American
International
University-
Bangladesh

Students & Supervisors

Abstract

Accurate brain tumor categorization by magnetic resonance imaging (MRI) is essential for early identification and therapy planning. Traditional deep learning algorithms might be challenging to utilize in clinical settings due to varying patient anatomy and interpretability issues. This study presents a new approach to brain tumor classification that combines Template Registration (TR) preprocessing and a customized TRIncepResNet architecture. Template registration reduces spatial variance in MRI data, enabling the model to concentrate on tumor-specific properties. The proposed model uses Inception modules for multi-scale feature extraction and residual connections, resulting in consistent training and performance. Tested on a balanced dataset of 10,000 improved MRI images from four classes (glioma, meningioma, pituitary tumor, and no tumor), the proposed technique achieves 99.6% accuracy and an AUC of 1.000, exceeding ResNet-based baselines. Grad-CAM++ is also used to improve interpretability by creating visual heatmaps that corroborate the model's emphasis on clinically significant tumor locations. These findings validate TRIncepResNet as a cutting-edge, interpretable, and therapeutically promising approach for automated brain tumor classification. Future work will expand the framework to include 3D and multi-modal MRI data, as well as evaluate its effectiveness in multi-center clinical trials.

Keywords

Publication Details

• DOI: N/A
• Type of Publication:
• Conference Name: ICCIT 2025 28th INTERNATIONAL CONFERENCE ON COMPUTER AND INFORMATION TECHNOLOGY
• Date of Conference: 19/12/2025 - 19/12/2025
• Venue: Long Beach Hotel, Cox's Bazar, Bangladesh
• Organizer: IEEE Bangladesh Section