Cervical cell automatic detection method based on YOLOv8 model
Through the automatic cervical cell detection method based on the YOLOv8 model, the large kernel separation convolution attention module and image enhancement algorithm are introduced to solve the problems of low efficiency and misdiagnosis and missed diagnosis in traditional methods, realize efficient and accurate cervical cell detection, and improve the quality of cervical cancer screening and the efficiency of medical resource utilization.
Patent Information
- Application Number
- CN202510447328.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-10
- Publication Date
- 2025-09-09
AI Technical Summary
Traditional cervical cell detection methods are inefficient and prone to misdiagnosis and missed diagnoses. Existing computer vision methods have insufficient feature extraction capabilities, limited detection accuracy, and poor real-time performance.
An automatic cervical cell detection method based on the YOLOv8 model is adopted. The large kernel separated convolutional attention module (LSKA) is introduced. Combined with the image enhancement algorithm and bounding box recognition technology, the image quality is improved through multi-angle imaging and preprocessing, and the model feature extraction capability is improved to achieve efficient and accurate recognition of cervical cells.
It significantly improves the accuracy and efficiency of cervical cell detection, reduces false detections and missed detections, reduces computational complexity and memory usage, promotes medical automation and early diagnosis, and improves the reliability and efficiency of cervical cancer screening.
Smart Images
Figure CN120612689A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an automatic cervical cell detection method based on the YOLOv8 model, and specifically to the fields of computer vision, deep learning, and medical image processing, in particular to a cell defect recognition technology used in cervical cell image analysis, pathological diagnosis, and medical automated testing. Background Art
[0002] Traditional cervical cell detection methods rely primarily on microscopic observation and manual annotation by pathologists. This approach is not only inefficient but also prone to misdiagnosis or missed diagnosis due to subjective factors. Furthermore, while early computer vision-based detection methods introduced automation, they often suffered from insufficient feature extraction capabilities, limited detection accuracy, and poor real-time performance when processing cell images.
[0003] This paper proposes an automatic cervical cell detection method based on the YOLOv8 model. This method incorporates a large kernel segregated convolutional attention module (LSKA) into the model head. Through the large kernel segregated convolution and attention mechanism, it can more efficiently extract the multi-scale features of cervical cells, especially the shape and texture characteristics of the cells. This significantly improves the detection accuracy of the model while maintaining low computational complexity.
[0004] This method provides the medical industry with a more efficient and accurate automated detection solution, which can meet the needs of modern medicine for large-scale screening and early diagnosis, and improve the automation level of medical services. Summary of the Invention
[0005] The purpose of the present invention is to provide a method for automatic detection of cervical cells based on the YOLOv8 model.
[0006] The present invention comprises the steps of:
[0007] S1: In high-throughput microscopy cell image acquisition systems, multi-angle shooting and full-view coverage technology enable high-resolution imaging of cells from multiple dimensions, ensuring that every area of the cell surface is fully captured. This multi-angle and full-view imaging method not only captures the two-dimensional morphological information of cells, but also captures their three-dimensional structure and dynamic characteristics.
[0008] S2: Use image enhancement algorithms to process the collected cell images to increase the brightness of the cell images, thereby improving the quality and details of the cell images. This step can significantly improve the quality and recognizability of the cell images by adjusting the brightness, contrast, and details of the images. This preprocessing operation not only enhances the details of the cell images, but also reduces the complexity of subsequent analysis and improves the accuracy and efficiency of the detection algorithm.
[0009] S3: The Large Separable Kernel Attention (LSKA) module improves the YOLOv8 model. This module effectively integrates feature information at different scales, improving the model's feature representation and adaptability to complex backgrounds. Furthermore, the LSKA module enhances the model's preference for target shape, enabling it to excel at recognizing cells of varying shapes and sizes.
[0010] S4: Based on the improved YOLOV8 model, different types of cervical cells are accurately identified, the cell area is identified by the bounding box, and different types of cervical cells are classified according to their characteristics;
[0011] S5: The trained and optimized network, combined with bounding box regression technology, identifies cervical cells and classifies them by outputting category names.
[0012] The object of the present invention is achieved like this:
[0013] This YOLOv8-based automatic cervical cell detection method uses the Zero-DCE unsupervised low-light image enhancement algorithm to effectively improve image brightness, contrast, and color saturation. In cervical cell detection, pathological images may have problems such as uneven lighting and low contrast. Zero-DCE can enhance the dynamic range of the image, making the cell structure clearer, thereby providing a better image foundation for subsequent detection.
