Automated machine vision enabled detection of movement disorders from hand drawn spirals

TL;DR

This study demonstrates that a CNN-based machine vision system can accurately classify Parkinson's disease, Essential tremor, and controls from scanned hand-drawn spirals, offering an affordable diagnostic tool.

Contribution

It introduces a CNN approach using scanned paper drawings for movement disorder diagnosis, avoiding expensive digitizing equipment.

Findings

Discrimination accuracy of 98.2% for PD vs. controls

92% accuracy for PD vs. ET and controls

Hyperparameter tuning improves accuracy by up to 4.33%

Abstract

A widely used test for the diagnosis of Parkinson's disease (PD) and Essential tremor (ET) is hand-drawn shapes,where the analysis is observationally performed by the examining neurologist. This method is subjective and is prone to bias amongst different physicians. Due to the similarities in the symptoms of the two diseases, they are often misdiagnosed.Studies which attempt to automate the process typically use digitized input, where the tablet or specialized equipment are not affordable in many clinical settings. This study uses a dataset of scanned pen and paper drawings and a convolutional neural network (CNN) to perform classification between PD, ET and control subjects. The discrimination accuracy of PD from controls was 98.2%. The discrimination accuracy of PD from ET and from controls was 92%. An ablation study was conducted and indicated that correct hyper parameter optimization can increases the accuracy up to 4.33%. Finally, the study indicates the viability of using a CNN-enabled machine vision system to provide robust and accurate detection of movement disorders from hand drawn spirals.