AI-based Fall-Risk Assessment during Daily Activities in Post Stroke Survivors using Smartphones

点击翻译按钮获取中文摘要

Project Summary Falls in post-stroke survivors are up to 73% in the first year after stroke, and at least 70% of ambulatory stroke survivors experience an annual fall. The detrimental effects of falls include serious injuries, increased morbidity and mortality, dwindling functional mobility and quality of life, and high health-related costs. Most fall risk assessments for ambulatory post-stroke survivors are based on an ordinal scale of functional measurements, lack objectivity and accuracy, and are limited to clinical or laboratory environments. Early identification of post- stroke survivors at risk of falling is crucial for developing timely tailored interventions to reduce falls. This project aims to develop a machine learning (ML) based fall risk assessment tool for ambulatory stroke survivors by using inertial sensor data from smartphone worn at the waist during activities of daily living. This endeavor will involve graduate and undergraduate (UG) students at each stage of the project and expose them to multiple facets of rigorous scientific research. Chapman University is at the forefront of stroke rehabilitation and organizes Stroke Boot Camp (SBC), a free rehabilitation program every semester. In addition, CSU Long Beach’s pro bono clinic will provide us with easy access to the nearby stroke population. The overall goal of this project is to develop a portable decision support system for clinicians to diagnose fall risk even when the patient is away from the clinic. This study aims 1) To establish if digital biomarkers extracted from the smartphone data while performing prescribed ADLs significantly differ between high and low-risk fallers in laboratory settings. The remaining data needed to build the ML model is collected entirely in the participant’s home setting. The participants will wear the smartphone on their waist during all waking hours and perform regular activities of daily living. 2) To train three ML models that can classify fall-risk using different data modalities: using i) passively collected 3-day ADL data, ii) data from prescribed simple ADL tasks like turning, walking, and sit-to-stand, iii) combined subjective and objective data. 3) To assess the predictive validity of the ML models against actual fall occurrences after six months. The successful implementation of the project will enhance stroke care by an accurate fall risk assessment for ambulatory stroke survivors. Identifying post-stroke individuals at high risk of falling will allow early intervention to improve care and quality of life in these individuals. In addition, this study has the potential for developing a product that could track progress during stroke rehabilitation.