not greater than a preset threshold, the candidate lesion location located within the lesion reference area among the multiple candidate lesion locations is taken as the target lesion location.

[0021] Optionally, after predicting the location of the lesion in the image to be detected, each of the multiple candidate lesion locations is associated with a predicted lesion probability;

[0022] The second processing module is further used for:

[0023] After determining the number of candidate lesion locations within the lesion reference area among the plurality of candidate lesion locations, if the number of locations is greater than the preset number threshold, the lesion rejection probability corresponding to each of the plurality of candidate lesion locations is obtained; wherein, the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location;

[0024] From the candidate lesion locations located within the lesion reference area, the candidate lesion locations with a lesion rejection probability greater than the rejection probability threshold are selected as the target lesion locations.

[0025] Optionally, the second processing module is further configured to:

[0026] Before obtaining the lesion rejection probability corresponding to each of the multiple candidate lesion locations, the proportion of pixels belonging to the at least one lesion reference region among the pixels covered by each of the multiple candidate lesion locations is calculated to obtain the corresponding region location proportion.

[0027] The predicted lesion probability and the regional location ratio corresponding to each of the multiple candidate lesion locations are combined to obtain the corresponding lesion rejection probability.

[0028] Optionally, the second processing module is specifically used for:

[0029] The number of pixels covered by each of the multiple candidate lesion locations is counted to obtain the corresponding number of pixels;

[0030] The number of pixels belonging to at least one lesion reference region is counted in each of the pixels covered by the multiple candidate lesion locations to obtain the corresponding number of region points.

[0031] For each of the multiple candidate lesion locations, the following steps are performed: determine the corresponding region location percentage based on the ratio between the number of region points and the number of pixels corresponding to a candidate lesion location.

[0032] Optionally, when the at least one lesion reference region is multiple lesion reference regions, the second processing module is further configured to:

[0033] Before selecting a target lesion location whose positional relationship satisfies the positional association condition from the plurality of candidate lesion locations based on the positional relationship between each of the plurality of candidate lesion locations and a lesion reference region, the reference location in the image to be detected is identified based on the biological part features extracted from the image to be detected, and the location of the reference location is obtained.

[0034] A reference coordinate system is established with the location of the reference part as the origin;

[0035] Based on the reference coordinate system, the multiple lesion reference regions, and the multiple candidate lesion locations, location association conditions are set.

[0036] Optionally, the second processing module is further configured to:

[0037] For each of the multiple lesion reference regions, the following steps are performed: determining the region inclusion relationship between each of the multiple candidate lesion locations and a lesion reference region in the image to be detected;

[0038] For each quadrant region divided by the reference coordinate system, the following steps are performed: determining the quadrant inclusion relationship between each of the multiple candidate lesion locations and a quadrant region, and the overlap relationship between each of the multiple lesion reference regions and the quadrant region;

[0039] For each of the multiple lesion reference regions, the following steps are performed: based on the obtained inclusion relationships of each region, inclusion relationships of each quadrant, and overlap relationships, the positional relationship between the multiple candidate lesion locations and a lesion reference region is determined.

[0040] Optionally, after predicting the location of the lesion in the image to be detected, each of the multiple candidate lesion locations is associated with a predicted lesion probability;

[0041] The second processing module is specifically used for:

[0042] Obtain the lesion rejection probability corresponding to each of the multiple candidate lesion locations; wherein, the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location;

[0043] For the multiple lesion reference regions, the following steps are performed: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region and the obtained lesion rejection probabilities, a reference lesion location with a rejection probability greater than a rejection probability threshold and located within the lesion reference region is selected from the multiple candidate lesion locations.

[0044] Based on the quadrant region where each reference lesion location is located and the lesion reference region to which it belongs, the corresponding target lesion location is selected from the multiple candidate lesion locations for each of the multiple lesion reference regions.

[0045] Optionally, the second processing module is specifically used for:

[0046] Based on the quadrant regions of each obtained reference lesion location, at least one group of reference lesion locations is determined from the reference lesion locations; wherein, the quadrant regions of the multiple reference lesion locations included in each group of reference lesion locations are matched, and the lesion reference regions to which they belong are different; the quadrant regions are obtained based on the reference coordinate system.

[0047] For each of the at least one set of reference lesion locations, the following is performed: the reference lesion location with the highest probability of lesion selection among the multiple reference lesion locations contained in a set of reference lesion locations is taken as the target lesion location;

[0048] From the plurality of candidate lesion locations, at least one lesion reference region other than the lesion reference region to which each target lesion location belongs is selected from the plurality of lesion reference regions.

[0049] Optionally, the second processing module is specifically used for:

[0050] Among the plurality of reference lesion locations, at least one reference lesion location is identified, excluding the reference lesion location with the highest probability of lesion rejection.

[0051] For the at least one reference lesion location, the following steps are performed: from the plurality of candidate lesion locations, among each candidate lesion location located within the lesion reference region to which a reference lesion location belongs, select a secondary reference location whose lesion rejection probability is within the rejection probability range; wherein, the maximum value in the rejection probability range is less than the rejection probability threshold.

[0052] Based on the quadrant regions of each of the at least one secondary reference positions, if it is determined that there is no secondary reference position group among the at least one secondary reference positions, the at least one secondary reference position is taken as the corresponding target lesion location; wherein, the quadrant regions of the multiple secondary reference positions included in the secondary reference position group are matched, and the lesion reference regions to which they belong are different.

[0053] Optionally, the second processing module is further configured to:

[0054] For the at least one set of reference lesion locations, the following steps are performed: after selecting the reference lesion location with the highest lesion rejection probability among the multiple reference lesion locations included in the set of reference lesion locations as the target lesion location, the reference lesion locations other than the at least one set of reference lesion locations among the reference lesion locations are selected as the target lesion locations.

[0055] Optionally, the first processing module is specifically used for:

[0056] Images of the cervical tissue surface after treatment with acetic acid solution were acquired.

[0057] Tissue features are extracted from the image to be detected to obtain biological tissue features;

[0058] Based on the obtained biological tissue features, the regions where the biological tissues with abnormal colors are located in the image to be detected are divided into regions to obtain at least one lesion reference region.

[0059] Optionally, the first processing module is specifically used for:

[0060] Cervical features are extracted from the image to be detected to obtain biological site features;

[0061] Based on the obtained biological features, the cervix in the image to be detected is identified, and the location of the reference site is obtained.

[0062] Thirdly, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0063] Fourthly, a computer device is provided, comprising:

[0064] Memory, used to store program instructions;

[0065] A processor is configured to invoke program instructions stored in the memory and execute the method described in the first aspect according to the obtained program instructions.

[0066] Fifthly, a computer-readable storage medium is provided, the computer-readable storage medium storing computer-executable instructions for causing a computer to perform the method as described in the first aspect.

[0067] In this embodiment, the region containing abnormal tissue in the image to be detected is divided into regions to obtain at least one lesion reference region. Furthermore, the location of the lesion in the image to be detected is identified to obtain multiple candidate lesion locations. Using the positional relationship between the multiple candidate lesion locations and the at least one lesion reference region as a reference, the target lesion location whose positional relationship satisfies the positional association condition is selected as the target detection result, instead of using multiple candidate lesion locations as the target detection result. This ensures that the target lesion location included in the target detection result is the most representative lesion location, avoiding subsequent repeated biopsies of the same continuous lesions and improving the detection accuracy during target detection.

[0068] Furthermore, for each lesion reference region, target lesion locations that meet the location association conditions are selected, so that each lesion reference region can be detected by the subsequent lesion detection process. This avoids the situation where the most suspected lesion location is missed due to the small number of selected target lesion locations, which could lead to misdiagnosis. This can improve the detection accuracy of target lesion detection while ensuring the accuracy of lesion detection.

