FLFSI:Feature-Level Image Fusion and Improved Swarm Intelligence Optimization Algorithm for Endometrial Detection in Contrast-Enhanced Ultrasound Imaging

The detection of endometrial regions is crucial for early lesion diagnosis, yet existing models face challenges like redundant detection and high false-positive rates. To address these issues, we propose FLFSI, which enhances the YOLO11 model’s performance in detecting endometrial regions in transvaginal ultrasound contrast images. By extending 2DPCA to RGB channels and optimizing feature vector selection with an improved BPSO algorithm, FLFSI reconstructs images to improve rendering and highlight key features. Experimental results demonstrate that FLFSI achieves 2.8% improvement in Box mAP50-95 and 1.9% improvement in Mask mAP50-95 compared to the baseline, offering enhanced detection accuracy for endometrial regions.