MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641092984 A) filed by Dr. R. Dhanalakshmi; and Gokulakrishnan. S on July 31, 2026, for Multi-Scale Hierarchical Attention Network With Self-Supervised Pre-Training For Cross-Domain Histopathological Image Diagnosis.
Inventors include Dr. R. Dhanalakshmi; and Gokulakrishnan. S.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: ABSTRACT A Multi-Scale Hierarchical Attention Network (MHAN) system and method for histopathological image diagnosis is disclosed. The system comprises a plurality of scale-specific processing branches (103, 104, 105, 106), each associated with a scale-specific attention module (107), configured to process a histopathological image at a plurality of magnification levels, and a cross-scale attention fusion module (108) configured to combine the resulting attention-weighted feature maps into a fused multi-scale representation for diagnostic classification (109, 110). The said network is pre-trained using a self-supervised pre-training framework comprising a spatial context prediction pretext task (203) and a morphological consistency pretext task (204) performed on unlabelled whole-slide images, thereby reducing dependency on labelled training data. A visualization module (306) further generates scale-specific and fused attention-based saliency maps, providing interpretability of diagnostic decisions at both ti
Disclaimer: Curated by HT Syndication.