[0069] Furthermore, through an automated candidate lesion location screening process, targeted and comprehensive target detection results can be obtained. The highly accurate target detection process simplifies the lesion location screening steps in subsequent lesion detection processes, avoiding unnecessary waste of medical resources. Attached Figure Description

[0070] Figure 1A This is a schematic diagram illustrating an application field of the target detection method provided in the embodiments of this application;

[0071] Figure 1B This is one application scenario of the target detection method provided in the embodiments of this application;

[0072] Figure 2 This is a schematic flowchart of a target detection method provided in an embodiment of this application;

[0073] Figure 3A A schematic diagram illustrating the principle of the target detection method provided in this application embodiment;

[0074] Figure 3B A schematic diagram of the principle of the target detection method provided in the embodiments of this application. Figure 2 ;

[0075] Figure 4A A schematic diagram three illustrating the principle of the target detection method provided in this application embodiment;

[0076] Figure 4B A schematic diagram four illustrating the principle of the target detection method provided in this application embodiment;

[0077] Figure 4C A schematic diagram five illustrating the principle of the target detection method provided in the embodiments of this application;

[0078] Figure 4D A schematic diagram six illustrating the principle of the target detection method provided in the embodiments of this application;

[0079] Figure 4E A schematic diagram of the principle of the target detection method provided in the embodiments of this application. Figure 7 ;

[0080] Figure 5AA schematic diagram of the principle of the target detection method provided in the embodiments of this application. Figure 8 ;

[0081] Figure 5B A schematic diagram of the principle of the target detection method provided in the embodiments of this application. Figure 9 ;

[0082] Figure 6A A schematic diagram of the principle of the target detection method provided in the embodiments of this application is shown in Figure 10.

[0083] Figure 6B 11. A schematic diagram of the principle of the target detection method provided in the embodiments of this application;

[0084] Figure 7 A schematic diagram twelve illustrating the principle of the target detection method provided in this application embodiment;

[0085] Figure 8 A schematic diagram of the target detection device provided in the embodiments of this application is shown below;

[0086] Figure 9 A schematic diagram of the target detection device provided in the embodiments of this application. Figure 2 . Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0088] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0089] (1) Colposcopy image:

[0090] A colposcope is a gynecological clinical diagnostic instrument, one type of gynecological endoscope. It is suitable for diagnosing various cervical diseases and genital lesions, and is also an important method for the early diagnosis of male and female sexual diseases.

[0091] (2) Cervical biopsy:

[0092] A cervical biopsy is a biopsy of the cervix, in which a small piece or several pieces of tissue are taken from the cervix for pathological examination to confirm the diagnosis.

[0093] This application relates to Artificial Intelligence (AI) technology and cloud computing. AI technology and cloud computing can be applied to many fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, smart transportation, smart maps, driver assistance systems, vehicle terminals, aircraft, digital twins, virtual humans, robots, AI-generated content (AIGC), conversational interaction, smart healthcare, smart customer service, game AI, etc. It is believed that with the development of technology, AI technology and cloud computing will be applied in more fields and play an increasingly important role.

[0094] It should be noted that the embodiments of this application involve operations such as acquiring data such as images to be detected. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0095] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0096] The application fields of the target detection method provided in the embodiments of this application will be briefly introduced below.

[0097] With the continuous development of technology, more and more client or server devices can perform target detection on medical images and obtain target detection results to assist doctors in medical diagnosis.

[0098] Taking cervical precancerous lesions as an example, the client can acquire colposcopy images and send them to the server. Upon receiving the colposcopy images, the server performs target detection to identify all possible lesion locations within the images, obtaining the target detection results. Based on these results, doctors can accurately diagnose the nature of cervical precancerous lesions, such as cervical intraepithelial neoplasia, and provide accurate treatment plans.

[0099] Taking oral precancerous lesions as an example, the client can collect oral images and perform target detection on the oral images in real time to determine all possible lesion locations in the oral images, so as to help determine whether there are oral precancerous lesions, such as oral leukoplakia, oral erythema and other oral precancerous lesions.

[0100] Taking precancerous lesions of colorectal cancer as an example, the client can collect colonoscopy images and send them to the server. After receiving the colonoscopy images, the server performs target detection on the colonoscopy images to determine all possible lesion locations in the colonoscopy images, in order to help determine whether there are precancerous lesions of colorectal cancer, such as adenomatous polyps.

[0101] Taking precancerous lesions of the stomach as an example, the client can collect gastroscopy images and send them to the server. After receiving the gastroscopy images, the server performs target detection on the gastroscopy images to determine all possible lesion locations in the gastroscopy images, in order to help determine whether there are precancerous lesions of the stomach, such as gastric mucosal atrophy, intestinal metaplasia and other precancerous lesions of the stomach.

[0102] In related technologies, target detection methods typically involve, after obtaining the image to be detected, using a trained liveness location prediction model to predict multiple liveness locations in the image, thus obtaining the target detection result. Based on the target detection result, a liveness test can then be performed at each liveness location to obtain the corresponding liveness result.

[0103] Please refer to Figure 1A Taking cervical precancerous lesions as an example, after acquiring colposcopy images, a trained biopsy location prediction model can be used to predict multiple cervical biopsy locations contained in the colposcopy images. Figure 1A The images are shown within small rectangles. This allows for the separate biopsy of each obtained cervical biopsy site, yielding the corresponding biopsy results for each site.

[0104] However, since many biopsy locations can be predicted from the image to be detected, and many of these biopsy locations may come from the same continuous lesion, when biopsies are performed on the biopsy locations obtained by the target detection method provided by the relevant technology, it is easy to repeatedly detect the same continuous lesion, resulting in the consumption of a lot of unnecessary computing resources and reducing biopsy efficiency. This also makes doctors need to combine many biopsy results to perform diagnostic analysis, resulting in a waste of medical resources.

[0105] Therefore, a more accurate target detection method is needed to overcome the above-mentioned shortcomings.

[0106] To address the issue of low detection accuracy in target detection, this application proposes a target detection method. In this method, based on biological tissue features extracted from the image to be detected, the region containing abnormal tissue in the image is divided to obtain at least one lesion reference region. Based on visual pathological features extracted from the image to be detected, the location of the lesion in the image is predicted, obtaining multiple candidate lesion locations. For each of the at least one lesion reference region, the following steps are performed: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, a target lesion location whose positional relationship satisfies a positional association condition is selected from the multiple candidate lesion locations. Based on the obtained at least one target lesion location, a target detection result is obtained.

[0107] In this embodiment, the region containing abnormal tissue in the image to be detected is divided into regions to obtain at least one lesion reference region. Furthermore, the location of the lesion in the image to be detected is identified to obtain multiple candidate lesion locations. Using the positional relationship between the multiple candidate lesion locations and the at least one lesion reference region as a reference, the target lesion location whose positional relationship satisfies the positional association condition is selected as the target detection result, instead of using multiple candidate lesion locations as the target detection result. This ensures that the target lesion location included in the target detection result is the most representative lesion location, avoiding subsequent repeated biopsies of the same continuous lesions and improving the detection accuracy during target detection.

[0108] Furthermore, for each lesion reference region, target lesion locations that meet the location association conditions are selected, so that each lesion reference region can be detected by the subsequent lesion detection process. This avoids the situation where the most suspected lesion location is missed due to the small number of selected target lesion locations, which could lead to misdiagnosis. This can improve the detection accuracy of target lesion detection while ensuring the accuracy of lesion detection.

[0109] Furthermore, through an automated candidate lesion location screening process, targeted and comprehensive target detection results can be obtained. The highly accurate target detection process simplifies the lesion location screening steps in subsequent lesion detection processes, avoiding unnecessary waste of medical resources.

[0110] The application scenarios of the target detection method provided in this application are described below.

[0111] Please refer to Figure 1BThis is a schematic diagram illustrating an application scenario of the target detection method provided in this application. The application scenario includes a server 101 and a client 102, which can communicate with each other. The communication method can be wired, such as through a network cable or serial cable; or wireless, such as through Bluetooth or Wi-Fi. No specific limitation is imposed.

[0112] Client 102 generally refers to devices that can acquire images to be detected, such as terminal devices, third-party applications accessible by the terminal devices, or web pages accessible by the terminal devices. Server 101 generally refers to devices that can perform target detection on images to be detected, such as terminal devices or servers.

[0113] Terminal devices include, but are not limited to, mobile phones, computers, smart medical devices, smart home appliances, vehicle terminals, or aircraft. Servers include, but are not limited to, cloud servers, local servers, or associated third-party servers. Both client 101 and server 102 can use cloud computing to reduce the use of local computing resources; similarly, they can also use cloud storage to reduce the use of local storage resources.

