Medical image lesion auxiliary positioning method and device based on high-quality negative prompt

Abstract

This application discloses a method and apparatus for assisted lesion localization in medical images based on high-quality negative cues, belonging to the field of image analysis technology. The method includes: acquiring a medical image and generating an initial mask for the lesion region based on bounding box cues; extracting regional features of the lesion region and the surrounding background region based on the initial mask, and generating high-quality negative cue points based on the similarity analysis between the regional features of the lesion region and the background region; acquiring positive cue points within the lesion region, and inputting the positive cue points, high-quality negative cue points, and bounding box cues into a visual basic segmentation model to generate a first lesion mask; and fusing the initial mask and the first lesion mask to obtain the final lesion localization result of the medical image. This method can effectively improve the accuracy of assisted lesion localization in complex medical image scenes.

Description

Technical Field

[0001] This application belongs to the field of image analysis technology, specifically relating to a method and device for assisted localization of lesions in medical images based on high-quality negative cues. Background Technology

[0002] Localization of lesions in medical images is a crucial step in AI-assisted diagnosis, widely applied in various medical scenarios including endoscopic images, CT, MRI, ultrasound, pathological images, and fundus images. Accurate localization of lesion regions aids in disease screening, lesion assessment, lesion segmentation, and treatment decisions, significantly improving the efficiency and accuracy of clinical diagnosis. However, lesion regions in different medical images typically exhibit characteristics such as blurred boundaries, irregular shapes, large scale variations, low local contrast, and complex tissue structures. Furthermore, images may contain noise, artifacts, organ texture interference, bright reflections, occluded areas, or structures similar to normal tissue, making it difficult to effectively distinguish lesion regions from the background and increasing the challenge of lesion-assisted localization.

[0003] Taking ileocecal ulcer as an example, it is a common pathological manifestation in digestive system diseases, widely seen in Crohn's disease, intestinal tuberculosis, and intestinal Behcet's disease. Due to the significant overlap in clinical symptoms, lesion location, and endoscopic findings among these three diseases, differential diagnosis has long been a key and challenging issue in the diagnosis and treatment of digestive system diseases. Endoscopic examination is one of the important means of assessing ileocecal ulcers, allowing direct observation of lesions within the intestinal lumen, such as longitudinal ulcers, annular ulcers, deep ulcers, and irregular ulcers. However, the presentation of ileocecal ulcers in endoscopic images is very challenging. The lesion area typically exhibits blurred soft tissue boundaries, low local contrast, and irregular shape, accompanied by complex background factors such as reflectivity, bright areas, and intestinal fluid obscuring the lesion, making it difficult to establish a clear boundary between the lesion area and normal tissue, easily leading to localization deviation or misidentification of the background area.

[0004] In recent years, with the development of deep learning and large model technology, prompt-based image segmentation methods have received widespread attention in medical image processing. In particular, general segmentation models can quickly generate target region masks through bounding box prompts, point prompts, etc., demonstrating good segmentation capabilities in both natural images and some medical image scenarios. However, existing prompting mechanisms still have significant limitations in complex medical scenarios such as ileocecal ulcers.

[0005] First, existing methods mainly rely on positive cues to guide the target area, but lack effective constraints on the complex background areas surrounding the lesion. For example, in endoscopic images of ileocecal ulcers, due to the blurred boundaries of the lesion area and the presence of a large amount of local reflection and inflammatory texture, relying solely on positive cues can easily lead the model to focus on highlighted areas, background areas, or local texture noise, resulting in lesion localization misalignment.

[0006] Secondly, most existing negative prompts use random point selection or simple background area sampling methods, resulting in a lack of targeted generation and difficulty in covering background areas that are easily confused with lesions. Especially in complex medical images, the background areas near the lesion edges often have similar colors, textures, or inflammatory features to the lesion. Random negative prompts are unable to effectively suppress these interfering information, affecting the accuracy of lesion localization.

