Multimodal Artificial Intelligence Based Fall Risk Prediction in Parkinson's Disease

Overview

Parkinson's disease (PD) is characterized by motor symptoms such as bradykinesia, tremor, rigidity, and postural instability, often leading to gait disturbances and a high risk of falls. Dual-task walking assessments-requiring simultaneous motor and cognitive engagement-have gained importance in evaluating real-life mobility impairments in PD, as they more accurately reflect challenges faced during daily activities. While clinical tools such as the Timed Up and Go (TUG), Four Square Step Test (FSST), and Mini-BESTest are widely used, their in-person application may not always be feasible for individuals with mobility or access limitations. Telehealth-based assessment methods, therefore, offer practical alternatives. Recently, the integration of artificial intelligence (AI), particularly machine learning (ML), into clinical assessments has opened new possibilities for fall risk prediction by enabling the simultaneous analysis of motor, cognitive, and balance-related parameters. This study aims to predict fall risk in individuals with PD using AI-based models that incorporate multiple data sources. Furthermore, it compares the predictive accuracy of models derived from single-task and dual-task conditions, with the goal of developing a more precise and clinically useful decision-support tool for early intervention.

Parkinson's disease (PD) is a progressive neurodegenerative disorder that primarily affects the basal ganglia, particularly the substantia nigra, leading to hallmark motor symptoms such as bradykinesia, resting tremor, muscular rigidity, and impaired postural reflexes. These motor impairments often result in gait disturbances, postural instability, and ultimately, a significantly increased risk of falls. Fall-related injuries are a major source of morbidity, reduced mobility, and increased healthcare burden in individuals with PD, making early identification of fall risk a clinical priority. Traditional balance and gait assessments, such as the Timed Up and Go (TUG) test, the Four Square Step Test (FSST), and the Mini-Balance Evaluation Systems Test (Mini-BESTest), have been widely employed to evaluate static and dynamic balance components in clinical settings. However, these assessments are often conducted under single-task conditions, which may not fully capture the complex, real-life demands placed on individuals with PD. In contrast, dual-task paradigms-where individuals perform a cognitive or motor secondary task while walking-have demonstrated greater sensitivity in detecting subtle deficits in postural control, as they mimic everyday situations more closely. Nevertheless, the practical implementation of such assessments is often hindered by logistical constraints, particularly among individuals with limited mobility or geographic access to healthcare facilities. In this context, telehealth-based assessment strategies are gaining momentum due to their ability to facilitate remote monitoring and evaluation with minimal equipment and reduced resource requirements. Recent advancements in artificial intelligence (AI), especially machine learning (ML) techniques, offer promising solutions for enhancing the predictive power of clinical assessments. ML algorithms can integrate and analyze complex datasets encompassing motor, cognitive, and balance-related parameters without relying on predefined statistical assumptions. These models are capable of identifying nonlinear relationships and subtle patterns within the data, thereby enabling more individualized and accurate fall risk predictions. The primary objective of this study is to develop and validate AI-based predictive models for fall risk estimation in individuals with Parkinson's disease by incorporating multimodal data obtained from both single-task and dual-task walking assessments. Additionally, the study aims to compare the predictive performance of models derived under these two conditions to determine whether dual-task data enhance the sensitivity and specificity of fall risk classification. Through this approach, the research seeks to establish a clinically relevant, remote-friendly, and data-driven decision-support tool to inform timely interventions and personalized rehabilitation strategies.