Machine learning researcher aims to facilitate earlier falls intervention

A machine learning project being undertaken by Balance Mat Pty Ltd electronics engineer Abishek Shrestha has the ultimate aim of enabling earlier falls intervention for people over 65.

Abishek is conducting the research as part of his PhD studies at the University of Canberra under the supervision of Dr Maryam Ghahramani, whose research interest is in the field of human motion analysis using machine learning for rehabilitation purposes.

Dr Ghahramani is Senior Lecturer and Program Director in Engineering at the Faculty of Science and Technology, a member of the UC Human Centred Technology Research Centre and a long-time research partner of Balance Mat Pty Ltd – the manufacturer and intellectual property owner of Balance Mat systems and technology. A photographic slide show near the foot of the Balance Mat Pty Ltd company website home page makes for interesting viewing. It features Maryam (shown above), the Balance Mat and the Balance Mat calibration robot, back in May 2022. Abishek Shrestha and his software developer colleague Binod Shrestha are shown in the slide show too.

Abishek’s research will explore the Balance Mat system’s ability to identify clinically meaningful balance deficits – first, by conducting measurement assurance across technical, clinical and informatics domains; then by comparing balance data from existing clinical tools; and lastly by using machine learning to develop a predictive model of falls likelihood and the consequent need for timely preventative interventions.

The study will compare the Balance Mat’s capacity to accurately detect and grade previously undiagnosed balance impairments against existing gold standard balance tests. These include the Berg Balance Scale, Timed Up and Go test and timed single leg stance – all of which are established indicators of functional balance capabilities, fall likelihood and the need for intervention.

To do this, Abishek will recruit community-dwelling adults over 65 years of age who will be categorised into high-fall-risk and low-fall-risk groups based on validated cut-off thresholds for the Berg Balance Scale and Timed Up and Go test.

Participants will undergo a Balance Mat assessment in four static balance positions: normal with eyes open, normal with eyes closed, tandem (heel-to-toe) with eyes open and single leg stance with eyes open. Several sway metrics for each stance will be recorded: mean, sway path, sway range, RMS (root mean square) sway, maximum and minimum sway and sway velocity.

Using Analysis of Variance (ANOVA) statistical models, Abishek will compare sway scores between risk groups and across the four stances. He will also use receiver operating characteristic (ROC) analysis to determine heightened instability and fall risk thresholds.

The ultimate aim of the study is to provide evidence for community-based screening of undiagnosed balance decline among older people to activate supportive interventions earlier and prevent injury from falls.

Abishek is currently preparing a paper for submission to the 2024 Australasian Conference on Robotics and Automation (ACRA 2024) to be held in Auckland, New Zealand on 27 – 29 November 2024.

For more information please contact Ian Bergman (ian@balancemetrix.com.au or mob. 0457 123 852) or Abishek Shrestha (abishek.shrestha@balancemat.com.au or mob. 0449 970 096).