Author(s) :

Joshua Asirvatham.

Volume/Issue :

Abstract :

Early detection of neurodegenerative diseases can be challenging, where Deep Learning (DL) techniques have shown promise. Most DL techniques provide a robust and accurate classification of performance. However, due to the complex architectures of the DL models, the classification results are difficult to interpret, causing challenges for their adoption in the healthcare industry. To help improve the adoption of AI in healthcare, this study incorporates Transfer Learning, Ensemble Learning, and XAI techniques to propose an effective and interpretable model. This work compares the performances of pre-trained models for the early detection of Alzheimer’s Disease (AD) and Parkinson’s Disease (PD). Specifically, the XAI technique Saliency Map was used to overlay gradients on the MRI scan, elucidating the regions on the MRI scan that led the model to its diagnosis. The Kaggle dataset used in the study has three classes: Parkinson’s disease (PD), Alzheimer’s disease (AD), and control (healthy). This study compares the performance of various pretrained models. Additionally, the diagnoses produced by the pretrained models were ensembled to produce a final diagnosis. Combining the predictions of multiple pretrained models can boost the performance of the model because it combines the strengths of multiple pretrained models, achieving a higher performance than the pretrained models. The best pretrained model EfficientNetB7 received an accuracy of 94.58% with an F1-score of 95.81%. The proposed model of this study is the ensemble learning model with an accuracy of 97.04% and an F1-score of 97.69%.