1Department of Artificial Intelligence & Data Science, Datta Meghe College of Engineering, Navi Mumbai, India
2Department of Computer Science Engineering, IoT Lokmanya Tilak College of Engineering, Navi Mumbai, India
3Department of Electronics and Communication Engineering, CMR Institute of Technology, Bangalore, India
4Department of Electronics and Telecommunication Engineering, Lokmanya Tilak College of Engineering, Navi Mumbai, India
5Department of Artificial Intelligence, Vishwakarma University, Pune, India
6Department of Electronics and Telecommunication Engineering, MH Saboo Siddik College of Engineering, Mumbai, India
Accurate detection of brain tumors from magnetic resonance imaging (MRI) remains a clinically critical yet computationally challenging task due to high-dimensional image complexity and sensitivity to diagnostic errors. This study presents a unified hybrid diagnostic framework that systematically integrates classical machine learning algorithms and deep neural architectures within a standardized and reproducible benchmarking environment. Unlike model-centric investigations that evaluate isolated classifiers, the proposed framework establishes a controlled cross-paradigm experimental ecosystem in which regression-based, probabilistic, distance-driven, shallow neural, and convolutional models operate under identical preprocessing, validation, and testing protocols. Beyond comparative performance analysis, interpretability is incorporated through Gradient-weighted Class Activation Mapping (Grad-CAM), enabling visualization of spatial attention patterns underlying convolutional predictions. A multi-metric evaluation strategy including accuracy, precision, recall, and F1-score provides comprehensive assessment of diagnostic reliability. Experimental results demonstrate a consistent performance hierarchy, with convolutional neural networks achieving superior discriminative capability and improved tumor sensitivity relative to classical approaches. By combining standardized benchmarking, interpretability integration, statistical validation, and deployment-aware evaluation, the proposed framework contributes a reproducible methodological reference for evidence-guided algorithm selection in medical imaging. The study advances transparent and clinically aligned artificial intelligence for MRI-based brain tumor detection.
Keywords: Brain tumor classification, Radiomic features, Convolutional neural networks, MRI analysis, Explainable artificial intelligence, Hybrid diagnostic modeling.
How to cite this article: Jain P, Jain N, Panchikattil S, Chikhale D, Jadhav JS, Sankpal A. A unified and interpretable benchmarking framework for brain tumor classification using classical and deep learning models. Int J Drug Deliv Technol. 2026;16(8s): 1-6; DOI: 10.25258/ijddt.16.8s.1
Source of support: None.
Conflict of interest: None