Applications of Artificial Intelligence in the Diagnosis, Management and Control of Parkinsonism: A Systematic Review

Abstract

Background: Parkinson disease (PD) is the second most prevalent neurodegenerative disorder, affecting around ten million people worldwide and typically develops after age 50, with average age of onset around 60 years. It is characterised by progressive motor and non-motor symptoms. Traditional diagnostic methods often miss early signs, leading to delayed treatment. Recent advancements in artificial intelligence (AI) offer new hope, AI is increasingly used in the management of PD by utilising MRI analysis, smartwatches, sensors, wearable devices, and smartphone apps. While many researchers have explored AI for diagnosis, early detection, and monitoring, a considerable vacuum remains in its use for rehabilitation and therapeutic management of PD symptoms. Objective: To provide an overview of recent uses of AI in managing PD Methods: The review followed the PRISMA guidelines, using a systematic search across Google Scholar, ScienceDirect, and PubMed. An initial 47 relevant peer-reviewed studies on AI applications in Parkinsonism were identified. Studies were included if they presented original clinical or technical data. Reviews and articles with, irrelevant, or insufficient data were excluded. Results: After screening, 25 studies were selected for final analysis. Five AI application areas were identified as ‘diagnosis’, ‘therapy’, ‘monitoring’, ‘early detection’, and ‘rehabilitation’. In diagnosis, 3D and 2D Convolutional Neural Networks (CNNs) achieved 88.90% and 72.22% accuracy using MRI data. Electroencephalography (EEG)-based classifiers showed Area Under the Curves (AUC) up to 99.4%. Drug repurposing tools like Therapeutics Graph Neural Network (TxGNN) revealed potential in identifying novel therapeutics. Monitoring tools, including wearable and nonwearable systems, reported over 97% precision. A CNN model showed 0.79 AUC for early detection using Cell Painting images, while other models using spiral drawings, speech, and breathing patterns reached AUCs up to 0.9 and accuracies up to 97%. In rehabilitation telemedicine and Tele-rehabilitation enhance patient management. AI-powered Tele-rehabilitation enables remote access to care, personalized therapy, continuous monitoring and feedback at home. Conclusions: High-performance models, particularly those using neuroimaging, EEG, and both wearable and non-wearable devices, demonstrate the potential to improve the accuracy and efficiency. However, ongoing validation and integration with clinical workflows remain major challenges to fully realizing the potential of AI applications in PD.