Explainable Multimodal Deep Learning Framework for Longitudinal Prediction of Neurodegenerative Disease Progression with Clinical Validation
Abstract
Neurodegenerative diseases, particularly Alzheimer's Disease and Parkinson's Disease, are rapidly increasing as one of the leading causes of disability, cognitive decline, and death among the elderly population worldwide. Accurately predicting long-term disease progression remains a significant research challenge due to the disease's slow progression, multifaceted biological characteristics, and patient-specific clinical presentation. Most existing machine learning and deep learning models rely on a single data source, cannot effectively analyze time-dependent pathological changes, and have non-interpretable decision-making processes, thereby limiting their practical clinical use. To overcome this limitation, the current study proposes an Explainable Multimodal Deep Learning Framework that predicts long-term progression in neurodegenerative diseases by integrating neuroimaging, clinical information, cognitive assessments, demographic characteristics, and biomarker data. Experimental results indicate that the proposed model achieved 94.8% Accuracy, 93.9% Precision, 94.3% Recall, 94.1% F1 Score, and 96.2% area under the curve (AUC). The average accuracy in five-fold cross-validation was 94.8%, confirming the model's stability and ability to generalize. In addition, 96.3%, 94.8% and 92.6% accuracy were achieved in predicting disease progression at one year, three years and five years, respectively. Clinical Validation showed 94.5% agreement with expert physician evaluations and high acceptability in the Explainability Assessment. The results demonstrate that the proposed interpretable multimodal deep learning framework is more accurate, more transparent, and more clinically acceptable than existing methods. This research provides a strong foundation for the development of future personalized medicine, early risk identification, and artificial intelligence-based clinical decision support systems.