Abstract

Info

Title:: Application of YOLOv5 algorithm incorporating channel-spatial adaptive module for automatic recognition and segmentation of intracerebral hemorrhage in CT images

Author(s):: ZHANG Yuhang1; 2; YANG Tao3; LU Xueqi4; FU Liyuan2; HUANG Hao1; 1. Academy of Integrative Medicine, College of Integrative Medicine, Fujian University of Traditional Chinese Medicine, Fuzhou 350122, China 2. Department of Radiology, Fuzong Teaching Hospital (the 900th Hospital), Fujian University of Traditional Chinese Medicine, Fuzhou 350025, China 3. The First Clinical Medical College, Fujian University of Traditional Chinese Medicine, Fuzhou 350122, China 4. School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China

Keywords:: Keywords: cerebral hemorrhage YOLOv5 channel-spatial adaptive module attention mechanism computed tomography image

Abstract:: Abstract: Objective To propose an improved YOLOv5 deep learning algorithm integrated with a channel-spatial adaptive module (CSAM) for enhancing the performance of YOLO-based models in the automatic recognition and segmentation of cerebral hemorrhage lesions in computed tomography (CT) images. Methods Building upon the conventional YOLOv5 architecture, the proposed algorithm simultaneously integrated the CSAM at the terminal stages of both the Backbone and Neck components. A cerebral hemorrhage CT image dataset sourced from the public Kaggle platform was employed for model training and validation. Results When IoU-T was set to 0.6, the YOLOv5-CSAM algorithm achieved a mean average precision of 0.985 for the recognition and segmentation of cerebral hemorrhage lesions in CT images, outperforming both the baseline YOLOv5 algorithm and other existing improved YOLOv5 variants. Conclusion The YOLOv5-CSAM algorithm demonstrates superior performance in cerebral hemorrhage lesion identification and segmentation in CT images, while exhibiting stronger anti-interference capability and favorable clinical adaptability.