[0007] Relying solely on single-point cues to generate lesion masks is prone to issues such as localized missing areas, region offsets, and incomplete boundaries. When the model relies only on bounding box cues, it often introduces excessive background regions; when relying only on point cues, incomplete lesion coverage or deviation of the region of interest may occur. Therefore, how to construct a high-quality negative cue mechanism that can accurately represent the background interference information around lesions and further improve the auxiliary localization accuracy of lesion regions in complex medical images is a crucial problem that urgently needs to be solved in the field of medical image analysis. Summary of the Invention

[0008] The purpose of this application is to provide a method and device for assisted localization of lesions in medical images based on high-quality negative cues. Starting from the confusion relationship between the lesion area and the background area, the method generates high-quality negative cue points that are closer to the lesion boundary through feature similarity analysis between the lesion-related area and the background area. These negative cue points are then combined with positive cue points to guide the localization of the lesion area, thereby improving the accuracy and stability of lesion area recognition in complex medical images.

[0009] According to a first aspect of the embodiments of this application, a method for lesion-assisted localization in medical images based on high-quality negative cues is provided, comprising: Acquire medical images and generate an initial mask for the lesion region based on bounding box cues; Based on the initial mask, extract the regional features of the lesion area and the regional features of the background area surrounding the lesion area. Based on the similarity analysis between the regional features of the lesion area and the regional features of the background area, generate high-quality negative cue points. Obtain positive cue points within the lesion region, and input the positive cue points, high-quality negative cue points, and bounding box cue points into the visual basic segmentation model to generate the first lesion mask; The initial mask is fused with the first lesion mask to obtain the final lesion localization result of the medical image.

[0010] In some optional embodiments of this application, high-quality negative cue points are generated based on similarity analysis between the regional features of the lesion region and the regional features of the background region, specifically including: Medical images are input into a visual feature extraction model to extract deep semantic features from the medical images; Based on the initial mask, extract the regional features of the lesion region and the regional features of the background region from the deep semantic features respectively; Calculate the first similarity between each pixel in the medical image and the regional features of the lesion region to obtain the confidence map of the lesion region; Calculate the second similarity between the regional features of each pixel in the medical image and the background region to obtain the background region confidence map; Based on the confidence map of the background region, high background confidence regions are selected as candidate negative cue regions within the boundary neighborhood of the lesion region. Identify high-quality negative cue points from the candidate negative cue region.

[0011] In some optional embodiments of this application, determining high-quality negative cue points from the candidate negative cue region specifically includes: Cluster analysis was performed on the candidate negative suggestion regions to divide them into multiple local regions; Select the center location of each local area as the high-quality negative cue point.

[0012] In some optional embodiments of this application, based on the background region confidence map, high background confidence regions are selected as candidate negative cue regions within the boundary neighborhood of the lesion region, specifically including: Determine the boundaries of the lesion region based on the initial mask; With the boundary as the center, extend the preset range towards the background area to determine the spatial range of the boundary neighborhood; Within the boundary neighborhood, regions with a background confidence level higher than a preset threshold in the background region confidence map are selected as candidate negative cue regions.

[0013] In some optional embodiments of this application, obtaining positive markers within the lesion area specifically includes: Based on the confidence map of the lesion area, select areas within the lesion area where the confidence level of the lesion is higher than a preset threshold; Cluster analysis was performed on the selected region, dividing it into multiple local regions; Select the center position of each local area as the positive prompt point.

[0014] In some optional embodiments of this application, the initial mask is fused with the first lesion mask to obtain the final lesion localization result of the medical image, specifically including: Calculate the degree of overlap between the initial mask and the first lesion mask; The consistency between the initial mask and the first lesion mask is determined based on the degree of overlap; When the consistency exceeds a preset threshold, the overlapping area between the initial mask and the first lesion mask is retained, and the difference area between the initial mask and the first lesion mask is supplemented and fused. When the consistency is below a preset threshold, the confidence weight of the difference region is reduced; The final lesion localization result is generated based on the fusion results.

