Lightweight YOLO26n-seg-SAConv model achieves 74.2% mAP50 for automated Genant classification of osteoporotic vertebral fractures

Automated Genant classification of osteoporotic vertebral compression fractures on digital radiographs achieved a mean average precision (mAP50) of 74.2% (3.9% improvement) with a 16.67% reduction in computational load using an improved YOLO26n-seg-SAConv model.

• The model’s computational complexity (FLOPs) was reduced by 16.67%, making it lightweight and suitable for edge deployment.
• Classification performance improved for Type 1 and Type 3 fractures, which typically have fewer training samples.
• The retrospective study used 483 DR images from multiple institutions, divided 8:2 for training and testing with Gaussian noise augmentation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.