MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202631123737 A) filed by C. V. Raman Global University on October 15, 2026, for Hybrid Deep Learning Framework For Brain Tumor Segmentation And Classification Using U-Net And Efficientnetb0 On Mri Images.
Inventors include Dusari Sandeep; Arpita Gouda; and Dr. Pushpendra Kumar Gupta.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: A hybrid deep learning framework for simultaneous brain tumor segmentation and classification from magnetic resonance imaging (MRI) scans is disclosed. The framework comprises a preprocessing stage that resizes and normalizes input images and applies data augmentation, followed by a shared encoder based on a pretrained EfficientNetB0 network. A U-Net style decoder with attention-enhanced skip connections generates a pixel-level tumor segmentation mask indicating the tumor location. A classification head operating on the same deep features applies global average pooling, fully connected layers and a softmax output to determine the tumor type among glioma, meningioma, pituitary tumor and no tumor, together with a confidence score. The network is trained with a joint loss comprising Dice loss and cross-entropy loss, optimized using AdamW. The invention provides a single, end-to-end and lightweight pipeline that reduces computational redundancy, improves consistency between localization and classification, and supports deployment in resource-constrained clinical environments. In experimental evaluation, the framework reported an accuracy of about 99.67 percent, an F1-score of about 0.9968 and a Dice coefficient of about 0.8689
Disclaimer: Curated by HT Syndication.
