NeuroSense: A Radiomics-Enhanced Hybrid Deep Learning Framework for Brain Tumor Classification Integrated with Multitask Clinical Prediction

View at publisher Summarise with VIRAT
Authors: Dinesh Karthik. S, Ariharan. R, K Sanjay, Gowri S
Cited by: 0
References: 23

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Magnetic resonance imaging (MRI) plays a major role in identifying brain tumors at an early stage, but the manual reading of these scans is often slow and may vary from one clinician to another. Many existing deep learning approaches concentrate only on a single task, such as basic classification, and therefore miss important radiological clues that are useful in real clinical settings. In this work, a complete multi-task framework is introduced to analyse MRI scans in a more practical way. The proposed model is designed to perform three related tasks at the same time-identifying the tumor type, estimating its severity, and suggesting a suitable treatment option. To achieve this, the system combines features learned from convolutional neural networks with hand-crafted radiomic measurements such as texture patterns extracted from graylevel co-occurrence matrices. To improve transparency, the system includes an explainability module based on Grad-CAM that visually highlights the regions of the MRI scan that influence the model's predictions. Alongside this, a straightforward lesion-measurement approach is applied to estimate the proportion of brain tissue that appears abnormal, providing additional support for the severity assessment. The framework also automatically generates a clinical-style summary report containing relevant images, prediction probabilities, and short interpretive notes. Experiments conducted on MRI data obtained from multiple sources indicate that the model performs reliably across all three diagnostic tasks, while remaining interpretable and suitable for use in real-world clinical settings. Overall, the system aims to provide a more transparent, reliable, and clinically helpful tool for automated brain tumor assessment.

Fields of Study & Subjects

Brain Tumor Detection and Classification Radiomics and Machine Learning in Medical Imaging Glioma Diagnosis and Treatment

Publication details

Journal / source
International Conference on Intelligent Sustainable Systems (ICISS)
Pages
476–482
Publication date
4 March 2026
Language
EN