[0114] As one embodiment, the server 101 and the client 102 can be the same device, or they can be different devices, or they can be different devices with some modules shared, etc., and there are no specific restrictions.

[0115] Taking cervical precancerous lesions as an example, client 102 responds to the imaging operation by acquiring colposcopy images. Client 102 sends a cervical lesion detection request carrying the acquired colposcopy images to server 101. After receiving the cervical lesion detection request from client 102, server 101 performs target detection based on the colposcopy images and obtains the target detection results. Server 101 returns the target detection results to client 102, and client 102 receives and presents the target detection results returned by server 101.

[0116] Client 102 can also perform target detection in real time based on colposcopy images after acquiring colposcopy images in response to the shooting operation, and obtain and present the target detection results.

[0117] The following is based on Figure 1B This document provides a detailed description of the target detection method provided in the embodiments of this application. Please refer to [link / reference]. Figure 2 This is a flowchart illustrating a target detection method provided in an embodiment of this application.

[0118] S201, Based on the biological tissue features extracted from the image to be detected, the region where the abnormal tissue is located in the image to be detected is divided into regions to obtain at least one lesion reference region.

[0119] The image to be detected can be an image taken of the part of an organism to be detected, or it can be a direct image taken of the part of an organism to be detected, or it can be an image taken after medical processing of the part of an organism to be detected, etc., there are no specific restrictions. The image to be detected can be the original image or an image after image processing, etc., there are no specific restrictions.

[0120] Tissue features are extracted from the image to be detected to obtain biological tissue features. These features describe the appearance characteristics of the biological tissue in the image, such as texture, color, tissue edges, spots, and smoothness. Based on the obtained biological tissue features, the areas containing abnormal tissue in the image are divided into regions to obtain at least one lesion reference region. When abnormal tissue is found among the biological tissues described by the biological tissue features, that tissue is identified as abnormal, and the area containing the abnormal tissue is designated as the lesion reference region.

[0121] For example, in the context of precancerous cervical lesions, when acetic acid solution is applied to the surface of cervical tissue, the acetic acid causes the proteins in normal cells to coagulate, making the cervical tissue transparent. Abnormal cells, on the other hand, will appear white or pale blue. Based on the medical characteristics of biological tissues, the areas containing abnormal tissue in the image to be detected can be segmented, which can then be used as a reference for selecting the location of target lesions, thus helping to improve the accuracy of target detection.

[0122] Therefore, please refer to Figure 3A This method can acquire images of cervical tissue after treatment with acetic acid solution. Tissue feature extraction is then performed on these images to obtain biological tissue characteristics. These biological tissue characteristics can describe features such as the transparency of the cervical tissue.

[0123] Therefore, based on the obtained biological tissue features, the regions containing abnormally colored biological tissue in the image to be detected are divided into areas to obtain at least one lesion reference region. Figure 3A Three lesion reference areas are shown, highlighted against a white background. The transparency of the abnormally colored biological tissue is lower than that of the other biological tissues; that is, the biological tissue within at least one lesion reference area is less transparent than the biological tissue within other areas.

[0124] As one embodiment, based on the biological tissue features extracted from the image to be detected, the region containing the abnormal tissue in the image is divided into regions to obtain at least one lesion reference region. This can be achieved using a trained region segmentation model, such as the U-Net model, etc., without any specific limitations. The trained region segmentation model can be trained on a large number of sample images with labeled lesion reference regions. The lesion reference region and the sample image will also be different depending on the target detection scenario. For example, in the target detection scenario in the field of oral precancerous lesions, the sample image can be an oral cavity image, and the lesion reference region can be a region containing oral leukoplakia or oral erythema, etc.

[0125] S202, based on the visual pathological features extracted from the image to be detected, predict the location of the lesion in the image to be detected and obtain multiple candidate lesion locations.

[0126] Lesions in biological tissues exhibit specific pathological features, such as particular tissue color, shape, texture, and size. Therefore, extracting pathological features from an image to be tested yields visual pathological features, which describe the location of pathological abnormalities within the image. Based on these visual pathological features, all possible lesion locations in the image are predicted, resulting in multiple candidate lesion locations.

[0127] Please refer to Figure 3B In the context of precancerous cervical lesions, the lesions may appear as small, well-defined white or pale white areas. Therefore, pathological feature extraction can be performed on the image to be detected to obtain visual pathological features. Based on these visual pathological features, multiple candidate lesion locations in the image can be predicted.

[0128] As one embodiment, the process of predicting the location of lesions in the image to be detected based on the visual pathological features extracted from the image to be detected and obtaining multiple candidate lesion locations can be implemented using a trained target detection model (You Only Look Once, YOLO) or YOLO V3, etc., without any specific limitation.

[0129] The trained object detection model can be trained based on a large number of sample images with labeled candidate lesion locations. The candidate lesion locations and sample images will vary depending on the lesion detection scenario. For example, in the target detection scenario in the field of precancerous skin lesions, the sample image can be a skin image, and the candidate lesion location can be a location of actinic keratosis, a location containing erythema or scaling, etc.

[0130] S203, for at least one lesion reference region, respectively perform: based on the positional relationship between multiple candidate lesion locations and a lesion reference region, select the target lesion location whose positional relationship satisfies the positional association condition from multiple candidate lesion locations.

[0131] After obtaining multiple candidate lesion locations contained in the image to be detected, and at least one lesion reference region delineated in the image, the following steps can be performed for each of the at least one lesion reference regions: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, the target lesion location whose positional relationship satisfies the positional association condition can be selected from the multiple candidate lesion locations. This allows for the selection of a target lesion location for each lesion reference region, avoiding the existence of lesion reference regions that do not contain the target lesion location, and improving the accuracy of target detection.

[0132] The number of target lesion locations selected for each lesion reference region can be one or more. The number of target lesion locations selected can be adjusted by adjusting the location association conditions, and there are no specific restrictions.

[0133] For example, the more location association conditions are included, the fewer target lesion locations will be selected; the fewer location association conditions are included, the more target lesion locations will be selected. Similarly, the higher the difficulty in satisfying location association conditions, the fewer target lesion locations will be selected; the lower the difficulty in satisfying location association conditions, the more target lesion locations will be selected, and so on.

[0134] Please refer to Figure 4A Positional relationships include, for example, inclusion relationships, partial overlap relationships, and distance relationships between nearest neighbors, etc., without specific limitations. The area with a white background represents the lesion reference area, and the area with a diagonal striped background represents the candidate lesion location.

[0135] The location association condition indicates that there is a relationship between the target lesion location and the lesion reference area. For example, the lesion reference area contains the target lesion location, or the overlap between the lesion reference area and a location area of ​​a preset size centered on the target lesion location accounts for a proportion of the location area that is greater than a proportion threshold, or the shortest distance between the lesion reference area and the target lesion location is less than a distance threshold, etc. There are no specific restrictions.

[0136] By selecting target lesion locations that meet the location association conditions through location relationships, associated target lesion locations can be selected for each lesion reference region. This allows each lesion reference region to be subsequently detected through its associated target detection location, which helps improve the accuracy of target detection.

[0137] As an example, taking the process of selecting a target lesion location for a lesion reference area as an example, when there are fewer candidate lesion locations within the lesion reference area among multiple candidate lesion locations, the candidate lesion locations within the lesion reference area can be used as the target lesion location.

[0138] Since the lesion reference region and candidate lesion locations represent the lesion condition in the image to be detected in different ways, candidate lesion locations located within the lesion reference region can more accurately reflect the lesion condition within that region compared to other candidate lesion locations. Therefore, based on the positional relationship between multiple candidate lesion locations and a lesion reference region, the number of candidate lesion locations located within a single lesion reference region can be determined. When the number of locations does not exceed a preset threshold, the candidate lesion locations located within a single lesion reference region are taken as the target lesion location.

[0139] In this scenario, without using excessive redundant calculations, a smaller number of target lesion locations can be selected for each lesion reference region. This ensures efficient target detection while maintaining the accuracy of lesion detection in each lesion reference region.

