Analysis and comparison of gait impairments in patients with Parkinson’s disease and normal pressure hydrocephalus using wearable sensors and machine learning algorithms

Objectives. Gait impairments in patients with Parkinson’s disease (PD) and normal pressure hydrocephalus (NPH) are visually assessed by movement disorders experts for diagnoses and to decide on pharmaceutical and surgical interventions. Despite standardised tests and clinicians’ expertise, such approaches entail a considerable level of subjectivity. The recent development of wearable sensors and machine learning offers complementary approaches providing more objective, quantitative assessments of gait impairments. We aim to employ the data gathered from an inertial measurement unit synchronized with a novel foot pressure sensor embedded in the patient’s shoes to characterize gait impairments. We focus on distinguishing PD from NPH and on assessing gait impairment before and after surgical intervention.
Methods. A cohort of 10 PD and 10 NPH patients was assembled and patients performed standardised walking tests. Measurements were performed employing wearable sensors comprising a three-axes gyroscope, a three-axes accelerometer and eight pressure sensors embedded in each patient’s shoe. To analyse the generated data, existing algorithms were implemented and adapted. These allow to compute gait cycle parameters such as step time and metrics characterizing the swing and stance phases. Machine learning algorithms where employed to identify major changes in gait cycle parameters between the two groups of patients, and for individual patients before and after surgical intervention as DBS implantation in PD and Shunt implantation in NPH.
Results. The gait impairments of both disease groups were measured and quantified. An algorithm to extract gait cycle parameters from sensors was implemented, tested and employed on such patients. Gait cycle parameters within and between the groups of PD and NPH patients were compared, assessing what gait cycle parameters allow to distinguish between these groups. Gait cycle impairments of patients before and after surgery were compared, assessing the effect of DBS or Shunt implantation and which gait cycle parameters allow to monitor symptoms improvement.
Conclusions. Wearable sensors measuring pressure, combined with gait cycle parameters extraction and machine learning algorithms, have a great potential for objective evaluation of gait impairment. In particular, they allow to characterize what differentiate such impairments between PD and NPH patients, and what allow to assess motor symptoms improvement after surgery.