Scholarlink

<p>In the medical field, detecting benign or malignant endometrial lesions is of paramount importance. However, hysteroscopic images are characterized by uneven illumination, bloodstain interference, and significant variations in lesion scales, which hinder existing detection algorithms from achieving precise detection. To further enhance detection performance, this paper proposes a high-precision detection model, YOLOv11-LC. We designed the LCA (Laplacian Coordinate Attention) module, which constructs a residual Laplacian matrix by computing similarity distances in feature space and channels, along with incorporating coordinate information of features. This enhances the aggregation of similar features while suppressing dissimilar ones, enabling the model to focus more on critical regions and improving detection accuracy. Additionally, this paper introduces the CMA_C3k2 module, which employs Conditional Convolution (CondConv) to dynamically adjust convolution kernel weights based on input features, adapting to lesions with varying scales and textures. Combined with multi-scale attention fusion, integrating spatial and channel attention, it enhances focus on small lesion targets while preserving global contextual information. Compared to YOLOv11 on the HS-CMU dataset, YOLOv11-LC improves mAP50 by 1.6%, achieving 0.815. Furthermore, it outperforms other YOLO-series detection models in comparative evaluations, demonstrating significant value for medical imaging scenarios requiring high-precision localization and classification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

YOLOv11-LC: accurate detection of benign and malignant endometrial lesions using LCA-enhanced YOLOv11 with CMA-C3k2

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • Jie Zhang,
  • Xiao Ling Peng

摘要

In the medical field, detecting benign or malignant endometrial lesions is of paramount importance. However, hysteroscopic images are characterized by uneven illumination, bloodstain interference, and significant variations in lesion scales, which hinder existing detection algorithms from achieving precise detection. To further enhance detection performance, this paper proposes a high-precision detection model, YOLOv11-LC. We designed the LCA (Laplacian Coordinate Attention) module, which constructs a residual Laplacian matrix by computing similarity distances in feature space and channels, along with incorporating coordinate information of features. This enhances the aggregation of similar features while suppressing dissimilar ones, enabling the model to focus more on critical regions and improving detection accuracy. Additionally, this paper introduces the CMA_C3k2 module, which employs Conditional Convolution (CondConv) to dynamically adjust convolution kernel weights based on input features, adapting to lesions with varying scales and textures. Combined with multi-scale attention fusion, integrating spatial and channel attention, it enhances focus on small lesion targets while preserving global contextual information. Compared to YOLOv11 on the HS-CMU dataset, YOLOv11-LC improves mAP50 by 1.6%, achieving 0.815. Furthermore, it outperforms other YOLO-series detection models in comparative evaluations, demonstrating significant value for medical imaging scenarios requiring high-precision localization and classification.