EXPLAINABILITY IN ALZHEIMER'S DISEASE DIAGNOSIS: A DEEP LEARNING APPROACH WITH NEUROIMAGING DATA

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that is the most common form of dementia and one of the leading causes of mortality. AD affects a subject's memory and behaviour; over time, it can severely impair daily functioning. AD research is lacking, so treatments are ineffective, especially in advanced stages. However, early detection is necessary for managing the symptoms and slowing down AD progression. Therefore, it is essential early identify AD. For this reason, artificial intelligence has started to be exploited. Convolutional neural networks (CNNs) have been revolutionary for image-based applications and are widely used for clinical image analysis. In recent years, studies have used 2D CNNs on magnetic resonance imaging (MRI) scans for AD detection. In this dissertation, a CNN is trained over the image slices to carefully distinguish subjects with Alzehimer from cognitively normal ones and then apply Explainable AI to assess prediction quality and find new patterns. The CNN used in this work was VGG-19 pre-trained on an ImageNet dataset. The input data employed were the 2D slices from the OASIS-3 dataset. Several configurations were implemented and compared for AD detection, and the best models were selected based on loss function, accuracy, precision, recall, f1-score, and Explainable AI (XAI). The proposed model achieved the best classification performance with 70\% accuracy with Nifti images and 76\% accuracy with jpeg images. Results show that applying CNN models with transfer learning can improve classification accuracy. Although the model performs better on RGB, the XAI analysis shows better outcomes on nifti images.