MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202641084649 A) filed by Koneru Lakshmaiah Education Foundation; Nagaraju Arumalla; and Veerraju Gampala on July 10, 2026, for Adaptive Neurograph Memory Fusion (anmf): A Self-Evolving Explainable Artificial Intelligence Framework For Brain Tumor Classification Via Dynamic Graph Memory And Physics-Guided Neuromultimodal Learning.
Inventors include Nagaraju Arumalla; and Veerraju Gampala.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The invention provides a computer-implemented artificial intelligence (AI) framework, Adaptive NeuroGraph Memory Fusion (ANMF), to formulate an accurate and explainable brain tumor classification from the specific multimodal MRI. In the architecture, adaptive graph learning, neural memory retrieval, physics-guided attention, multimodal feature fusion and uncertainty-aware decision making are integrated into one framework. Multimodal MRI data is the first preprocessed using skull stripping, bias field correction, denoising, intensity normalization and image registration (first localization of automated tumors). Dynamic NeuroGraph Construction Module (DNGCM) A DNGCM represents anatomical brain regions as graph nodes, with edges in the graph that adapt during inference over time using local representation of features and spatial relationships to capture the inherent heterogeneity among brain function and structure. Historical memory from past predictions is also considered when defining the strength of connections across different regions, together with prediction uncertainty for each region informs how strongly connected the two nodes will be in a specific way. Memory—Neuro One approach we have developed is a Neuro Memory Bank which stores the latent representations of previously diagnosed tumors and allows us to perform continual learning that retrieves clinically relevant cases without needing to completely retrain. They proposed a Physics-Guided Attention Module, that integrates the MRI acquisition parameters (repetition time, echo time, magnetic field strength, voxel spacing and slice thickness) to induce robustness against different scanners. A Confidence Evolution Engine sequentially processes the fused graph, memory, and physics-aware features to progressively refine uncertain predictions and produce an Explainable Decision Graph that explains how a diagnostic decision was made. Conclusion: This proposed framework improves the classification performance, interpretability, continual learning ability and cross-institutional generalizability as requirements for deployment in clinical decision support systems, hospital radiology workflows, telemedicine platforms and intelligent medical imaging applications.
Disclaimer: Curated by HT Syndication.