[0015] In some optional embodiments of this application, the final lesion localization result is generated based on the fusion result, specifically including: The merged mask region is evaluated for regional confidence, and regions with confidence scores higher than the preset confidence threshold are retained. The final lesion localization result is generated based on the preserved region.

[0016] According to a second aspect of the embodiments of this application, a medical image lesion-assisted localization device based on high-quality negative feedback is provided, comprising: The image acquisition module is used to acquire medical images; The initial mask generation module is used to generate an initial mask for the lesion region based on bounding box prompts. The negative prompt generation module is used to extract the regional features of the lesion area and the regional features of the background area surrounding the lesion area based on the initial mask, and generate high-quality negative prompt points based on the similarity analysis between the regional features of the lesion area and the regional features of the background area. The positive feedback generation module is used to obtain positive feedback points within the lesion area; The segmentation module is used to input positive cue points, high-quality negative cue points, and bounding box cue points into the visual basic segmentation model to generate the first lesion mask; The mask fusion module is used to fuse the initial mask with the first lesion mask to obtain the final lesion localization result of the medical image.

[0017] According to a third aspect of the embodiments of this application, an electronic device is provided, which may include: processor; Memory used to store processor-executable instructions; The processor is configured to execute instructions to implement a medical image lesion-assisted localization method based on high-quality negative cues, as shown in any embodiment of the first aspect.

[0018] According to a fourth aspect of the embodiments of this application, a storage medium is provided, wherein when instructions in the storage medium are executed by a processor of an information processing device or a server, the information processing device or server implements the medical image lesion-assisted localization method based on high-quality negative cues as shown in any embodiment of the first aspect.

[0019] The above-mentioned technical solution of this application has the following beneficial technical effects: (1) Improve the accuracy of auxiliary localization of lesion regions in complex medical images. Existing cue-based lesion localization methods mainly rely on bounding box cues or positive cue points to guide the target region. They lack effective constraints on complex background regions and are easily affected by noise, artifacts, tissue textures, highlighted areas, and normal tissue structures, leading to lesion region localization deviation. This application generates high-quality negative cue points by analyzing highly confusing background regions in the lesion neighborhood. This enables the model to simultaneously focus on the differences between the lesion region and easily confused background regions, thereby improving the accuracy of lesion region identification and localization.

[0020] (2) Enhanced background suppression capability under complex background interference conditions. Existing negative prompting methods typically employ random sampling, fixed-region sampling, or background sampling methods far from the target area, which are difficult to cover background areas that are easily confused with the lesion area, thus limiting the background suppression capability. This application prioritizes selecting background areas with high feature similarity to the lesion area within the lesion boundary neighborhood to generate negative prompting points, making the negative prompting more targeted and representative, and more effectively suppressing complex background areas, interference areas near the boundary, and local artifact areas, thereby improving the lesion area boundary identification capability.

[0021] (3) Improve the integrity and localization stability of lesion region boundaries. Existing single-prompt methods are prone to problems such as region missing, boundary discontinuity, or local missegmentation. This application uses a positive and negative prompting collaborative guidance mechanism to enable the model to learn not only the target features of the lesion region, but also the rejection features of the background region, reducing misidentification problems near the lesion boundary and improving the integrity and localization consistency of the lesion region boundaries.

[0022] (4) Reduce the impact of errors caused by the instability of single prompt results. Existing methods usually rely on single prompt results, which are prone to instability in localization results due to prompt position deviation or local noise. This application further proposes a lesion mask fusion mechanism based on consistency analysis of results under multiple prompt conditions, which dynamically fuses lesion masks generated under different prompt conditions, reduces the risk of mislocalization caused by single prompt deviation, and improves the stability of lesion auxiliary localization results.

