MUMBAI, India, July 13 -- Intellectual Property India has published a patent application (202621067634 A) filed by Parul University Parul Institute Of Engineering And Technology on May 29, 2026, for Multi-Encoder Deep Learning System For Multi-Modal Mri Tumor Segmentation.
Inventors include Sonia Flora; and Dr Amit Ganatra.
The application for the patent was published on July 10, 2026, under issue no. 28/2026.
Abstract: A medical image segmentation system for multi-modal MRI tumor segmentation comprises a plurality of modality-specific encoder networks based on Swin Transformer architecture, each independently processing a corresponding MRI modality input (T1, T1ce, T2, FLAIR) to generate modality-specific feature representations. A feature fusion module integrates the modality-specific feature representations via concatenation along a channel dimension to generate a unified multi-modal feature representation. A decoder network generates a segmentation map from the unified multi-modal feature representation, delineating tumor regions. A Convolutional Block Attention Module (CBAM) positioned downstream of the decoder applies channel-wise attention and spatial attention to refine the segmentation map, enhancing boundary delineation and suppressing background noise. The system captures complementary information from multiple MRI sequences through independent parallel encoding, avoiding conflation of modality- specific features. The post-decoder CBAM attention mechanism improves segmentation accuracy for tumors of varying sizes and morphologies compared to single-encoder architectures.
Disclaimer: Curated by HT Syndication.
