STCNN: Combining SMOTE-TOMEK With CNN for Imbalanced Classification of Alzheimer's Disease

Abstract

The most frequent cause of dementia worldwide is Alzheimer's disease (AD). Progressing from mild to severe, it gradually deteriorates, making independent tasks more challenging. Due to the aging population and the timing of diagnoses, its prevalence has exceeded expectations. Existing models for categorizing cases include magnetic resonance imaging (MRI), cognitive testing, and medical history. However, these methods lack precision and sensitivity and are not always effective. A framework for identifying particular features of AD from MRI images is developed using the convolutional neural network (CNN). To prevent the issue of class imbalance, the synthetic minority oversampling technique is used, which exists in the MRI image dataset from Kaggle. An STCNN model is proposed to predict the different dementia stages from MRI and achieves 99.36% and 99% accuracy and F1-score, respectively. We compared the proposed model with the benchmark models and discovered that the STCNN model outperformed the state-of-the-art models in terms of accuracy, efficiency, and performance.