On The Feasibility of Multimodal Dialog Based Remote Balance Assessment

A variety of tests exist to assess balance, which are usually administered by a clinician in clinic. This limits how often they can be administered and requires patients to travel to the clinic, which can often be inconvenient or difficult, especially in cases involving movement disorders. To address this issue, we evaluate the feasibility of using a multimodal dialog-based platform for remote balance assessment. The platform uses the participants’ webcams to track their body poses while a virtual agent guides them through a series of tasks to assess balance inspired by the Berg Balance Scale (BBS). We recorded 62 assessment sessions from 58 healthy participants to evaluate whether participants were able to perform the tasks as instructed and to evaluate whether the obtained data can be used to automatically extract analytically valid metrics to assess balance. The results show that the assessment is feasible but participant compliance with instructions is crucial to ensure good overall data quality. Furthermore, we show that the time taken to stand from a seated position (TSS) can be accurately calculated if participants comply with instructions to perform the task correctly.