Application and Improvement of YOLO11 for Carotid Plaque Detection in Handheld Ultrasound Images.
Authors
Affiliations (2)
Affiliations (2)
- Department of Medical Imaging, The Affiliated Hospital of Yangzhou University, Yangzhou, China.
- Department of Medical Imaging, The Affiliated Hospital of Yangzhou University, Yangzhou, China. [email protected].
Abstract
Early and accurate detection of carotid plaques is clinically significant for stroke risk assessment and primary care screening. To address challenges such as low resolution, small target size, and blurred boundaries in handheld ultrasound images of carotid plaques, this paper proposes an improved YOLO11n model named YOLO11-MGD. The model introduces the C2PSA_Mona module to suppress interference from normal anatomical structures and incorporates the C3k2_GCConv module to enhance global context modeling capabilities, thereby alleviating semantic fragmentation caused by heterogeneous plaques. Additionally, it integrates a Dynamic Head to achieve joint dynamic attention allocation across scale, space, and task dimensions, and it designs the Inner_MPDIoU loss function to improve bounding box regression accuracy for irregular plaques. Experimental results show that on a self-constructed dataset containing 2078 carotid plaque images, YOLO11-MGD achieves a mean mAP50 of 93.16%, outperforming several mainstream object detection models. While maintaining its lightweight advantage, the proposed method significantly enhances the accuracy and robustness of carotid plaque detection in low-quality ultrasound images and demonstrates good application value for primary care stroke risk screening.