Computer vision, a promising artificial intelligence tool to detect bradykinesia in Parkinson’s disease

Abstract

Objective: To analyze the accuracy of assessing bradykinesia in Parkinson’s disease (PD) using Computer vision.

Background: PD is a chronic and progressive neurodegenerative disorder characterized by motor and non-motor symptoms. Bradykinesia is one of the cardinal symptoms of PD, and it is characterized by slowness and difficulty initiating voluntary and involuntary movements. Computer Vision is a field of Artificial Intelligence that deals with the ability of computers to interpret images and videos. Detection of PD bradykinesia using computer vision is an open area of research and has been used to extract hand movement characteristics during a finger-tapping test. However, the accuracy of machine learning models has not yet been established in PD.

Methods: Multicenter, cross-sectional, case-control study. Patients diagnosed with PD and age, and gender-matched controls were included. We collected a set of videos of the finger-tapping test (left and right hands) from PD patients and controls. MediaPipe, a computer vision tool, was selected to extrapolate finger positions features in each frame, using complex computed statistics. Feature selection techniques were applied to a set of time-series features extracted from the videos. These features were used to feed various machine-learning models for detecting PD bradykinesia. Then, the best machine learning model was selected based on different evaluation metrics.

Results: 45 patients with PD (56.25%), and 35 controls (43.75%), 41 males (51%%), and 39 females (49%) with a mean age of 69.52 + 9.22 years, and median Hoehn Yahr stage of 2 (1-3) were included. The accuracy of Computer Vision for detecting bradykinesia in PD will be reported.

Conclusions: The diagnosis of PD is based on clinical criteria established by a doctor. By utilizing these techniques, the goal is to develop a non-invasive and accurate method for detecting PD, facilitating an early diagnosis and management of the disease. In addition, computer vision can also assist in tracking the progression of the disease, providing insight into how the disease is evolving, which can help to determine the effectiveness of pharmacological and non-pharmacological interventions.

José Luis Garrido-Labrador

Assistant Lecturer in Computer Languages and Systems

PhD in Machine Learning, researching in semi-supervised learning and restricted set classification. Assistant Lecturer in Computer Languages and Systems at Universidad de Burgos.