Title: TLEABLCNN: Brain Disorder Detection and Alzheimer's Using Explainable Attention-Based Deep Learning and SMOTE on Imbalanced MRI Data
Authors: Salah Aldin Shadi Aldaya
Duha Khalil Ahmed AlDaya
Samy S. Abu-Naser
Volume: 9
Issue: 12
Pages: 120-135
Publication Date: 2025/12/28
Abstract:
Alzheimer's disease (AD) is a major cause of dementia, and its early diagnosis using MRI is crucial, particularly at the mild cognitive impairment (MCI) stage. However, deep learning-based diagnosis is often hindered by limited data and severe class imbalance in medical imaging datasets.This study proposes TLEABLCNN, a lightweight transfer learning-based explainable attention convolutional neural network for early AD detection using imbalanced brain MRI data. The framework integrates a pre-trained EfficientNet backbone with attention mechanisms and lightweight layers to enhance discriminative feature learning while maintaining computational efficiency. To address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) is applied only to the training set on extracted deep features, ensuring unbiased evaluation. The model is evaluated on MRI data comprising normal control (NC), early MCI (EMCI), late MCI (LMCI), and AD classes, achieving strong classification performance across multiple metrics. Grad-CAM is employed to provide visual explainability by highlighting clinically relevant brain regions. Comparative results demonstrate that TLEABLCNN outperforms existing methods in terms of accuracy, generalization, and interpretability, making it a promising tool for early Alzheimer's disease detection.