[0014] The automatic cervical cell detection method based on the YOLOv8 model adds an LSKA attention module, which makes the model more inclined to focus on the shape features of objects rather than textures. This can effectively improve the detection accuracy of cells, while maintaining high performance, significantly reducing computational complexity and memory usage, and achieving more efficient detection.
[0015] The improved network structure of the automatic cervical cell detection method based on the YOLOv8 model significantly improves the performance and reliability of cervical cell detection by optimizing feature extraction, enhancing the detection capability of small targets and complex backgrounds, improving detection efficiency and real-time performance, enhancing generalization capability, and optimizing the model structure.
[0016] The automatic cervical cell detection method based on the YOLOv8 model not only improves the quality of cervical cancer screening, reduces human errors, promotes the development of medical imaging technology, but also optimizes the utilization efficiency of medical resources, providing strong support for the widespread application of artificial intelligence in the medical field.
[0017] The automatic cervical cell detection method based on the YOLOv8 model mainly includes the following core modules:
[0018] The image acquisition module is used to obtain image data of cervical cell samples in real time. It uses professional, high-precision microscope imaging equipment and a supporting image acquisition system to ensure that key information such as the morphology, texture, and fine structure of cervical cells can be clearly captured. The collected images will be passed as input data to the next processing stage after pre-processing such as noise reduction and enhancement, laying a solid data foundation for subsequent accurate analysis and detection.
[0019] The data preprocessing module is used to systematically optimize the collected cervical cell image data. With the help of advanced image algorithms and processing technologies, through image enhancement algorithms, it can restore the true details of the cells, highlight the cell contours and internal structures, and improve the clarity and recognizability of the image; and normalize the image to unify the size and pixel value range so that the data meets the input requirements of the model, providing high-quality and standardized data input for efficient and accurate detection of the model, greatly improving the stability and reliability of model detection.
[0020] The cell recognition module uses an optimized YOLOv8 model to accurately analyze preprocessed cervical cell images. Leveraging the model's powerful feature extraction and classification capabilities, the module accurately identifies various types of cervical cells, both normal and abnormal, from complex image data. By deeply studying and comparing multi-dimensional features such as cell morphology, size, texture, and the ratio of nucleus to cytoplasm, efficient cell classification and location are achieved. The module's output details the cell's category and location, providing intuitive and critical data support for subsequent diagnosis, significantly improving the automation and diagnostic efficiency of cervical cell testing.
[0021] The method for automatic cervical cell detection based on the YOLOv8 model is described, wherein the cell recognition module uses an improved YOLOv8 convolutional neural network model to perform deep feature extraction on the image.
[0022] The improved YOLOv8 model enhances the model's ability to recognize multi-scale targets by introducing a modular structure, optimizes its ability to capture cell morphology and texture, and improves detection accuracy and robustness.
[0023] The described automatic cervical cell detection method based on the YOLOv8 model adopts the improved CSPDarknet as the backbone network, extracts multi-scale features through deep separable convolution and residual structure, further improves the efficiency and expression ability of feature extraction, provides more efficient and accurate technical support for the early screening and diagnosis of cervical cancer, and helps to improve the efficiency and reliability of pathological analysis.
[0024] The described method for automatic cervical cell detection based on the YOLOv8 model combines the feature pyramid network (FPN) and the path aggregation network (PAN) in its Neck part. By bidirectionally fusing features through upsampling and downsampling, the model can more accurately locate the target boundary when detecting cells, thereby enhancing the detection capability of target cells of different sizes.
[0025] The described automatic cervical cell detection method based on the YOLOv8 model integrates the LSKA attention module into the C2f module of the Neck part. The improved model significantly enhances the ability to capture cervical cell features while maintaining real-time processing capabilities, thereby improving detection accuracy.
[0026] The automatic cervical cell detection method based on the YOLOv8 model can present the detection results to medical staff in real time through a display module or transmit them to a data storage system, intuitively displaying the location and type information of cervical cells, providing auxiliary decision support for clinical diagnosis, and facilitating subsequent processing.
[0027] The advantage of the automatic cervical cell detection method based on the YOLOv8 model is that it can achieve full process automation from sample identification to result output by relying on the optimized network structure and algorithm, greatly improving the detection accuracy, effectively reducing the dependence on professional medical personnel, and efficiently utilizing existing medical resources at a lower cost, thereby facilitating the early and accurate diagnosis of cervical diseases.
