Multi-class classification and feature analysis of FTM drawing tasks in a digital assessment of tremor

Drawing tasks, such as completing an Archimedes spiral, are commonly used to diagnose and evaluate tremor severity in individuals presenting with tremor-like symptoms. Clinical evaluation of these tasks is generally restricted to having a clinician examine and rate performance using one of several rating scales. Although this method is relatively effective in tremor assessment, it restricts tremor assessment to clinical settings and requires each assessment be evaluated individually, placing considerable time demands on both patients and clinicians. Here is presented the first multi-class approach to evaluation of tremor severity based on data remotely recorded by a mobile- or tablet-based drawing application. Taking a divergent approach from previously published work, which focuses solely on binary diagnostic capacity of similar systems, our work seeks to differentiate between healthy subjects, essential tremor patients receiving deep brain stimulation treatment, and those same patients with treatment disabled. Our classification algorithm was highly effective, with overall accuracy of 97.04%. Of note is that all erroneously classified samples within this set were ”treated” samples mistakenly labelled as ”healthy” samples, implying that DBS treatment is capable of completely eliminating tremor. Future work will focus on adapting these findings into a regression algorithm capable of automatically evaluating tremor severity on an established tremor rating scale, allowing fully automated remote assessment of tremor severity for the coming era of telemedicine.