MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101040 A) filed by Sri Eshwar College Of Engineering on August 20, 2026, for Hybrid Deep Learning Framework For Automated Brain Tumor Detection And Anomaly Identification Using Mri.
Inventors include Dr Thamaraimanalan T; Ms. Sharmila Parveen M; Ms. Mohanapriya K; and Mr Hariharan S.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: The proposed invention reveals a strong deep learning–based hybrid system for computer-aided automatic brain tumor detection and anomaly identification from Magnetic Resonance Imaging (MRI) data, ensuring high diagnostic accuracy, computational efficiency, and clinical flexibility. The system integrates a supervised convolutional neural network (CNN) classifier and an unsupervised autoencoder-based anomaly detector within a unified framework to achieve accurate classification and efficient detection of novel or abnormal cases. Implemented in MATLAB R2023a, the architecture employs a custom 23-layer CNN consisting of convolutional, batch normalization, max-pooling, fully connected, dropout, and softmax layers optimized for classifying glioma, meningioma, pituitary tumor, and normal brain images. The autoencoder component enhances robustness by analyzing reconstruction loss between input and reconstructed MRI scans to detect abnormalities without the need for labeled data. Experimental evaluation using 7,023 MRI images, after preprocessing and augmentation, demonstrated a classification accuracy of 94.66% and an anomaly detection AUC of 0.962, confirming high sensitivity and specificity. Comparative analysis with ResNet50, VGG23, and SVM classifiers established the proposed model’s superior performance, low misclassification rate, and faster convergence. The hybrid two-stage design enables reliable operation in both labeled and unlabeled environments, making it suitable for telemedicine and real-time diagnostic applications. By integrating deep feature extraction and anomaly detection within a common architecture, the invention provides a scalable, interpretable, and efficient solution for early brain tumor diagnosis and intelligent medical image analysis, assisting clinicians in achieving faster and more accurate clinical decisions.
Disclaimer: Curated by HT Syndication.