[0028] The automatic cervical cell detection method based on the YOLOv8 model has excellent robustness to cope with complex and diverse sample situations, and is therefore suitable for scenarios such as routine physical examinations in hospitals, primary medical diagnosis, and large-scale cervical cancer screening.
[0029] How it works: This network is an improved version of YOLOv8, further optimized based on its original structure, incorporating various techniques to improve the efficiency and accuracy of feature extraction. This network introduces the Large Separable Kernel Attention (LSKA) module within the C2f module. This module extracts rich spatial information through large kernel separable convolution operations. It also incorporates an attention mechanism to weight feature channels, thereby enhancing the model's ability to capture cervical cell features. This improvement significantly improves detection accuracy while maintaining efficient computational performance.
[0030] The preprocessing of cervical cell images uses image enhancement algorithms to improve image quality and enhance the characteristics of cervical cells so that subsequent detection algorithms can identify and analyze them more accurately.
[0031] The preprocessed image is fed into the improved YOLOv8 model. During the feature extraction phase, the improved YOLOv8 model progressively extracts low-level and high-level features from the image. The C2f module in this feature extraction step achieves effective feature extraction and fusion through carefully designed convolutional layers and skip-layer connections. Furthermore, the introduction of the LSKA attention module further enhances the model's focus on key features, improving detection accuracy.
[0032] The LSKA attention module is a large separable kernel attention mechanism that decomposes the 2D convolution kernel of the depth convolution layer into cascaded horizontal and vertical 1-D kernels. This decomposition allows the deep convolution layer with large kernels to be used directly in the attention module without the need for any additional modules, thereby reducing computational complexity and memory usage.
[0033] By adding the LSKA attention mechanism to the C2f module, the model can focus more on key features of cervical cells, such as morphology, texture, and boundaries. By dynamically adjusting the position and size of the bounding box, the model can accurately locate and classify cervical cells, maintaining high detection accuracy and speed even when dealing with multi-scale and densely packed objects.
[0034] The improved YOLOv8 model will output cervical cell detection results, including location, size, and category information. These results will be presented to the user in the form of annotated boxes and category labels for subsequent analysis and processing.
[0035] The beneficial effects produced by the present invention are:
[0036] 1. By introducing the LSKA attention module, the improved YOLOv8 model can more accurately capture the subtle features of cervical cells, such as morphology, texture, and boundaries, thereby significantly improving detection accuracy. This helps reduce false positives and missed detections, and improves the reliability of cervical cancer screening.
[0037] 2. The present invention further optimizes computational efficiency by reducing computational complexity and memory usage. It can process a large number of cervical cell images in a short period of time, speed up the screening process, and provide patients with more timely diagnostic results.
[0038] 3. The present invention reduces the need for manual review and the potential for errors caused by human factors by using automated testing methods. This not only improves work efficiency but also ensures the consistency and objectivity of the testing process.
[0039] 4. By improving detection accuracy and speed, the present invention helps to detect abnormal cells in the early stages of cervical cancer development, thereby taking timely treatment measures and improving patient prognosis.
[0040] 5. By providing fast and accurate screening results, the present invention not only helps medical institutions to achieve efficient allocation of resources, but also promotes the optimization of diagnosis and treatment processes, providing patients with more timely, efficient and high-quality medical services. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:
[0042] Figure 1 It is a system architecture diagram of the present invention;
[0043] Figure 2 This is a schematic diagram of the DCE-net network structure of the image enhancement algorithm;
[0044] Figure 3 This is a schematic diagram of the LSKA attention module;
[0045] Figure 4 is the system performance index of the present invention;
[0046] Figure 5 This is an example schematic diagram of the present invention in actual detection application. DETAILED DESCRIPTION
[0047] The specific embodiments of the present invention are described in further detail below with reference to the accompanying drawings.
[0048] like Figure 1 As shown, the automatic cervical cell detection method based on the YOLOv8 model includes the following steps:
[0049] S1: In high-throughput microscopy cell image acquisition systems, multi-angle shooting and full-view coverage technology enable high-resolution imaging of cells from multiple dimensions, ensuring that every area of the cell is fully captured. This multi-angle and full-view imaging method can not only obtain two-dimensional morphological information of cells, but also capture their three-dimensional structure and dynamic characteristics;
[0050] S2: Use image enhancement algorithms to process the collected cell images to increase the brightness of the cell images, thereby improving the quality and details of the cell images. This step can significantly improve the quality and recognizability of the cell images by adjusting the brightness, contrast, and details of the images. This preprocessing operation not only enhances the details of the cell images, but also reduces the complexity of subsequent analysis and improves the accuracy and efficiency of the detection algorithm.