[0140] Please refer to Figure 4B The image to be detected contains two lesion reference regions, namely the first lesion reference region and the second lesion reference region. The image to be detected contains four candidate lesion locations, namely the first candidate lesion location (denoted as A in the figure), the second candidate lesion location (denoted as B in the figure), the third candidate lesion location (denoted as C in the figure), and the fourth candidate lesion location (denoted as D in the figure).

[0141] The positional relationships between the four candidate lesion locations and the first lesion reference region are as follows: containment relationship, partial overlap relationship, and distance relationships 2 and 3 between the nearest edges. The positional relationships between the second lesion reference regions are as follows: distance relationships 3 and 1 between the nearest edges, containment relationship, and distance relationship 2 between the nearest edges.

[0142] The number of candidate lesion locations within each lesion reference area, i.e., the number of candidate lesion locations corresponding to the inclusion relationship, is 1. Taking a preset number threshold of 1 as an example, the candidate lesion locations corresponding to the inclusion relationship can be used as the target lesion locations of the corresponding lesion reference areas.

[0143] As one example, after predicting the location of a lesion in the image to be detected, multiple candidate lesion locations are each associated with a predicted lesion probability. The predicted lesion probability represents the probability that a lesion exists at the corresponding candidate lesion location. The higher the predicted lesion probability, the more likely a lesion is to exist at the corresponding candidate lesion location. The lower the predicted lesion probability, the less likely a lesion is to exist at the corresponding candidate lesion location.

[0144] Therefore, when selecting the target lesion location for each lesion reference region, selection can also be based on the predicted lesion probability. For example, the candidate lesion location located within the lesion reference region with the highest predicted lesion probability can be selected as the target lesion location corresponding to that lesion reference region. Alternatively, a location score can be calculated based on the positional relationship between the lesion reference region and the candidate lesion locations. For instance, the score for a location with an inclusion relationship is higher than that for a location with a partial overlap relationship, and the score for a location with a partial overlap relationship is higher than that for a location with a distance relationship between the nearest edges. Based on the weighted sum of the location score and the predicted lesion probability, the candidate lesion location with the largest weighted sum is selected as the target lesion location.

[0145] For example, after determining the number of candidate lesion locations within a lesion reference region from multiple candidate lesion locations, if the number of locations exceeds a preset threshold, the lesion rejection probability corresponding to each candidate lesion location can be obtained. The rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location. From the candidate lesion locations within a lesion reference region, the candidate lesion locations with rejection probabilities greater than a rejection probability threshold are selected as the target lesion locations.

[0146] The probability of lesion selection or rejection can be the predicted probability of the lesion itself, or it can be calculated based on the predicted probability of the lesion, etc., without any specific restrictions. The threshold for selection or rejection probability can be determined based on the selection or rejection probability of each lesion and the expected number of target lesion locations selected for each lesion reference region. For example, if the selection or rejection probabilities of 5 lesions are 0.4, 0.7, 0.2, 0.5, and 0.8, and the expected number is 1, then the threshold for selection or rejection probability can be set to 0.7; if the expected number is 2, then the threshold for selection or rejection probability can be set to 0.6, and so on.

[0147] When there are many candidate lesion locations within a lesion reference area, the lesion rejection probability, which is positively correlated with the predicted lesion probability, can be combined to select the most suspected lesion location, thereby improving the accuracy of target lesion location selection and thus improving the accuracy of target detection.

[0148] Please refer to Figure 4CIn a lesion reference region within the image to be detected, there are four candidate lesion locations, denoted as A, B, C, and D. The lesion rejection probabilities for these four candidate lesion locations are 0.7, 0.2, 0.5, and 0.8, respectively, with a rejection probability threshold of 0.7. Therefore, the candidate lesion location represented by D is taken as the target lesion location corresponding to the lesion reference region.

[0149] The following example illustrates a method for determining the probability of lesion selection based on the predicted lesion probability. Other methods include using the area of ​​the lesion reference region as the weight and multiplying the predicted lesion probability by the weight as the lesion selection probability; or using the number of pixels involved in the candidate lesion location within the lesion reference region as the weight and multiplying the predicted lesion probability by the weight as the lesion selection probability, etc. These will not be listed here.

[0150] Before obtaining the lesion rejection probability for each of the multiple candidate lesion locations, the proportion of pixels belonging to at least one lesion reference region among the pixels covered by each of the multiple candidate lesion locations is calculated to obtain the corresponding region location proportion. The predicted lesion probability and region location proportion corresponding to each of the multiple candidate lesion locations are then fused to obtain the corresponding lesion rejection probability.

[0151] By identifying the proportion of pixels belonging to at least one lesion reference region among the pixels covered by the candidate lesion location, we can reflect how many pixels in the candidate lesion location were again identified as lesions during lesion reference region segmentation. This proportion can then be used as the corresponding region location proportion to measure the lesion rejection probability. This allows the lesion rejection probability to reflect the probability that the candidate lesion location is doubly identified as a lesion location. Therefore, to a certain extent, this can improve the accuracy of selecting the most suspected lesion location based on the lesion rejection probability, thereby improving the accuracy of target detection.

[0152] Please refer to Figure 4D (1) The image to be detected contains three lesion reference regions, shown in irregular shapes against a white background, and ten candidate lesion regions, shown as rectangular boxes against a diagonal background. Based on the overlap between the candidate lesion regions and the lesion reference regions, the respective regional proportions of these ten candidate lesion regions are 50%, 100%, 80%, 10%, 45%, 100%, 80%, 95%, 60%, and 90%.

[0153] Please refer to Figure 4D (2) The predicted lesion probabilities corresponding to these 10 candidate lesion regions are 95%, 72%, 96%, 85%, 88%, 71%, 82%, 88%, 97% and 82%, respectively.

[0154] Taking the average as an example, by combining the predicted lesion probability with the regional location proportion, the lesion rejection probabilities corresponding to these 10 candidate lesion regions are 72.5%, 86%, 88%, 47.5%, 66.5%, 85.5%, 81%, 91.5%, 78.5%, and 86%, respectively.

[0155] Please refer to Figure 4E ,based on Figure 4D Taking weighted summation as an example, please refer to formula (1). The predicted lesion probability is denoted as Pred, and the regional location proportion is denoted as Rate. aw .

[0156] U = α·Rate aw +β·Pred

[0157] Taking a case where both α and β are 1, the probability of lesion prediction and the proportion of regional location are fused. The lesion rejection probabilities corresponding to these 10 candidate lesion regions are 145%, 172%, 176%, 95%, 133%, 171%, 162%, 183%, 157%, and 172%, respectively.

[0158] As one embodiment, when determining the proportion of a region, the number of pixels covered by each of the multiple candidate lesion locations can be counted separately to obtain the corresponding number of pixels. The number of pixels belonging to at least one lesion reference region among the pixels covered by each of the multiple candidate lesion locations can be counted separately to obtain the corresponding number of region points. For multiple candidate lesion locations, the following is performed: based on the ratio between the number of region points and the number of pixels corresponding to a candidate lesion location, the corresponding proportion of the region is determined.

[0159] Please refer to formula (2) for the region location rate corresponding to a candidate lesion location. aw The calculation process involves the number of pixels covered by the candidate lesion location, denoted as Area. biop The number of pixels belonging to at least one lesion reference region is denoted as Piexl. aw .

[0160] Rate aw =Piexl aw / Area biop (2)

[0161] The number of region points can be the total number of pixels belonging to all lesion reference regions, or the maximum number of pixels belonging to each lesion reference region, etc. There is no specific restriction.

[0162] The ratio between the number of region points and the number of pixels can be used to intuitively reflect the proportion of a region's location through a simple calculation process. Alternatively, the proportion of a region's location can be determined by first calculating the difference between the number of pixels and the number of region points, and then by calculating the ratio of the number of region points to the difference. Or, the proportion of a region's location can be determined by calculating the sum between the number of pixels and the number of region points, and then by calculating the ratio of the number of region points to the sum. There are no specific restrictions.

[0163] As one example, in order to further improve the accuracy of selecting the target lesion location and thus improve the accuracy of target detection, clinical guidelines can also be considered.

