Alzheimer’s and Brain Disorder Detection Using TLEABLCNN with SMOTE-Augmented Imbalanced MRI for Enhanced Diagnostic Accuracy

Abstract

Alzheimer's disease (AD) and brain tumors are two of the most serious neural diseases that need to be accurately diagnosed with magnetic resonance imaging (MRI) in order to be treated effectively. This study looks at MRI datasets that aren't balanced and uses the Synthetic Minority Oversampling Technique (SMOTE) to make classification work better. A number of machine learning and deep learning models were used, such as Random Forest (RF), Support Vector Machine (SVM), Voting Classifier, Convolutional Neural Network (CNN), VGG16, ResNet50, InceptionV3, AlexNet with LSTM, Xception, and the new Transformer and LSTM Enhanced Attention-Based Explainable CNN (TLEABLCNN). A comparison test with three different situations showed that TLEABLCNN regularly did better than baseline models. It achieved classification accuracy of 96.1% in Scenario A, 97.4% in Scenario B, and 98.2% in Scenario C, with better macro-precision, recall, and F1-scores. Object recognition models YOLOv5, YOLOv8, YOLOv9, and YOLOv11 were used to find brain abnormalities. With a mean average precision (mAP) of 91.5%, YOLOv9 had the best performance, while YOLOv8 had the best accuracy of 91.8%. Also, explainable AI (XAI) methods like Grad-CAM were used to make heatmaps that showed the most important parts of the MRI scans that affected the model's predictions. This made the forecasts easier to understand and increased clinical trust. It's clear from these results that the strategy for correctly identifying neurological disorders is very strong