[0051] S3: The Large Separable Kernel Attention (LSKA) module improves the YOLOv8 model. This module effectively integrates feature information at different scales, improving the model's feature representation and adaptability to complex backgrounds. Furthermore, the LSKA module enhances the model's preference for target shape, enabling it to excel at recognizing cells of varying shapes and sizes.
[0052] S4: Based on the improved YOLOV8 model, different types of cervical cells are accurately identified, the cell area is identified by the bounding box, and different types of cervical cells are classified according to their characteristics;
[0053] S5: The trained and optimized network, combined with bounding box regression technology, identifies cervical cells and classifies them by outputting category names.
[0054] like Figure 2 As shown in the figure, the YOLOv8-based method for automatic cervical cell detection uses a Zero-DCE image enhancement algorithm model consisting of seven convolutional layers and activation functions. During feature extraction, the image width and height remain unchanged; only the dimensions change. The algorithm processes the collected cell images, increasing their brightness and thus improving their quality and detail.
[0055] like Figure 3 As shown in the figure, the automatic cervical cell detection method based on the YOLOv8 model incorporates the LSKA attention mechanism, which greatly reduces the computational complexity and memory requirements by decomposing the traditional two-dimensional weighted convolution kernel into two cascaded one-dimensional separable convolution kernels. The LSKA attention mechanism is shown in the following formula:
[0056]
[0057]
[0058] A C =W 1×1 *Z C (3)
[0059]
[0060] in, Represents the output after depth convolution, H and W represent the height and width of the feature map, A C represents the attention map, Indicates the rounding down operation, d indicates the expansion rate, k is the convolution kernel size, * indicates convolution, represents the Hadamard product, represents the input feature map, and C represents the number of input channels.
[0061] The automatic cervical cell detection method based on the YOLOv8 model mainly includes the following core modules:
[0062] The image acquisition module is used to obtain image data of cervical cell samples in real time. It uses professional, high-precision microscope imaging equipment and a supporting image acquisition system to ensure that key information such as the morphology, texture, and fine structure of cervical cells can be clearly captured. The collected images will be passed as input data to the next processing stage after pre-processing such as noise reduction and enhancement, laying a solid data foundation for subsequent accurate analysis and detection.
[0063] The data preprocessing module is used to systematically optimize the collected cervical cell image data. With the help of advanced image algorithms and processing technologies, through image enhancement algorithms, it can restore the true details of the cells, highlight the cell contours and internal structures, and improve the clarity and recognizability of the image; and normalize the image to unify the size and pixel value range so that the data meets the input requirements of the model, providing high-quality and standardized data input for efficient and accurate detection of the model, greatly improving the stability and reliability of model detection.
[0064] The cell recognition module uses an optimized YOLOv8 model to accurately analyze preprocessed cervical cell images. Leveraging the model's powerful feature extraction and classification capabilities, the module accurately identifies various types of cervical cells, both normal and abnormal, from complex image data. By deeply studying and comparing multi-dimensional features such as cell morphology, size, texture, and the ratio of nucleus to cytoplasm, efficient cell classification and location are achieved. The module's output details the cell's category and location, providing intuitive and critical data support for subsequent diagnosis, significantly improving the automation and diagnostic efficiency of cervical cell testing.
[0065] The method for automatic cervical cell detection based on the YOLOv8 model, wherein the cell recognition module uses an improved YOLOv8 convolutional neural network model to perform deep feature extraction on images.
[0066] The improved YOLOv8 model enhances the model's ability to recognize multi-scale targets by introducing a modular structure, optimizes its ability to capture cell morphology and texture, and improves detection accuracy and robustness.
[0067] The described automatic cervical cell detection method based on the YOLOv8 model adopts the improved CSPDarknet as the backbone network, extracts multi-scale features through deep separable convolution and residual structure, further improves the efficiency and expression ability of feature extraction, provides more efficient and accurate technical support for the early screening and diagnosis of cervical cancer, and helps to improve the efficiency and reliability of pathological analysis.