[0164] Based on biological features extracted from the image to be detected, reference sites are identified within the image, and their locations are obtained. In practical applications, the reference sites differ depending on the precancerous lesion scenario. A reference site is a location that is helpful in analyzing precancerous lesions. For example, in a cervical precancerous lesion scenario, the cervix could be a reference site; in an oral precancerous lesion scenario, teeth and other structures could be reference sites.

[0165] When the reference part appears as a region in the image to be detected, the location of the reference part can be any point in the region, the center point of the region, a point in the region with a pixel value greater than a certain threshold, or a point in the region with a texture degree higher than a certain value, etc. There are no specific restrictions.

[0166] By establishing a reference coordinate system with the reference site location as the origin, location association conditions can be set based on the reference coordinate system, multiple lesion reference regions, and multiple candidate lesion locations. The reference coordinate system can be a two-dimensional or three-dimensional Cartesian coordinate system, or any other coordinate system; there are no specific restrictions.

[0167] By establishing a reference coordinate system based on a reference site and setting positional association conditions based on the reference coordinate system, the selection of the target lesion location can fully consider the relationship between the reference site and the lesion location. This improves the accuracy of target detection and enhances the comprehensiveness and precision of subsequent lesion detection based on the target lesion location.

[0168] As one embodiment, when setting position association conditions based on a reference coordinate system, multiple lesion reference regions, and multiple candidate lesion locations, the positional relationship between each of the multiple candidate lesion locations and a lesion reference region can also be determined using the following method:

[0169] For multiple lesion reference regions, the following steps are performed: determine the regional inclusion relationship between each of the multiple candidate lesion locations in the image to be detected and a lesion reference region, that is, whether each of the multiple candidate lesion locations is located within the lesion reference region. If the candidate lesion location is represented by a very small region, the regional inclusion relationship can also indicate the overlapping area between each of the multiple candidate lesion locations and the lesion reference region.

[0170] For each quadrant region defined by the reference coordinate system, the following steps are performed: determining the quadrant inclusion relationship between each of the multiple candidate lesion locations and a quadrant region, and determining the overlap relationship between each of the multiple lesion reference regions and a quadrant region. That is, the reference coordinate system can, for example, divide the image to be detected into four quadrant regions. This allows determining which quadrant region each candidate lesion location falls within. If a candidate lesion location is represented by a very small area, the quadrant inclusion relationship can also indicate the overlap area between each candidate lesion location and that quadrant region. Furthermore, it can determine which quadrant region each lesion reference region overlaps with.

[0171] For multiple lesion reference regions, the following steps are performed: based on the obtained inclusion relationships of each region, inclusion relationships of each quadrant, and overlap relationships, the positional relationship between multiple candidate lesion locations and a lesion reference region is determined.

[0172] By considering the location relationships of multiple candidate lesion locations, including which quadrant region each lesion reference region is located in, and which quadrant region each lesion reference region overlaps with, the target lesion location can be selected for each quadrant region divided by the reference coordinate system with the reference site as the origin. This avoids selecting the target lesion location only for a certain range in the image to be detected, thus missing other potential lesion locations. The system automatically performs target detection according to clinical guidelines, improving the accuracy of target detection.

[0173] Please refer to Figure 5A The reference coordinate system divides the image to be detected into four quadrants: quadrant I, quadrant II, quadrant III, and quadrant IV. The image contains three lesion reference regions and ten candidate lesion locations. Region inclusion relationships include the inclusion relationships and partial overlap relationships described above.

[0174] Quadrant inclusion relationships include, for example, the first quadrant containing 2 candidate lesion locations, the second quadrant containing 4 candidate lesion locations, the third quadrant containing 4 candidate lesion locations, and the fourth quadrant containing 1 candidate lesion location. Overlap relationships include, for example, the first quadrant overlapping with the upper right lesion reference area, the second quadrant overlapping with both upper lesion reference areas, the third quadrant overlapping with the lower lesion reference area, and the fourth quadrant overlapping with the lower lesion reference area.

[0175] For example, the positional relationship is such that the candidate lesion is located in the first quadrant and partially overlaps with the lesion reference area on the upper right side, etc., which will not be listed one by one here.

[0176] As one example, in the scenario of precancerous cervical lesions, the reference site can be the cervical os. Therefore, when obtaining the location of the reference site, cervical os features can be extracted from the image to be detected to obtain biological site features. Based on the obtained biological site features, the cervical os in the image to be detected is identified, and the location of the reference site is obtained.

[0177] Please refer to Figure 5B In the context of precancerous cervical lesions, the cervix can appear as a round or oval opening, pink or red, etc. Therefore, cervical os features can be extracted from the image to be tested, yielding biological characteristics. Based on these biological characteristics, the cervix in the image can be identified. After determining the location of the cervix in the image, a reference coordinate system can be established with the center point of the cervix as the origin, the horizontal axes as the x-axis, and the vertical axes as the y-axis.

[0178] As one embodiment, if after predicting the location of a lesion in the image to be detected, multiple candidate lesion locations are each associated with a predicted lesion probability, and the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location (see the previous description for details), then, considering quadrant regions, the process of selecting a target lesion location for each lesion reference region can be as follows: First, obtain the lesion rejection probabilities corresponding to each of the multiple candidate lesion locations. For multiple lesion reference regions, respectively, perform the following: Based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region and the obtained lesion rejection probabilities, select reference lesion locations from the multiple candidate lesion locations whose lesion rejection probabilities are greater than a rejection probability threshold and are located within a lesion reference region. Based on the quadrant region where each reference lesion location is located and its respective lesion reference region, select the corresponding target lesion location from the multiple candidate lesion locations for each of the multiple lesion reference regions.

[0179] First, based on the probability of lesion inclusion and rejection and their location relationships, at least one sample reference lesion location is selected for each lesion reference region. Then, combined with quadrant regions, a further screening process is performed on the at least one sample reference lesion location obtained for each lesion reference region. Finally, the target lesion location is selected for each lesion reference region. This avoids selecting lesion locations only for a certain area in the image to be detected, thus missing other potential lesion locations. The system automates target detection according to clinical guidelines, improving the accuracy of target detection.

[0180] As one embodiment, when further screening at least one sample reference lesion location obtained for each lesion reference region, at least one group of reference lesion locations can be determined from each reference lesion location based on the quadrant region where each obtained reference lesion location is located. Each group of reference lesion locations contains multiple reference lesion locations whose quadrant regions are matched, but whose respective lesion reference regions are different. For each of the at least one group of reference lesion locations, the following steps are performed: the reference lesion location with the highest lesion rejection probability among the multiple reference lesion locations contained in the group is selected as the target lesion location. From multiple candidate lesion locations, at least one lesion reference region other than the lesion reference region to which each target lesion location belongs is selected as the corresponding target lesion location.

[0181] Please refer to Figure 6A Taking two lesion reference regions and two reference lesion locations contained in the image to be detected, denoted as A and B respectively, these two reference lesion locations can be divided into a group of reference lesion locations because they are located in different quadrants and belong to different lesion reference regions.

[0182] The lesion selection probabilities for these two reference lesion locations are 72.5% and 86%, respectively. Therefore, the reference lesion location with the highest lesion selection probability, i.e., the reference lesion location marked B, is selected as the target lesion location, while the reference lesion location marked A is discarded.

[0183] First, the target lesion location is determined in the lesion reference area where the reference lesion location with the highest lesion rejection probability value is located. Then, the target lesion location is selected for other lesion reference areas. This can ensure that the reference lesion location with the highest lesion rejection probability value will be selected as the target detection location, avoiding the omission of reference lesion locations with high lesion rejection probability and causing false detection, thus improving the accuracy of target detection.

[0184] As one embodiment, when continuing to select target lesion locations for other lesion reference regions, at least one reference lesion location can be identified among multiple reference lesion locations, excluding the reference lesion location with the highest lesion rejection probability. For at least one reference lesion location, the following steps are performed: from multiple candidate lesion locations, among each candidate lesion location located within the lesion reference region to which a reference lesion location belongs, a secondary reference location is selected whose lesion rejection probability is within the rejection probability range, where the maximum value in the rejection probability range is less than the rejection probability threshold. Based on the quadrant regions of each of the at least one secondary reference location, if it is determined that there is no secondary reference location group among the at least one secondary reference location, the at least one secondary reference location is taken as the corresponding target lesion location; wherein, the quadrant regions of the multiple secondary reference locations included in the secondary reference location group are matched, and the lesion reference regions to which they belong are different.