[0023] This application addresses the problems of insufficient background constraint, strong randomness of negative prompts, incomplete lesion boundaries, and poor stability of single prompt results in existing methods by modeling high-confusion background regions in the vicinity of lesions, generating high-quality negative prompts, co-assisted localization of positive and negative prompts, and lesion mask fusion mechanism. It can effectively improve the accuracy of lesion region assisted localization in complex medical image scenes. Attached Figure Description

[0024] Figure 1 is a schematic flowchart of a medical image lesion-assisted localization method based on high-quality negative cues in an exemplary embodiment of this application; Figure 2 is a schematic diagram of the structure of a medical image lesion-assisted localization device based on high-quality negative cues in an exemplary embodiment of this application; Figure 3 is a schematic diagram of the structure of an electronic device in an exemplary embodiment of this application; Figure 4 is a schematic diagram of the hardware structure of an electronic device in an exemplary embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of this application. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0026] The accompanying drawings illustrate layer structure diagrams according to embodiments of this application. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0027] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0028] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0030] While existing lesion-assisted localization technologies can achieve target area identification to a certain extent, they still have significant shortcomings in complex medical image scenarios.

[0031] Existing cue-based segmentation methods primarily rely on positive cues for target guidance, lacking effective constraints on background regions. Due to the blurred lesion boundaries, complex tissue structures, and low local contrast in complex medical images, relying solely on lesion-internal cues can easily lead to models incorrectly focusing on bright reflective areas, local texture noise, or normal tissue areas, resulting in lesion localization misalignment.

[0032] Existing negative cue generation methods lack specificity. Most methods employ random background sampling or fixed-region negative cueing, but randomly generated negative cue points are often far from the lesion area, failing to cover the neighboring background regions that are easily confused with the lesion. Therefore, the model may still incorrectly retain interfering regions near the lesion edge, reducing the accuracy of lesion localization.

[0033] Existing suggestion mechanisms typically rely on a single suggestion generation result, lacking a comprehensive evaluation of the reliability of different suggestion results. When bounding box suggestions and point suggestions conflict, the model lacks an effective mechanism to determine the credibility of different regions, which can easily lead to problems such as mask region offset, boundary discontinuity, and missing lesion regions.

[0034] Therefore, the technical problem to be solved by this application is to provide a method for assisted localization of lesions in complex medical images based on high-quality negative cues. In complex endoscopic image scenarios, high-quality negative cue points closer to the lesion boundary are generated through feature analysis of the lesion region and the background region, which enhances the model's ability to identify background interference regions in the vicinity of the lesion. Furthermore, by combining positive and negative cue collaborative guidance and mask fusion mechanism, the accuracy, completeness and stability of assisted localization of lesion regions in complex medical images are improved.

[0035] To address the challenges of blurred lesion boundaries, low local contrast, significant interference from reflective areas, and difficulty in effectively distinguishing lesion areas from background areas in complex medical images, this application proposes a high-quality negative cueing-based lesion localization method for complex medical images. This method generates high-quality negative cueing points through feature analysis of the lesion and background areas, and combines positive cueing with a multi-mask fusion mechanism to achieve accurate lesion localization in complex medical image scenarios.

[0036] The method described in this application includes the following steps: (1) Acquire medical images and generate an initial lesion region mask using bounding box prompts; (2) Generate high-quality negative cue points based on the highly confusing background areas in the vicinity of the lesion area; (3) Use positive and negative cues to guide the re-localization of the lesion area; (4) The lesion masks generated under different prompting conditions are fused to obtain the final lesion auxiliary localization result.

[0037] (a) Formation of the initial lesion area To reduce the impact of complex backgrounds on subsequent lesion localization, this application first generates an initial mask for the lesion region using bounding box prompts. For the input medical image, coarse bounding box information of the lesion region is first obtained. The bounding boxes can be generated by manual annotation, physician experience selection, coarse localization algorithms, or other automatic detection methods. Subsequently, the bounding box prompts are input into the visual basic segmentation model to obtain the initial mask for the lesion region.

[0038] Because relying solely on bounding box cues can easily introduce a large amount of background area, the generated initial lesion region often suffers from incomplete boundaries, local offsets, and background redundancy. This application further analyzes the feature relationship between the lesion region and the background region to improve the accuracy of subsequent lesion region localization.