[0068] The described method for automatic cervical cell detection based on the YOLOv8 model combines the feature pyramid network (FPN) and the path aggregation network (PAN) in its Neck part. By bidirectionally fusing features through upsampling and downsampling, the model can more accurately locate the target boundary when detecting cells, thereby enhancing the detection capability of target cells of different sizes.
[0069] The described automatic cervical cell detection method based on the YOLOv8 model integrates the LSKA attention module into the C2f module of the Neck part. The improved model significantly enhances the ability to capture cervical cell features while maintaining real-time processing capabilities, thereby improving detection accuracy.
[0070] The automatic cervical cell detection method based on the YOLOv8 model can present the detection results to medical staff in real time through a display module or transmit them to a data storage system, intuitively displaying the location and type information of cervical cells, providing auxiliary decision support for clinical diagnosis, and facilitating subsequent processing.
[0071] like Figure 4 As shown in the figure, during the training process, the evaluation indicators such as precision, recall, mAP@50 and mAP@50-95 of the automatic cervical cell detection method based on the improved YOLOv8 model all showed a steady improvement trend, indicating the continuous optimization of the model performance.
[0072] Table 1 Comparison of training results of different models By analyzing the data in the above table, we can see that the recall rate of the improved YOLOv8 model has been improved, and both mAP50 and mAP50-95 have been improved, indicating that the improved YOLOv8 model can accurately predict bounding boxes at a higher IoU threshold.
[0073] In this method for automatic cervical cell detection based on the YOLOv8 model, an image enhancement algorithm is used in the preprocessing step to optimize the original image. By adjusting the image's brightness, contrast, and color balance, the detailed features of cervical cells are enhanced, while the texture and boundaries of abnormal cells are highlighted, improving the overall image quality. After image enhancement, the image is standardized to ensure consistency in feature expression across images, providing high-quality input for subsequent model training and inference, further improving the accuracy and reliability of cervical cell detection.
[0074] This automated cervical cell detection method, based on the YOLOv8 model, uses target detection technology to efficiently identify and locate abnormal cell areas in cervical cell images. Furthermore, combined with attention mechanism optimization and post-processing techniques, this method further improves detection accuracy and stability, effectively reducing false positives and missed detections, and providing strong support for accurate cervical cell detection.
[0075] The automatic cervical cell detection method based on the YOLOv8 model is integrated into the medical diagnostic system after optimized training. It can analyze and detect cervical cell images in real time, quickly identify and locate abnormal cells, provide efficient and accurate diagnostic support for early screening of cervical cancer, and improve the quality of medical services and disease prevention and control.
Claims
1. A method for automatic cervical cell detection based on an improved YOLOv8 model, characterized in that: The method comprises the following steps: S1: In the high-throughput microscope cell image acquisition system, multi-angle shooting and full-view coverage ensure that every area of the cell surface can be fully captured, thereby achieving complete acquisition of cell information; S2: Use image enhancement algorithms to process the collected cell images to increase the brightness of the cell images, thereby improving the quality and details of the cell images; S3: The Large Separable Kernel Attention (LSKA) module is introduced to improve the YOLOv8 model. The improved model can effectively integrate feature information of different scales and improve the feature expression ability of the model. S4: Based on the improved YOLOV8 model, different types of cervical cells are accurately identified, the cell area is identified using a bounding box, and different types of cervical cells are classified according to their characteristics; S5: The trained and optimized network, combined with bounding box regression technology, identifies cervical cells and classifies them by outputting category names.
2. The method according to claim 1, characterized in that The acquisition system uses high-throughput image acquisition technology to quickly acquire a large amount of high-resolution image data, providing a rich information basis for subsequent image analysis, processing and application.
3. The method according to claim 1, characterized in that The image enhancement algorithm adopts the Zero-DCE (ZeroReference Deep Curve Estimation) algorithm, which eliminates noise and artifacts in the image by using parameter-free convolution, thereby improving the quality and details of the image.
4. The method according to claim 1, wherein The cervical cells are divided into eight categories: abnormal keratinization cells in high-grade squamous intraepithelial lesions, abnormal keratinization cells in squamous cell carcinoma, intermediate squamous cells, metaplastic cells, mild koilocytes, parabasal cells, severe koilocytes, and superficial squamous cells.
5. The method according to claim 1, wherein The improved YOLOV8 model introduces the large separable kernel attention mechanism LSKA, which weights the original features through the attention map, highlights important features and suppresses unimportant features, thereby improving the model's detection accuracy for cervical cells.