Deep-EFNet: An MRI-Based Clinical Decision Support System for Parkinson’s Disease and Alzheimer’s Disease Classification

The growing prevalence of neurodegenerative disorders (NDs), particularly Background: Parkinson’s disease (PD) and Alzheimer’s disease (AD) are among the most prevalent neurodegenerative disorders and are frequently associated with progressive functional impairment, mobility limitations, and reduced quality of life. Early and accurate differentiation between these conditions remains clinically challenging because of overlapping neurological manifestations and structural brain abnormalities. Magnetic resonance imaging (MRI) has emerged as an important non-invasive tool for supporting diagnostic evaluation; however, manual interpretation is time-consuming and subject to inter-observer variability. Objective: This study aimed to develop and evaluate an efficient deep learning framework for automated MRI-based classification of PD, AD, and cognitively normal individuals to support neurological assessment and clinical decision-making. Methods: A transfer learning framework, termed Deep-EFNet, was developed using an EfficientNetB4 backbone combined with customized classification layers incorporating Global Average Pooling and dropout regularization. The model was trained and evaluated on a publicly available dataset comprising 7,839 structural brain MRI images categorized into AD, PD, and healthy control groups. Standard preprocessing, data augmentation, and stratified sampling techniques were employed to enhance model robustness and generalization. Results: Deep-EFNet achieved an overall classification accuracy of 99%, outperforming benchmark transfer learning models including DenseNet121, ResNet50, Xception, and MobileNet. The model demonstrated high precision, recall, and F1-scores across all diagnostic categories, indicating excellent discriminatory capability for differentiating neurodegenerative disorders from healthy controls. Receiver operating characteristic analysis further confirmed strong classification performance. Conclusions: The proposed Deep-EFNet framework provides a highly accurate and computationally efficient approach for automated MRI-based classification of PD and AD. The findings suggest potential utility as a clinical decision-support tool for neurological evaluation and early disease identification. Further validation using multicenter datasets and prospective clinical studies is recommended to establish real-world applicability.