Efficient multi-modal fusion framework with advanced AI-driven approaches for automated Parkinson's disease detection

Parkinson’s Disease is a neurological disorder, characterized by gradual loss of dopaminergic neurons in the substantia nigra resulting in: tremors, muscle rigidity, bradykinesia and postural instability. Other symptoms which involve other parts of the body and organs are as follows: loss of smell, difficulty to sleep, and changes in cognition. Being a neurodegenerative disorder, it becomes important to detect the disease in early manner. Different researchers across all over the world are trying to develop such techniques that can be helpful for disease detection process. Manually detection process based on the medical images is complex and time consuming where accuracy and reliability is also questionable. Here, deep learning came to the picture to make the process automatic and reliable where deep neural network-based models are being used to classify different diseases quite accurately and efficiently. Utilizing the potentiality of Artificial Intelligence (AI), a novel work on Parkinson’s disease diagnosis has been performed with comprehensive personalized management strategies. Here in this work, AI-powered detection frameworks have been designed for Parkinson’s disease classification. Seven Machine Learning models (Logistic Regression, K-Nearest Neighbors, Perceptron, Support Vector Machine, XGBoost, Decision Tree and Random Forest) and five Deep Learning Models (ResNet101, VGG19, Xception, Inception and EfficientNet) were trained and best models have been selected based on the performance analysis. Feature fusion technique with modified classification layers with hyperparameter tuning ensures optimized and remarkable output. LR and VGG19 have been selected where accuracies of 95.74% for EEG data with LR model, 96.78% for MRI image-based classification and 97.7% for spiral and wave-based drawings with proposed fusion VGG19 model. • A thorough examination of recent developments in the field was conducted, comparing existing state-of-the-art methodologies. This systematic review identifies current gaps and establishes the novelty and contributions of the proposed work. • Three robust classification frameworks were developed—EEG data-based, MRI image-based, and spiral-wave drawing data-based—using optimized machine learning models and advanced deep learning fusion techniques. • The proposed frameworks leverage cutting-edge AI methodologies to ensure superior performance metrics, including high accuracy, sensitivity, efficiency, and reliability across various datasets. • The study demonstrates practical applications of the proposed frameworks in real-world scenarios, showcasing their potential to enhance disease diagnosis, classification, and management. • The novel fusion of multiple data modalities within the proposed frameworks highlights a unique approach to addressing complex disease analysis and challenges, setting a benchmark for future research in the domain. • A systematic comparative analysis has been done to figure out the current gaps and compared the novelty of the proposed work with other existing works. The comparison demonstrates that the proposed frameworks are superior to previous approaches in accuracy, sensitivity, and reliability, as well as dealing with limitations identified in the literature. • In addition to model development, an IoT-based integrated framework is proposed as a novel extension pathway for clinical application to achieve continuous monitoring, real-time analysis and remote patient management thus closing the gap between the computational research and the actual implementation of healthcare. Major improvements have been done in the manuscript during the revision including: • Dataset Clarification: The dataset description has been thoroughly revised to clearly specify that the “Parkinson Diseases EEG Dataset” used in this study consists of EEG-based numerical features rather than voice recordings. The section has been rewritten to eliminate any ambiguity regarding data modality and feature type. • Exact Statistical Values: All statistical results now include the exact p-values (e.g., 0.021, 0.013, 0.018, 0.034) instead of generic expressions such as “p < 0.05.” This enhancement improves transparency and interpretability of the reported findings. • IoT Integration: In the manuscript, the importance of the IoT integration is justified, proposed as a novel extension pathway for clinical application to achieve continuous monitoring, real-time analysis, and remote patient management, thus closing the gap between the computational research and the actual implementation of healthcare. • Language and Grammar Improvements: The entire manuscript underwent a detailed English language revision to correct grammatical errors, improve sentence flow, and remove redundancies. The text has been polished for better readability and academic tone.