NEURONET: AN ADVANCED DEEP LEARNING FRAMEWORK FOR EARLY DIAGNOSIS AND PREEMPTIVE TREATMENT OF NEURODEGENERATIVE DISEASES

Globally, citizens are bearing a heavy burden about the ravages of neurodegenerative diseases such as Alzheimer’s and Parkinson’s. These diseases develop slowly and are frequently diagnosed at a late stage when therapy possibilities are restricted. Currently level of diagnostics rely on clinical examination and imaging but many times early pathological clues are overlooked. AI driven decay prediction models, including Capsule Networks (CapsNets) and Sparse Learning Models are developed and achieve improvement, but not efficient enough to fuse multimodal data jointly and obtain spatial-temporal features which are critical for early detection and prediction. To fill these gaps, we propose a novel hybrid deep learning framework named NeuroNet, which is composed of 3D Convolutional Neural Networks (CNN) for learning spatial features, Long Short-Term Memory (LSTM) networks for modeling temporal sequences, attention mechanism for selecting diagnostic regions and clinical biomarkers for multimodal information fusion. This architecture is fine-tuned by Bayesian hyperparameter optimization, and validated through ADNI dataset. The primary contribution of this work is the evidence that multimodal fusion with attention-informed deep learning, drives a dramatic improvement in diagnostic accuracy and interpretability. Accuracy, precision and recall achieved by NeuroNet were 98.62%, 98.50%, 98.70%, and an AUC–ROC value of 0.998, respectively, showing a superior performance over several state-of-the-art models. Explainable AI techniques like Grad-CAM and SHAP increase model transparency even more. This work provides new insights in the form of a clinically viable and interpretable deep learning model which pushes the frontier of AI-facilitated neurodegenerative disease diagnosis. The model can be easily incorporated in clinical diagnostical systems for early intervention and better prognosis