An Explainable and Class-Balanced Deep Learning Framework for Multi-Class Alzheimer's Disease Detection using MRI Scans

Bolukbasi Z., Yurtoglu H., Kaplan R. D., Yetgin D., Kuğu E.

10th International Conference on Computer Science and Engineering, UBMK 2025, İstanbul, Turkey, 17 - 21 September 2025, pp.935-940, (Full Text) identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/ubmk67458.2025.11206828
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.935-940
  • Keywords: Alzheimer's Disease, Deep Learning, Explainable AI, SMOTETomek
  • TED University Affiliated: Yes

Abstract

Alzheimer's Disease (AD) is one of the most common neurodegenerative diseases and early diagnosis is essential for decelerating its advancement. A Deep Learning (DL) based architecture is introduced for the classification o f A D stages using brain MRI scans. Three different models Custom Convolutional Neural Network (CNN), VGG16, and ResNet50 were implemented and compared in terms of accuracy, precision, recall, and F1-score. To handle class imbalance, SMOTETomek was applied to resample the dataset. Among the models, the VGG16-based transfer learning model delivered the highest accuracy of 0.98. Grad-CAM, a visual explainability technique, was used to interpret the VGG16 model's decisions. The heatmaps generated by Grad-CAM highlighted brain regions consistent with medical knowledge, supporting the reliability of model predictions. This study shows that combining DL with explainable artificial i n telligence ( X AI) m e thods c a n support more accurate and trustworthy AD diagnosis.