[0039] (ii) High-quality negative point suggestion construction To construct more reliable confidence information for lesion regions, this application further introduces a feature similarity analysis mechanism between lesion regions and background regions. The input medical image is fed into a visual feature extraction model to extract deep semantic features. Based on the initial lesion region mask, lesion region features and background region features are extracted separately, and the similarity relationship between each pixel in the image and the lesion region and background region is calculated. Confidence maps for the lesion region and the background region are obtained using cosine similarity, distance function, or other similarity metrics. The lesion region confidence map describes the degree of association between each pixel and the lesion region, while the background region confidence map describes the degree of similarity between each pixel and the background region.

[0040] Existing negative cue generation methods typically employ random sampling, which struggles to accurately cover background areas that are actually confused with the lesion region, easily leading to residual background areas or lesion boundary shifts. This application filters regions with high background confidence based on a background region confidence map and further restricts the spatial range of candidate regions by combining lesion region boundary information. Unlike random negative cueing, this application prioritizes selecting background regions in the lesion's neighborhood that share similar texture, color, or inflammatory characteristics with the lesion region as candidate negative cueing regions.

[0041] To avoid negative alerts being too concentrated or randomly located, this application employs clustering analysis to process the candidate regions. The clustering algorithm divides the candidate negative regions into multiple local regions, and the center of each region is selected as the negative alert point.

[0042] Compared with traditional negative prompting methods, the negative prompting points generated in this application have the following characteristics: First, it is closer to the boundary of the lesion area. Since the negative cue point is located in the background area adjacent to the lesion, it can more effectively suppress the easily confused background area near the lesion boundary.

[0043] Second, it has higher background representativeness. Negative cue points are derived from high-confidence regions of background rather than randomly sampled regions, thus more accurately representing complex background information.

[0044] Third, the distribution is more uniform. Cluster analysis can avoid local clustering of negative cue points and improve the coverage of complex background areas.

[0045] Through the above methods, this application is able to generate negative prompt information with higher quality and stronger binding force.

[0046] (III) Positive and negative cueing coordinated assisted positioning To further improve the accuracy of assisted localization of lesion areas, this application uses high-quality negative and positive prompts in combination to re-localize the lesion areas.

[0047] Based on the confidence map of the lesion region, high-confidence lesion areas are selected as positive cue points. These positive cue points are mainly distributed within the lesion region, enhancing the model's ability to focus on the core lesion area. The positive cue points, negative cue points, and initial bounding box cues are jointly input into the visual basic segmentation model. Positive cue points guide the model to focus on the lesion region, negative cue points suppress the background region surrounding the lesion, and bounding box cues limit the approximate spatial extent of the lesion.

[0048] By employing a combined positive and negative cueing approach, the model can learn not only what to focus on but also what to ignore, reducing interference from reflective areas, local texture noise, and complex background regions. Compared to methods relying solely on positive cueing, this application effectively reduces lesion localization shift and improves the integrity of lesion region boundaries.

[0049] (iv) Lesion masking fusion mechanism To further improve the stability of lesion localization, this application proposes a lesion mask fusion mechanism based on mask reliability analysis. First, a consistency analysis is performed on the lesion mask generated by bounding box prompts and the lesion mask co-generated by positive and negative prompts. The consistency relationship between different masks is evaluated by calculating the degree of overlap. When the two types of masks have high consistency, overlapping areas are retained first, and locally differing areas are supplemented; when the consistency is low, the influence of unreliable areas is reduced to prevent localization deviation caused by a single prompt result. This application can combine regional confidence information to dynamically fuse different regions, retaining lesion regions with high confidence. Finally, a final lesion region mask is generated for auxiliary localization of ileocecal ulcer lesions.

[0050] The above methods can improve the integrity of the lesion area boundary and the stability of its location, and reduce the problem of missegmentation under complex background conditions.

[0051] The following description, in conjunction with the accompanying drawings, details the medical image lesion-assisted localization method based on high-quality negative cues provided in this application through specific embodiments and application scenarios.

