A Multimodal Framework for Parkinson's Disease Detection Using Image, Audio, and Clinical Data
Year: 2026
Type: conference-paper
Abstract: Parkinson's disease is a neurodegenerative disorder that gradually affects movement control and speech production. Detecting the disease at an early stage is important for improving treatment outcomes and slowing disease progression. This research proposes a multimodal machine learning framework for identifying Parkinson's disease using three different sources of information: image-based drawing tests, voice recordings, and clinical measurements. The image modality analyzes spiral and wave drawings to identify motor abnormalities using a convolutional neural network model. The audio modality evaluates speech characteristics such as jitter and shimmer extracted from voice signals. In addition, structured medical attributes from the Parkinson's telemonitoring dataset are used in the clinical modality. Experimental analysis shows that the clinical model achieved the highest individual accuracy of 90%, while the image and audio models obtained accuracies of 89% and 88% respectively. The results indicate that each modality captures different symptoms associated with Parkinson's disease. Integrating multiple data sources can therefore improve the reliability of automated Parkinson's disease detection systems.
Authors: Archana Yashodhar, Shashidhar Kini
Institutions: St. Aloysius (Deemed to Be University), Srinivas University
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7167203003