MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641118473 A) filed by Sadiqa Kushtar. Kasimsab Attar; Dr. Pavan Narayan Kunchur; Deepa Basavaraj Benni; Fatima Tuz Zohra M Gadwal; Gautami R Kali; and Nisha Karjagi on October 03, 2026, for System For Multi-Scale Ovarian Cancer Subtype Classification Using Vision Transformer And Attention-Based Multiple Instance Learning.

Inventors include Sadiqa Kushtar. Kasimsab Attar; Dr. Pavan Narayan Kunchur; Deepa Basavaraj Benni; Fatima Tuz Zohra M Gadwal; Gautami R Kali; and Nisha Karjagi.

The application for the patent was published on October 09, 2026, under issue no. 41/2026.

Abstract: A computer-implemented system for multi-scale classification of ovarian cancer subtypes from digitized whole-slide histopathological images is disclosed. The system includes a whole-slide image input module, a preprocessing engine for tissue selection, artefact reduction and appearance normalization, and a multi-scale patch generator that produces spatially registered tissue instances at different effective image scales. A vision transformer encoder converts the instances into token-derived embeddings using self-attention, and a cross-scale feature fusion module combines complementary cellular and tissue-architectural representations. An attention-based multiple instance learning aggregator assigns learned contribution weights to fused instances and forms a weighted slide-level representation. A classification head generates a subtype score vector, while an output module maps attention weights to slide coordinates to provide spatially traceable evidence associated with the classification. A training controller may provide case-level partitioning, class-aware optimization, cross-scale consistency and confidence calibration. The architecture enables variable-size whole-slide analysis without requiring exhaustive patch-level subtype labels and supports benchmark evaluation against alternative deep-learning architectures. Accompanied Drawing [FIGS. 1-2].

Disclaimer: Curated by HT Syndication.