[0052] like Figure 1 As shown, in a first aspect of this application, a method for lesion-assisted localization in medical images based on high-quality negative cues is provided, comprising: S110: Acquire medical images and generate an initial mask for the lesion region based on bounding box prompts; S120: Based on the initial mask, extract the regional features of the lesion area and the regional features of the background area surrounding the lesion area. Based on the similarity analysis between the regional features of the lesion area and the regional features of the background area, generate high-quality negative cue points. S130: Obtain positive cue points within the lesion region, and input the positive cue points, high-quality negative cue points, and bounding box cue points into the visual basic segmentation model to generate the first lesion mask; S140: The initial mask is fused with the first lesion mask to obtain the final lesion localization result of the medical image.

[0053] Based on the similarity analysis between the regional features of the lesion area and the regional features of the background area, high-quality negative cue points are generated, specifically including: Medical images are input into a visual feature extraction model to extract deep semantic features from the medical images; Based on the initial mask, extract the regional features of the lesion region and the regional features of the background region from the deep semantic features respectively; Calculate the first similarity between each pixel in the medical image and the regional features of the lesion region to obtain the confidence map of the lesion region; Calculate the second similarity between the regional features of each pixel in the medical image and the background region to obtain the background region confidence map; Based on the confidence map of the background region, high background confidence regions are selected as candidate negative cue regions within the boundary neighborhood of the lesion region. Identify high-quality negative cue points from the candidate negative cue region.

[0054] The first and second similarities are measured by cosine similarity or Euclidean distance.

[0055] The visual basic segmentation model is the Segment Anything Model or its medical image-adapted variant.

[0056] The above-described method addresses the problems of insufficient background constraint, strong randomness of negative prompts, incomplete lesion boundaries, and poor stability of single prompt results in existing methods by modeling high-confusion background regions in the vicinity of lesions, generating high-quality negative prompts, co-assisted localization of positive and negative prompts, and lesion mask fusion mechanism. It can effectively improve the accuracy of assisted localization of lesion regions in complex medical image scenes.

[0057] The above embodiments propose a high-quality negative cue generation mechanism based on the analysis of highly confusing background regions in the lesion's neighborhood. This application constructs confidence maps for the lesion and background regions by analyzing the characteristic relationships between the lesion and background regions, and generates high-quality negative cue points based on background regions in the lesion's neighborhood that have high similarity to the lesion. Unlike existing random background sampling or fixed-region negative cue methods, this application prioritizes screening background regions within the lesion boundary neighborhood that are easily confused with the lesion as candidate negative cue regions, and combines spatial distribution constraints or cluster analysis to generate representative negative cue points, enhancing the model's ability to suppress complex background regions and improving the accuracy of lesion region auxiliary localization.

[0058] The above embodiments also propose a lesion-assisted localization mechanism based on the collaborative guidance of positive and negative cues. This application inputs positive cues, high-quality negative cues, and bounding box cues within the lesion region into the visual basic segmentation model to achieve re-localization of the lesion region. Positive cues enhance the model's ability to focus on the core lesion region, negative cues suppress easily confused background regions near the lesion boundary, and bounding box cues constrain the approximate spatial range of the lesion. Through the collaborative guidance of positive and negative cues, the model not only learns the target features of the lesion region but also the rejection features of the background region, reducing lesion localization offset caused by complex backgrounds, noisy regions, artifact regions, or tissue texture interference, and improving the integrity of the lesion boundary and localization accuracy.

[0059] Furthermore, the above embodiments propose a lesion mask fusion mechanism based on consistency analysis of results under multiple prompting conditions. To address the potential issues of region offset, local missing parts, or inconsistent boundaries in lesion masks generated under different prompting conditions, this application further integrates the bounding box prompting results with the results generated collaboratively by positive and negative prompts, and combines mask consistency analysis or region reliability assessment mechanisms to dynamically fuse different regions. When different mask results have high consistency, overlapping regions are preferentially retained and locally differing regions are supplemented; when consistency is low, the influence of unreliable regions is reduced, improving the stability, completeness, and robustness of the lesion-assisted localization results.