[0185] Please refer to Figure 6B ,based on Figure 6A The image to be detected also contains two reference lesion locations, denoted as C and D. The lesion rejection probabilities for the two reference lesion locations are 47.5% and 66.5%, respectively. Taking a rejection probability range of 50% to 86% as an example, since the lesion rejection probability of the reference lesion location marked D is within the rejection probability range, and there is no secondary reference location group within the quadrant region where the reference lesion location marked D is located, the reference lesion location marked D can be used as the target lesion location.

[0186] For at least one reference lesion location other than the reference lesion location with the highest probability of lesion selection, a secondary reference location with the second highest probability of lesion selection is selected in other quadrants. This ensures that the reference lesion location with the highest probability of lesion selection is not discarded, while also selecting a secondary reference location with a relatively high probability of lesion selection for each quadrant. This balances multiple location association conditions and helps improve the accuracy of target detection.

[0187] As one embodiment, if a target lesion location has been selected for each quadrant region, but there are still independent lesion reference regions where no target lesion location has been selected, the following steps can be performed for at least one set of reference lesion locations: the reference lesion location with the highest lesion selection probability among the multiple reference lesion locations included in a set of reference lesion locations can be selected as the target lesion location, and then the reference lesion locations other than the at least one set of reference lesion locations can be selected as the target lesion locations.

[0188] This ensures that the lesion reference area is not discarded, and that the target lesion location is selected in each quadrant. This balances multiple location correlation conditions and helps improve the accuracy of target detection.

[0189] S204, based on the obtained location of at least one target lesion, obtain the target detection result.

[0190] After obtaining at least one target lesion location, the at least one target lesion location can be directly used as the target detection result; or the at least one target lesion location can be represented in various data forms in the target detection result, such as in image form, coordinate form, text description form, relative position form, etc., without any specific restrictions.

[0191] As one embodiment, after obtaining the target detection results, for at least one target lesion location, the following can be performed: based on the biopathological features extracted from the target lesion location, lesion detection can be performed on the target lesion location to obtain lesion detection results. Lesion detection can be a biopsy or other detection methods, without limitation. The lesion detection results indicate whether precancerous lesions, etc., exist at the corresponding target lesion location, without limitation.

[0192] The following uses cervical precancerous lesions as an example. The reference site is the cervical os, the reference lesion area is the acetowhite epithelial lesion area, and one acetowhite epithelial lesion area represents one acetowhite epithelium involved clinically. The candidate lesion location or target lesion location is the biopsy point. Based on this, the target detection method provided in the embodiments of this application will be introduced. Please refer to... Figure 7 .

[0193] The image to be tested is divided into regions to obtain 3 reference lesion regions. The image to be tested is also divided into regions to obtain the location of the cervix. Finally, the image to be tested is divided into regions to obtain 10 candidate lesion locations.

[0194] After establishing a reference coordinate system with the cervical os as the origin, positional association conditions were set based on the positional relationship between the reference coordinate system, 10 candidate lesion locations, and 3 reference lesion regions. These conditions were used to limit the number of biopsy points for the same acetowhite epithelium to one, and to ensure that each of the four quadrants centered on the cervical os has only one biopsy point, and that each acetowhite epithelium can be biopsied.

[0195] For each of the three lesion reference regions, target lesion locations that satisfy the location association condition are selected from 10 candidate lesion locations. For example, if there are multiple different acetowhite epithelial lesion regions, including acetowhite epithelial lesion region A, acetowhite epithelial lesion region B, and acetowhite epithelial lesion region C, the biopsy points with the highest lesion rejection probability in each acetowhite epithelial lesion region are biopsy point A-1, biopsy point B-1, and biopsy point C-1. Within the same quadrant region (such as the first quadrant region), the biopsy point with the highest lesion rejection probability in that quadrant region (such as biopsy point A-1) is retained. Therefore, for the discarded biopsy points (biopsy point B-1 and biopsy point C-1) within this quadrant, biopsy points with the second highest lesion rejection probability are selected from their corresponding acetowhite epithelial lesion areas (acetowhite epithelial lesion area B and acetowhite epithelial lesion area C). This process continues until biopsy points are selected in all four quadrants, but there are still acetowhite epithelial lesion areas where no biopsy points are selected. In this case, the biopsy point with the highest lesion rejection probability in that acetowhite epithelial lesion area is selected. Thus, the target lesion locations corresponding to the three lesion reference areas mentioned above can be obtained.

[0196] Based on the same inventive concept, embodiments of this application provide a target detection device capable of achieving the functions corresponding to the aforementioned target detection method. Please refer to... Figure 8 The device includes a first processing module 801 and a second processing module 802, wherein:

[0197] First processing module 801: used to divide the region where the abnormal tissue is located in the image to be detected based on the biological tissue features extracted from the image to be detected, and obtain at least one lesion reference region;

[0198] The first processing module 801 is also used to: predict the location of lesions in the image to be detected based on the visual pathological features extracted from the image to be detected, and obtain multiple candidate lesion locations;

[0199] The second processing module 802 is used to perform the following for at least one lesion reference region: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, select the target lesion location whose positional relationship satisfies the positional association condition from the multiple candidate lesion locations;

[0200] The second processing module 802 is also used to: obtain target detection results based on the obtained at least one target lesion location.

[0201] In one possible embodiment, the second processing module 802 is specifically used for:

[0202] Based on the positional relationship between multiple candidate lesion locations and a lesion reference region, determine the number of candidate lesion locations that are located within a lesion reference region among the multiple candidate lesion locations;

[0203] When the number of locations is not greater than a preset threshold, the candidate lesion location located within a lesion reference area among multiple candidate lesion locations is taken as the target lesion location.

[0204] In one possible embodiment, after predicting the location of a lesion in the image to be detected, multiple candidate lesion locations are each associated with a predicted lesion probability.

[0205] The second processing module 802 is also used for:

[0206] After determining the number of candidate lesion locations within a lesion reference area among multiple candidate lesion locations, if the number of locations exceeds a preset threshold, the lesion rejection probability corresponding to each of the multiple candidate lesion locations is obtained; among them, the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location;

[0207] From the candidate lesion locations located within a lesion reference area, the candidate lesion locations with a lesion rejection probability greater than a rejection probability threshold are selected as the target lesion locations.

[0208] In one possible embodiment, the second processing module 802 is further configured to:

[0209] Before obtaining the lesion rejection probability corresponding to each of the multiple candidate lesion locations, the proportion of pixels belonging to at least one lesion reference region among the pixels covered by each of the multiple candidate lesion locations is counted to obtain the corresponding region location proportion.

[0210] The predicted lesion probability and regional location proportion of each of the multiple candidate lesion locations are combined to obtain the corresponding lesion rejection probability.

[0211] In one possible embodiment, the second processing module 802 is specifically used for:

[0212] The number of pixels covered by each of the multiple candidate lesion locations is counted to obtain the corresponding number of pixels;

[0213] The number of pixels belonging to at least one lesion reference region is counted in each of the pixels covered by multiple candidate lesion locations to obtain the corresponding number of region points.

[0214] For multiple candidate lesion locations, the following steps are performed: Based on the ratio between the number of region points and the number of pixels corresponding to a candidate lesion location, the corresponding region location proportion is determined.

[0215] In one possible embodiment, when at least one lesion reference region is multiple lesion reference regions, the second processing module 802 is further configured to:

[0216] Before selecting the target lesion location whose positional relationship satisfies the positional association condition from multiple candidate lesion locations based on the positional relationship between multiple candidate lesion locations and a lesion reference region, the reference location is obtained by identifying the reference location in the image to be detected based on the biological part features extracted from the image to be detected.

[0217] Establish a reference coordinate system with the reference location as the origin;

[0218] Based on a reference coordinate system, multiple lesion reference regions, and multiple candidate lesion locations, location association conditions are set.

