MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641074257 A) filed by Srm Institute Of Science And Technology, Ramapuram Campus; and Easwari Engineering College on June 15, 2026, for Skin Cancer Detection Using Deep Learning: A Comprehensive Study On Efficientnet-Based Models With Explainability And Clinical Evaluation.
Inventors include Dr. Rama Chaithanya Tanguturi; Dr. S. Sajini; Dr. Sankar Ram; Dr. Sandhya; Dr. G. Deena; Dr. Mohan; Mr. Sree Varshan; and Mr. Piyush.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: Abstract: Skin cancer, encompassing melanoma and non-melanoma types, represents a formidable global public health challenge with rising incidence rates. Early and accurate detection is the critical determinant of patient prognosis. This invention presents a methodologically rigorous system for automated skin lesion classification using the ISIC and HAM10000 datasets. We propose an EfficientNet-B0 based architecture, selected for its optimal balance of parameter efficiency and predictive power, benchmarking it against a MobileNetV3 baseline. Our methodology integrates extensive ablation studies, examining the impact of data augmentation, loss functions, and transfer learning. Crucially, we address the "black box" nature of deep learning through Explainable AI (XAI) using Gradient-weighted Class Activation Mapping (Grad-CAM) to generate clinically interpretable heatmaps. Furthermore, we implement advanced uncertainty quantification via Monte Carlo Dropout to ensure safe clinical deployment. Empirical results demonstrate that EfficientNet-B0 achieves state-of-the-art performance with 89.7% accuracy on ISIC and 87.9% on HAM10000, significantly outperforming the baseline. The system concludes with a critical analysis of ethical considerations, including dataset bias and the necessity for algorithmic transparency in diverse healthcare settings.
Disclaimer: Curated by HT Syndication.
