Scaphoid Fracture Model Shows Sensitivity to Degradation

Researchers from Chang Gung Memorial Hospital and collaborators (JMIR Med Inform, 2026) evaluated a CNN-based scaphoid fracture detection model's robustness under simulated image perturbations including Gaussian noise, blur, JPEG compression, contrast enhancement, resizing, and geometric offsets. They found Gaussian blur and reduced resolution caused the largest accuracy drops and that PSNR–SSIM correlated strongly with performance, implying stricter image quality control improves diagnostic reliability.