MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641075269 A) filed by Vellore Institute Of Technology on June 18, 2026, for Ai-Based Explainable Framework For Skin Cancer Classification Using Resnet50 And Grad-Cam With Imbalance-Aware Learning.
Inventors include Priyanka N; Dr. Arumuga Arun R; Akhil Kanagabalan; Aravind Ajay; and M Vishnu.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: ABSTRACT The present invention discloses an integrated imbalance-aware explainable deep learning framework for automated skin cancer detection and classification from dermoscopic or clinical skin lesion images. The framework employs a refined ResNet50-based transfer learning architecture (102) configured for multi-class skin lesion classification. To address severe class imbalance in dermoscopic datasets, the system incorporates weighted cross-entropy loss optimization and weighted random sampling techniques during the training phase, thereby improving balanced accuracy and enhancing sensitivity toward minority malignant lesion classes including melanoma. The invention further integrates Gradient-weighted Class Activation Mapping (Grad-CAM) (105) during the inference phase to generate clinically understandable visual explanations highlighting lesion-specific discriminative regions influencing prediction outcomes. The unified integration of imbalance correction mechanisms and explainability techniques improves reliability, transparency, and clinical interpretability of automated skin cancer detection systems. The invention thereby provides a reproducible and computationally efficient framework suitable for medical image analysis, computer-aided diagnosis, and practical clinical deployment.
Disclaimer: Curated by HT Syndication.
