Brain MRI Analysis Using CNN-Based Feature Extraction and Machine Learning Techniques to Diagnose Alzheimer's Disease

This research focuses on developing a complex framework for automatic diagnosis purposes of Alzheimer's disease from MRI brain data, with emphasis placed on advancement in accuracy towards early detection, using CNN-based feature extraction and machine learning classifiers. The models were trained, validated, and tested on two public accessible datasets: ADNI and MIRIAD. Prior to scaling the MRI images for the ResNet-50 architecture, preprocessing methods such as noise reduction and skull stripping were applied. The paper has analyzed the data gathered using three classifiers: Random Forest (RF), Support Vector Machine (SVM), and SoftMax. The experimental result finds ResNet50-Softmax outperforming other classifiers. Its accuracy was at level 98% along with specificity at 99%, sensitivity at 98%, and F-measure at 99% on the ADNI dataset. Other impressive performances were witnessed on MIRIAD dataset with accuracy 98%, specificity 97%, sensitivity 99% and 98% F-measure. These results, combined with ResNet-50, prove the resilience of the SoftMax classifier for accurate AD classification. Overall findings in this study show that the proposed CNN-based method can serve as a viable avenue to automatically and accurately diagnose Alzheimer's, thus aiding its treatment and early intervention.