[0060] It should be noted that the lesion-assisted localization method based on high-quality negative cues in medical images provided in this application can be executed by a lesion-assisted localization device based on high-quality negative cues, or by a control module within that device for executing the lesion-assisted localization method based on high-quality negative cues. This application uses the execution of the lesion-assisted localization method based on high-quality negative cues by the lesion-assisted localization device as an example to illustrate the device for lesion-assisted localization based on high-quality negative cues provided in this application.

[0061] like Figure 2 As shown, in a second aspect of the embodiments of this application, a medical image lesion-assisted localization device based on high-quality negative cues is provided, comprising: Image acquisition module 210 is used to acquire medical images; The initial mask generation module 220 is used to generate an initial mask for the lesion region based on bounding box prompts; The negative prompting generation module 230 is used to extract the regional features of the lesion area and the regional features of the background area surrounding the lesion area based on the initial mask, and generate high-quality negative prompting points based on the similarity analysis between the regional features of the lesion area and the regional features of the background area. Positive prompt generation module 240 is used to obtain positive prompt points within the lesion area; The segmentation module 250 is used to input positive cue points, high-quality negative cue points, and bounding box cue points into the visual basic segmentation model to generate the first lesion mask. The mask fusion module 260 is used to fuse the initial mask with the first lesion mask to obtain the final lesion localization result of the medical image.

[0062] The medical image lesion-assisted localization device based on high-quality negative cues in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0063] The medical image lesion-assisted localization device based on high-quality negative cues in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0064] The medical image lesion-assisted localization device based on high-quality negative feedback provided in this application embodiment can achieve... Figure 1The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0065] Optionally, such as Figure 3 As shown, this application embodiment also provides an electronic device 300, including a first processor 301, a first memory 302, and a program or instructions stored in the first memory 302 and executable on the first processor 301. When the program or instructions are executed by the first processor 301, they implement the various processes of the above-described embodiment of the medical image lesion-assisted localization method based on high-quality negative cues, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0066] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0067] Figure 4 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application.

[0068] The hardware structure 400 of the electronic device includes, but is not limited to, the following components: radio frequency unit 401, network module 402, audio output unit 403, input unit 404, sensor 405, display unit 406, user input unit 407, interface unit 408, second memory 409, and second processor 410.

[0069] Those skilled in the art will understand that the hardware structure 400 of the electronic device may also include a power supply (such as a battery) for supplying power to various components. The power supply can be logically connected to the second processor 410 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0070] It should be understood that, in this embodiment, the input unit 404 may include a graphics processing unit (GPU) 4041 and a microphone 4042. The GPU 4041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 406 may include a display panel 4061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, etc. The user input unit 407 includes a touch panel 4071 and other input devices 4072. The touch panel 4071 is also called a touch screen. The touch panel 4071 may include a touch detection device and a touch controller. Other input devices 4072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, joysticks, etc., which will not be described in detail here. The second memory 409 can be used to store software programs and various data, including but not limited to applications and operating systems. The second processor 410 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understandable that the aforementioned modem processor may not be integrated into the second processor 410.

[0071] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiments of the medical image lesion-assisted localization method based on high-quality negative cues, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0072] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0073] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described embodiments of the medical image lesion-assisted localization method based on high-quality negative cues, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0074] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0075] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0076] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0077] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for lesion-assisted localization in medical images based on high-quality negative cues, characterized in that, include: Acquire medical images and generate an initial mask for the lesion region based on bounding box cues; Based on the initial mask, the regional features of the lesion area and the regional features of the background area surrounding the lesion area are extracted. Based on the similarity analysis between the regional features of the lesion area and the regional features of the background area, high-quality negative cue points are generated. Obtain positive cue points within the lesion region, and input the positive cue points, the high-quality negative cue points, and the bounding box cue points into the visual basic segmentation model to generate a first lesion mask; The initial mask is fused with the first lesion mask to obtain the final lesion localization result of the medical image.

