Extraction and position fusion of abdominal aortic lipid plaques in rabbits based on photoacoustic-ultrasound dual-modality

Photoacoustic-ultrasound dual-modality endoscopic imaging compensates for the limitations of single-modality techniques in visualizing intravascular lipid plaques. In this study, the technique is applied specifically to the abdominal aorta of rabbits, a well-established animal model for atherosclerosis research, aiming to enhance detection accuracy and analytical robustness for vascular lesions. A multimodal fusion approach is proposed, integrating the ultrasonic VH-IVUS feature extraction with K-means clustering and a photoacoustic-based YOLOv11-seg model for plaque detection and localization. Initially, vessel wall segmentation is performed on ultrasound images, followed by plaque classification using K-means clustering based on grayscale and frequency features derived from VH-IVUS-seg. Clustering performance is validated using silhouette coefficients, and Davies–Bouldin indices, demonstrating high cluster compactness and separability. Concurrently, the YOLOv11-seg model is employed to detect plaque regions in photoacoustic images, achieving a detection precision of 96.8%. The detection results from both imaging modalities are then fused based on spatial correlation and detection consistency, enabling more robust and comprehensive plaque localization. Experimental results in rabbit models demonstrate that the fused imaging data significantly improve plaque detection stability and spatial accuracy compared to single-modality methods. The proposed method, combining YOLOv11-seg with VH-IVUS-based ultrasound clustering, presents a high-precision, multimodal imaging strategy for lipid plaque detection in the rabbit abdominal aorta, offering valuable insights for preclinical atherosclerosis research and its potential future clinical translation.