Performance Disparities in Very Deep Super-Resolution Learning for Malignant versus Benign Breast Histopathological Images

Abstract

VDSR networks enhance breast histopathological image resolution, but performance is significantly higher for malignant (PSNR=38dB, SSIM=0.992) than benign lesions (PSNR=36dB, SSIM=0.980), necessitating optimization. Prospective studies on segmentation and classification pipelines will support scalable AI-assisted pathology.