2. The method for lesion-assisted localization in medical images based on high-quality negative cues according to claim 1, characterized in that, The generation of high-quality negative warning points based on the similarity analysis between the regional features of the lesion region and the regional features of the background region specifically includes: The medical image is input into a visual feature extraction model to extract deep semantic features of the medical image; Based on the initial mask, extract the regional features of the lesion region and the regional features of the background region from the deep semantic features, respectively; Calculate the first similarity between each pixel in the medical image and the regional features of the lesion region to obtain a confidence map of the lesion region; Calculate the second similarity between each pixel in the medical image and the regional features of the background region to obtain a background region confidence map; Based on the background region confidence map, high background confidence regions are selected as candidate negative cue regions within the boundary neighborhood of the lesion region. The high-quality negative cue points are determined from the candidate negative cue regions.

3. The method for assisted lesion localization in medical images based on high-quality negative cues according to claim 2, characterized in that, The step of determining the high-quality negative suggestion point from the candidate negative suggestion region specifically includes: Cluster analysis is performed on the candidate negative suggestion regions to divide them into multiple local regions; The center position of each of the aforementioned local regions is selected as the high-quality negative prompt point.

4. The method for lesion-assisted localization in medical images based on high-quality negative cues according to claim 2, characterized in that, The step of selecting high background confidence regions as candidate negative cue regions within the boundary neighborhood of the lesion region based on the background region confidence map specifically includes: The boundary of the lesion region is determined based on the initial mask; With the boundary as the center, a preset range is extended towards the background area to determine the spatial range of the boundary neighborhood; Within the boundary neighborhood, regions in the background confidence map with a background confidence level higher than a preset threshold are selected as candidate negative cue regions.

5. The method for assisted lesion localization in medical images based on high-quality negative cues according to claim 1, characterized in that, The acquisition of positive indicator points within the lesion area specifically includes: Based on the confidence map of the lesion region, regions within the lesion region whose lesion confidence is higher than a preset threshold are selected. Cluster analysis was performed on the selected region, dividing it into multiple local regions; The center position of each of the aforementioned local regions is selected as the positive prompt point.

6. The method for assisted lesion localization in medical images based on high-quality negative cues according to claim 1, characterized in that, The step of fusing the initial mask with the first lesion mask to obtain the final lesion localization result of the medical image specifically includes: Calculate the degree of overlap between the initial mask and the first lesion mask; The consistency between the initial mask and the first lesion mask is determined based on the degree of overlap; When the consistency exceeds a preset threshold, the overlapping area between the initial mask and the first lesion mask is retained, and the difference area between the initial mask and the first lesion mask is supplemented and fused. When the consistency is lower than a preset threshold, the confidence weight of the difference region is reduced. The final lesion localization result is generated based on the fusion result.

7. The method for lesion-assisted localization in medical images based on high-quality negative cues according to claim 6, characterized in that, The process of generating the final lesion localization result based on the fusion result specifically includes: The merged mask region is evaluated for regional confidence, and regions with confidence scores higher than the preset confidence threshold are retained. The final lesion localization result is generated based on the preserved region.

8. A medical image lesion localization aid based on high-quality negative cues, characterized in that, include: The image acquisition module is used to acquire medical images; The initial mask generation module is used to generate an initial mask for the lesion region based on bounding box prompts. The negative prompt generation module is used to extract the regional features of the lesion area and the regional features of the background area surrounding the lesion area based on the initial mask, and generate high-quality negative prompt points based on the similarity analysis between the regional features of the lesion area and the regional features of the background area. A positive feedback generation module is used to obtain positive feedback points within the lesion area; The segmentation module is used to input the positive cue points, the high-quality negative cue points, and the bounding box cue points into the visual basic segmentation model to generate a first lesion mask. The mask fusion module is used to fuse the initial mask with the first lesion mask to obtain the final lesion localization result of the medical image.

9. An electronic device, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the medical image lesion-assisted localization method based on high-quality negative cues as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the medical image lesion-assisted localization method based on high-quality negative cues as described in any one of claims 1-7.

Images