[0219] In one possible embodiment, the second processing module 802 is further configured to:

[0220] For multiple lesion reference regions, the following steps are performed: determine the region inclusion relationship between each of the multiple candidate lesion locations and a lesion reference region in the image to be detected;

[0221] For each quadrant region divided by the reference coordinate system, the following steps are performed: determine the quadrant inclusion relationship between each of the multiple candidate lesion locations and a quadrant region, and the overlap relationship between each of the multiple lesion reference regions and a quadrant region;

[0222] For multiple lesion reference regions, the following steps are performed: based on the obtained inclusion relationships of each region, inclusion relationships of each quadrant, and overlap relationships, the positional relationship between multiple candidate lesion locations and a lesion reference region is determined.

[0223] In one possible embodiment, after predicting the location of a lesion in the image to be detected, multiple candidate lesion locations are each associated with a predicted lesion probability.

[0224] The second processing module 802 is specifically used for:

[0225] Obtain the lesion rejection probability corresponding to each of the multiple candidate lesion locations; wherein, the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location;

[0226] For multiple lesion reference regions, the following steps are performed: Based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region and the obtained lesion rejection probabilities, a reference lesion location is selected from the multiple candidate lesion locations whose rejection probability is greater than the rejection probability threshold and which is located within a lesion reference region.

[0227] Based on the quadrant region where each reference lesion location is located and the lesion reference region to which it belongs, the corresponding target lesion location is selected from multiple candidate lesion locations for multiple lesion reference regions.

[0228] In one possible embodiment, the second processing module 802 is specifically used for:

[0229] Based on the quadrant regions of each obtained reference lesion location, at least one group of reference lesion locations is determined from each reference lesion location; wherein, the quadrant regions of the multiple reference lesion locations in each group of reference lesion locations are matched, and the reference lesion regions to which they belong are different; the quadrant regions are obtained based on the reference coordinate system.

[0230] For at least one set of reference lesion locations, the following steps are performed: the reference lesion location with the highest probability of lesion selection among the multiple reference lesion locations contained in a set of reference lesion locations is selected as the target lesion location;

[0231] From multiple candidate lesion locations, each is a lesion reference region other than the lesion reference region to which each target lesion location belongs, and the corresponding target lesion location is selected.

[0232] In one possible embodiment, the second processing module 802 is specifically used for:

[0233] Among multiple reference lesion locations, identify at least one reference lesion location other than the one with the highest probability of lesion acceptance / rejection.

[0234] For at least one reference lesion location, the following steps are performed: From multiple candidate lesion locations, among each candidate lesion location located within the lesion reference region to which a reference lesion location belongs, select the secondary reference location whose lesion rejection probability is within the rejection probability range; wherein, the maximum value in the rejection probability range is less than the rejection probability threshold.

[0235] Based on the quadrant regions of each of the at least one secondary reference locations, if it is determined that there is no secondary reference location group among the at least one secondary reference locations, the at least one secondary reference location is taken as the corresponding target lesion location; wherein, the quadrant regions of the multiple secondary reference locations included in the secondary reference location group are matched, and the lesion reference regions to which they belong are different.

[0236] In one possible embodiment, the second processing module 802 is further configured to:

[0237] For at least one set of reference lesion locations, the following steps are performed: the reference lesion location with the highest probability of lesion selection among the multiple reference lesion locations included in a set of reference lesion locations is selected as the target lesion location. Then, the reference lesion locations other than the at least one set of reference lesion locations are selected as the target lesion locations.

[0238] In one possible embodiment, the first processing module 801 is specifically used for:

[0239] Images of the cervical tissue surface after treatment with acetic acid solution were acquired.

[0240] Tissue features are extracted from the image to be detected to obtain biological tissue features;

[0241] Based on the obtained biological tissue features, the regions where the biological tissues with abnormal colors are located in the image to be detected are divided into regions to obtain at least one lesion reference region.

[0242] In one possible embodiment, the first processing module 801 is specifically used for:

[0243] Cervical features are extracted from the images to be tested to obtain biological features;

[0244] Based on the obtained biological features, the cervix in the image to be detected is identified, and the location of the reference site is obtained.

[0245] Please refer to Figure 9 This is a computer device 900 provided in the embodiments of this application. The computer device 900 can, for example, be... Figure 1B The client 101 or server 102 in the system. The current and historical versions of the data storage program and the application software corresponding to the data storage program can be installed on a computer device 900, which includes a processor 980 and a memory 920. In some embodiments, the computer device 900 may include a display unit 940, which includes a display panel 941 for displaying a user-interactive interface, etc.

[0246] In one possible embodiment, the display panel 941 may be configured in the form of a liquid crystal display (LCD) or an organic light-emitting diode (OLED).

[0247] The processor 980 is used to read a computer program and then execute the methods defined by the computer program. For example, the processor 980 reads a data storage program or file, thereby running the data storage program on the computer device 900 and displaying the corresponding interface on the display unit 940. The processor 980 may include one or more general-purpose processors, and may also include one or more DSPs (Digital Signal Processors) for performing related operations to implement the technical solutions provided in the embodiments of this application.

[0248] The memory 920 generally includes main memory and secondary storage. Main memory can be random access memory (RAM), read-only memory (ROM), and cache, etc. Secondary storage can be a hard disk, optical disk, USB flash drive, floppy disk, or magnetic tape drive, etc. The memory 920 is used to store computer programs and other data. The computer programs include applications corresponding to each client, and other data may include data generated after the operating system or applications are run, including system data (e.g., operating system configuration parameters) and user data. In this embodiment, the computer program is stored in the memory 920, and the processor 980 executes the computer program in the memory 920 to implement any of the methods described in the preceding figures.

[0249] The aforementioned display unit 940 is used to receive input digital information, character information, or contact touch operations / non-contact gestures, and to generate signal inputs related to user settings and function control of the computer device 900. Specifically, in this embodiment, the display unit 940 may include a display panel 941. The display panel 941, for example, is a touch screen, which can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or on the display panel 941), and drive corresponding connection devices according to a pre-set program.

[0250] In one possible embodiment, the display panel 941 may include two parts: a touch detection device and a touch controller. The touch detection device detects the player's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller. The touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 980. It can also receive and execute commands from the processor 980.

[0251] The display panel 941 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the display unit 940, in some embodiments, the computer device 900 may also include an input unit 930. The input unit 930 may include an image input device 931 and other input devices 932, wherein the other input devices may include, but are not limited to, one or more of the following: a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), a trackball, a mouse, and a joystick.

[0252] In addition to the above, the computer device 900 may also include a power supply 990 for powering other modules, an audio circuit 960, a near-field communication module 970, and an RF circuit 910. The computer device 900 may also include one or more sensors 950, such as an accelerometer, a light sensor, and a pressure sensor. The audio circuit 960 specifically includes a speaker 961 and a microphone 962, for example, the computer device 900 can use the microphone 962 to collect the user's voice and perform corresponding operations.

[0253] As one embodiment, the number of processors 980 can be one or more, and the processors 980 and the memory 920 can be coupled together or relatively independent.

[0254] As one example, Figure 9 The processor 980 in the middle can be used to implement, for example Figure 8 The functions of the first processing module 801 and the second processing module 802 in the process.

[0255] As one example, Figure 9 The processor 980 in the text can be used to implement the functions of the server or terminal devices discussed above.

[0256] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by a computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the computer program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0257] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of software products, for example, through computer program products. These computer program products are stored in a storage medium and include computer programs used to cause a computer device to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0258] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A target detection method, characterized in that, include: Based on the biological tissue features extracted from the image to be detected, the region where the abnormal tissue is located in the image to be detected is divided into regions to obtain at least one lesion reference region. Based on the visual pathological features extracted from the image to be detected, the location of the lesion in the image to be detected is predicted, and multiple candidate lesion locations are obtained. For the at least one lesion reference region, the following steps are performed: based on the positional relationship between each of the plurality of candidate lesion locations and a lesion reference region, a target lesion location whose positional relationship satisfies the positional association condition is selected from the plurality of candidate lesion locations; Based on the location of at least one target lesion, the target detection result is obtained.

