Objective: This study evaluated the utility of Artificial Intelligence (AI) in detection of patients with Parkinson’s disease from stride dynamics.

Background: Parkinson’s Disease (PD) is the second most common neuro-degenerative disease affecting mankind. The pathogenesis of PD is dopaminergic cell death which manifests with motor deficits. Diagnosing and treating PD right at its earlier stages is of paramount significance as it improves the Quality-of-Life (QOL) of patients. The Single Photon Emission Computed Tomography (SPECT) scans are used as an ancillary diagnostic technique. However, their absorption into clinics, particularly in resource-limited settings is restricted. Gait cycle exhibits regular pattern of foot movements. These possess biomarkers to differentiate between patients affected with Parkinson’s disease and healthy controls. We propose to develop a “non-invasive and cost-effective” AI technique for diagnosing PD using gait-based biomarkers.

Method: The PhysioNet dataset which included a total of 166 participants with 93 patients affected with PD and 73 healthy controls is analyzed. 8 sensors per foot i.e., 16 sensors in total record the vertical ground force reaction from the participants as they walked at their usual pace for 2 minutes. The data was recorded at the rate of 100 samples per second. The output of these recordings reflects the centre of pressure and other timing measures for each foot as function of time. The stride-to-stride dynamics was studied from these recordings.

These recordings were fed to the Recurrent Neural Network (RNN) model. The RNN model was appropriate as it is ideal for analyzing time-series data. The Training : validation datasets are in the ratio 80 : 20. The performance measures like accuracy, sensitivity, specificity, precision and F1-Score of the RNN model were evaluated.

Results: The RNN classifier yielded an accuracy of 88.37%, sensitivity of 88.70%, specificity of 87.80%, precision of 87.90% and F1-Score of 86.30%.

Conclusion: The performance of RNN classifier is quite impressive as it differentiated patients affected with PD from healthy controls. The key pros of the proposed work include non-invasiveness, cost-effectiveness, no requirement of sophisticated instruments and technicians and provides rapid turnaround times. This could provide a major impetus to the evidence-based diagnosis of Parkinson’s disease worldwide.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/pd-ai-detection-of-parkinsons-disease-with-gait-based-biomarkers-using-ai-techniques/