MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641118894 A) filed by Vallurupalli Nageswara Rao Vignana Jyothi Institute Of Engineering And Technology on October 05, 2026, for An Explainable Hybrid Deep Learning System And Method For Brain Tumor Detection And Classification From Mri Images.
Inventor includes Mrs. Anoosha.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: ABSTRACT [0032] A lightweight hybrid system is disclosed for automatic classification of brain magnetic resonance imaging scans into four clinically relevant categories comprising glioma, meningioma, pituitary tumor and absence of tumor. Magnetic resonance images are first collected from a structured multi class collection and subjected to validation, uniform resizing, intensity normalization and controlled augmentation operations that introduce controlled geometric and photometric variations. A pretrained MobileNetV2 network whose convolutional weights remain frozen extracts compact high level feature vectors from each preprocessed scan through global average pooling. The resulting high dimensional representations are then projected by principal component analysis onto a lower dimensional subspace that retains the majority of discriminative variance. [0033] The reduced feature vectors are classified by a support vector machine employing a radial basis function kernel. Concurrently, gradient weighted class activation mapping is applied to the same MobileNetV2 backbone to produce spatial heat maps that highlight the image regions most responsible for the predicted label. The complete pipeline is realized as a web accessible application that accepts a new scan, returns the predicted category together with a confidence score and displays the corresponding explanatory heat map. Experimental evaluation on a publicly available multi class magnetic resonance imaging collection demonstrates that the hybrid arrangement achieves high test accuracy and excellent cross validation stability while requiring only modest computational resources and no graphics processing unit for the core classification path. The combination of efficient deep feature extraction, statistical dimensionality reduction, robust classical classification and visual interpretability yields a practical and transparent decision support tool suitable for research and educational deployment in neurological imaging.
Disclaimer: Curated by HT Syndication.