2. The method according to claim 1, characterized in that, The step of selecting a target lesion location from the plurality of candidate lesion locations that satisfies the positional correlation condition based on the positional relationship between each of the plurality of candidate lesion locations and a lesion reference region includes: Based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, determine the number of candidate lesion locations located within the lesion reference region among the multiple candidate lesion locations; When the number of locations is not greater than a preset threshold, the candidate lesion location located within the lesion reference area among the multiple candidate lesion locations is taken as the target lesion location.

3. The method according to claim 2, characterized in that, After predicting the location of the lesion in the image to be detected, each of the multiple candidate lesion locations is associated with a predicted lesion probability; After determining the number of candidate lesion locations within the reference lesion region among the plurality of candidate lesion locations, the method further includes: When the number of locations exceeds the preset threshold, the lesion selection probability corresponding to each of the multiple candidate lesion locations is obtained; wherein, the lesion selection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location; From the candidate lesion locations located within the lesion reference area, the candidate lesion locations with a lesion rejection probability greater than the rejection probability threshold are selected as the target lesion locations.

4. The method according to claim 3, characterized in that, Before obtaining the lesion rejection probability corresponding to each of the multiple candidate lesion locations, the method further includes: The proportion of pixels belonging to at least one lesion reference region is calculated among the pixels covered by each of the multiple candidate lesion locations to obtain the corresponding region location proportion. The predicted lesion probability and the regional location ratio corresponding to each of the multiple candidate lesion locations are combined to obtain the corresponding lesion rejection probability.

5. The method according to claim 4, characterized in that, The step of calculating the percentage of pixels belonging to the at least one lesion reference region among the pixels covered by each of the multiple candidate lesion locations to obtain the corresponding region location percentage includes: The number of pixels covered by each of the multiple candidate lesion locations is counted to obtain the corresponding number of pixels; The number of pixels belonging to at least one lesion reference region is counted in each of the pixels covered by the multiple candidate lesion locations to obtain the corresponding number of region points. For each of the multiple candidate lesion locations, the following steps are performed: determine the corresponding region location percentage based on the ratio between the number of region points and the number of pixels corresponding to a candidate lesion location.

6. The method according to any one of claims 1 to 5, characterized in that, When the at least one lesion reference region is multiple lesion reference regions, before performing the following steps for each of the at least one lesion reference regions: selecting a target lesion location whose positional relationship satisfies the positional association condition from the multiple candidate lesion locations based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, the method further includes: Based on the biological features extracted from the image to be detected, a reference part in the image to be detected is identified, and the location of the reference part is obtained; A reference coordinate system is established with the location of the reference part as the origin; Based on the reference coordinate system, the multiple lesion reference regions, and the multiple candidate lesion locations, location association conditions are set.

7. The method according to claim 6, characterized in that, The positional relationship between each of the multiple candidate lesion locations and a lesion reference region is determined using the following method: For each of the multiple lesion reference regions, the following steps are performed: determining the region inclusion relationship between each of the multiple candidate lesion locations and a lesion reference region in the image to be detected; For each quadrant region divided by the reference coordinate system, the following steps are performed: determining the quadrant inclusion relationship between each of the multiple candidate lesion locations and a quadrant region, and the overlap relationship between each of the multiple lesion reference regions and the quadrant region; For each of the multiple lesion reference regions, the following steps are performed: based on the obtained inclusion relationships of each region, inclusion relationships of each quadrant, and overlap relationships, the positional relationship between the multiple candidate lesion locations and a lesion reference region is determined.

8. The method according to claim 6, characterized in that, After predicting the location of the lesion in the image to be detected, each of the multiple candidate lesion locations is associated with a predicted lesion probability; Then, for each of the multiple lesion reference regions, the following steps are performed: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, selecting the target lesion location from the multiple candidate lesion locations whose positional relationship satisfies the positional association condition, including: Obtain the lesion rejection probability corresponding to each of the multiple candidate lesion locations; wherein, the lesion rejection probability is positively correlated with the predicted lesion probability associated with the corresponding candidate lesion location; For the multiple lesion reference regions, the following steps are performed: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region and the obtained lesion rejection probabilities, a reference lesion location with a rejection probability greater than a rejection probability threshold and located within the lesion reference region is selected from the multiple candidate lesion locations. Based on the quadrant region where each reference lesion location is located and the lesion reference region to which it belongs, the corresponding target lesion location is selected from the multiple candidate lesion locations for each of the multiple lesion reference regions.

9. The method according to claim 8, characterized in that, The step of selecting a corresponding target lesion location from the multiple candidate lesion locations, based on the quadrant region where each reference lesion location is located and the lesion reference region to which it belongs, includes: Based on the quadrant regions of each obtained reference lesion location, at least one group of reference lesion locations is determined from the reference lesion locations; wherein, the quadrant regions of the multiple reference lesion locations included in each group of reference lesion locations are matched, and the lesion reference regions to which they belong are different; the quadrant regions are obtained based on the reference coordinate system. For each of the at least one set of reference lesion locations, the following is performed: the reference lesion location with the highest probability of lesion selection among the multiple reference lesion locations contained in a set of reference lesion locations is taken as the target lesion location; From the plurality of candidate lesion locations, at least one lesion reference region other than the lesion reference region to which each target lesion location belongs is selected from the plurality of lesion reference regions.

10. The method according to claim 9, characterized in that, The step of selecting a corresponding target lesion location from the plurality of candidate lesion locations, which are respectively at least one lesion reference region outside the lesion reference region to which each target lesion location belongs, includes: Among the plurality of reference lesion locations, at least one reference lesion location is identified, excluding the reference lesion location with the highest probability of lesion rejection. For the at least one reference lesion location, the following steps are performed: from the plurality of candidate lesion locations, among each candidate lesion location located within the lesion reference region to which a reference lesion location belongs, select a secondary reference location whose lesion rejection probability is within the rejection probability range; wherein, the maximum value in the rejection probability range is less than the rejection probability threshold. Based on the quadrant regions of each of the at least one secondary reference positions, if it is determined that there is no secondary reference position group among the at least one secondary reference positions, the at least one secondary reference position is taken as the corresponding target lesion location; wherein, the quadrant regions of the multiple secondary reference positions included in the secondary reference position group are matched, and the lesion reference regions to which they belong are different.

11. The method according to claim 9, characterized in that, After performing the following steps for each of the at least one set of reference lesion locations: selecting the reference lesion location with the highest lesion rejection probability among multiple reference lesion locations included in a set of reference lesion locations as the target lesion location, the method further includes: The reference lesion location other than the at least one group of reference lesion locations is taken as the target lesion location.

12. The method according to any one of claims 1 to 5, characterized in that, The method involves dividing the region containing abnormal tissue in the image to be detected based on biological tissue features extracted from the image to be detected, thereby obtaining at least one lesion reference region, including: Images of the cervical tissue surface after treatment with acetic acid solution were acquired. Tissue features are extracted from the image to be detected to obtain biological tissue features; Based on the obtained biological tissue features, the regions where the biological tissues with abnormal colors are located in the image to be detected are divided into regions to obtain at least one lesion reference region.

13. The method according to claim 6, characterized in that, The step of identifying reference regions in the image to be detected based on biological feature extraction from the image to be detected, and obtaining the location of the reference regions, includes: Cervical features are extracted from the image to be detected to obtain biological site features; Based on the obtained biological features, the cervix in the image to be detected is identified, and the location of the reference site is obtained.

14. A target detection device, characterized in that, include: First processing module: used to divide the region where the abnormal tissue is located in the image to be detected based on the biological tissue features extracted from the image to be detected, and obtain at least one lesion reference region; The first processing module is further configured to: predict the location of lesions in the image to be detected based on the visual pathological features extracted from the image to be detected, and obtain multiple candidate lesion locations; The second processing module is used to perform the following for the at least one lesion reference region: based on the positional relationship between each of the multiple candidate lesion locations and a lesion reference region, select the target lesion location whose positional relationship satisfies the positional association condition from the multiple candidate lesion locations; The second processing module is also used to: obtain target detection results based on the obtained at least one target lesion location.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 13.

16. A computer device, characterized in that, include: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the method as described in any one of claims 1 to 13 according to the obtained program instructions.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the method as described in any one of claims 1 to 13.

Images