Novel machine learning-based hybrid strategy for severity assessment of Parkinson's disorders

Parkinson's disease (PD) severity assessment in clinical settings largely depends on expertise level of clinicians which have inherent limitations and non-uniformity. Instrumented gait analysis plays a significant role in disease diagnosis and management. However, these are agonized from data dispersion due to demography, anthropometry, and self-selected walking speed of an individual. This research work aims to develop computationally efficient hybrid strategy using normalized features for PD severity evaluation. The relevance of each considered gait feature in demonstrating the outcomes is explained through feature importance and partial dependence plot (PDP) to build substantial insight for clinical needs. Gait, a biomarker, is used to access human mobility by utilizing vertical ground reaction force (VGRF) data of 72 healthy and 93 Parkinson's individuals. A multi-variate normalization approach identifies gait differences between PD and non-PD. The proposed hybrid model used is able to detect PD with accuracy of 99.39% and 99.9%, and its severity assessment based on MDS-UPDRS-III shows coefficient of determination (R²) as 97% and 98.7% using leave-one-out cross-validation (CV) and tenfold CV respectively. The significant features suitable for clinical implications are reported. Moreover, normalized gait parameters supplement capability to compare individuals with diverse physical properties, resulting in assistive system for evaluation of PD severity.