Intellectual Property India Publishes Patent Application for Explainable Neuro-Symbolic Meta-Learning-Driven Hybrid Abc-Ssa-Optimized Siamese Cnn-Gnb Framework For Dual-Modality Mri Fusion-Based Parkinson'S Disease Diagnosis

MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202611071867 A) filed by Mr. Sachin Kumar; Dr. Sourabh Shastri; Dr. Rajesh Kumar Chandel; and Prof. Vibhakar Mansotra on June 10, 2026, for Explainable Neuro-Symbolic Meta-Learning-Driven Hybrid Abc-Ssa-Optimized Siamese Cnn-Gnb Framework For Dual-Modality Mri Fusion-Based Parkinson'S Disease Diagnosis.

Inventors include Mr. Sachin Kumar; Dr. Sourabh Shastri; Dr. Rajesh Kumar Chandel; and Prof. Vibhakar Mansotra.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: This invention pertains to an explainable neuro-symbolic meta-learning-driven hybrid ABC-SSA-optimized Siamese CNN-GNB architecture for dual-modality MRI fusion-based PD diagnosis. The proposed system employs T1W-sMRI and rs-fMRI modalities to analyze Parkinson's-related neuroimaging traits. Before using the input data, the latter goes through MRI pre-processing steps such as slice time correction, slice extraction, removing motion artifacts, skull stripping, image normalization, image enhancement, resizing, and labeling the images. A meta-learning approach built on top of Reptile helps develop generic initialization parameters for robust model generalization and adaptation to other datasets. Subsequently, the Siamese CNN is used for feature extraction from both the T1W-sMRI and rs-fMRI, which are later subjected to multimodal fusion. The hybrid ABC-SSA algorithm aids in the learning of optimum parameters for the models, while the GNB model makes the final disease prediction. In order to help understand the predictions made from the disease prediction model clinically, an explainability model based on Grad-CAM offers explanation by way of mean activation, maximum activation, activation standard deviation, and active area percentage. Neuro-symbolic reasoning is used to help infer clinically meaningful decision-making rules from the explanation provided. The framework presented also utilizes the UMAP method to assess the separability of features. From the experiment conducted, it is found that there is excellent diagnostic performance of our framework, where the testing accuracy was 99.38%, the sensitivity was 100.00%, the specificity was 98.75%, and the ROC-AUC value was 99.37%.

Disclaimer: Curated by HT Syndication.