Smartphone based automated detection of Freezing of Gait in people with Parkinson's disease

Parkinson's disease (PD) is the second most common neurodegenerative disorder, with serious motor and non-motor complications associated. Freezing of gait (FOG) is a form of akinesia (loss of movement) affecting most of PD patients in advanced stages of the disease, and represents one of the most disabling symptoms of PD. Current therapies are highly effecting in ameliorating the symptoms of the disease, but their delivery need to be patient-specific, based on motor fluctuations and progression of the disease in each patient. Wereable sensors have been recently proposed for providing objective assessment of FOG during the real life. This study aimed to the construction of an algorithm capable of automated FOG detection, in order to prove the feasibility of a simple, small and low cost detection system for home monitoring of FOG episodes in PD patients. Data acquisition was executed with a commercial smartphone, which included several inertial sensors (e.g. accelerometers, gyroscopes). A total number of 59 partecipants were involved in the study, including PD freezers, PD non-freezers and control samples, leading to acquisition of more than 3 hours of acceleration signal, yet only 15 FOG episodes were observed. Processing of acceleration signals was executed offline; features able to differentiate FOG from other activities were identified and two Support vector machines (SVM) classifiers were used for solving a multi-class problem. Results were promising, with average sensitivity, specificity and accuracy reaching values over 90 \%, while average precision was not as good, reaching only 82 %. Despite the smallness of the FOG dataset does not allow to give the results statistical meaningfulness, detection algorithm demontrated great ability of generalization and robustness.