Tleablcnn: Brain and Alzheimers Disease Detection Using Attention-Based Explainable Deep Learning and Smote Using Imbalanced Brain Mri

Abstract

Alzheimer's disease (AD) is one of the primary causes of dementia. It degenerates the brain and reduces the activity of individuals by disrupting their memory and physiological functions. A comprehensive examination of specific brain tissue is required in order to accurately diagnose a brain condition using magnetic resonance imaging. The main aim of this research was to develop a rapid and effective technique for detecting healthy persons before the onset of brain tumours, including AD, pituitary tumours, gliomas, and meningiomas. This work presents a lightweight convolutional architecture based on EfficientNet with a Squeeze Attention Block using transfer learning. The proposed approach used lightweight layers with an L2 regularizer, global pooling (2D), and batch normalisation to construct the model, including two dropout layers. This paper utilises the synthetic minority oversampling approach to address the issues of overfitting and imbalanced samples by balancing two significantly imbalanced MRI datasets. We evaluated the efficacy of the minority strategy by doing experiments on both the training set and the entire dataset. The proposed model demonstrates high efficiency, with the risk of overfitting mitigated by applying SMOTE solely to the training data, while the test and validation datasets remain unaffected. The effectiveness of the proposed approach was evaluated using several performance measures and compared with previous published research. We evaluated the proposed approach for explainability via gradient-weighted class activation mapping to comprehend the model's behaviour and its predictions. The proposed framework offers advantages over existing models in terms of computing efficiency, explainability, generalization, and clinical significance.