2.2 - Minimizing the Latency of Freezing of Gait Detection on Wearable Devices
- Event
- iCCC2024 - iCampµs Cottbus Conference
2024-05-14 - 2024-05-16
Cottbus - Band
- Vorträge
- Chapter
- Gesundheit
- Author(s)
- A. Haddadi Esfahani, O. Maye, M. Frohberg, P. Langendörfer - IHP–Leibniz-Institut für innovative Mikroelektronik, Frankfurt/Oder
- Pages
- 49 - 52
- DOI
- 10.5162/iCCC2024/2.2
- ISBN
- 978-3-910600-00-3
- Price
- free
Abstract
Freezing of Gait (FoG) is a common and severe symptom that impacts persons who have been diag-nosed with Parkinson's disease. The detection of FoG is of greatest significance for precise diagnosis, preventing falls, and obtaining accurate measurements for severity of FoG episodes. These factors are critical for optimizing treatment approaches and enhancing the overall quality of life for those af-fected by FoG. Real-time FoG detection may be achieved by using a wearable system that can be worn by the patient. This configuration includes a sensor that is coupled with inference software on a computing device and a vibrator. The patient is alerted of a FoG incident by the vibrator that is activat-ed on demand. The key role in FoG detection is the latency between the initial moments of imminent FoG episode and the moment at which the patient is a notified by the vibrator. By using more ad-vanced FoG detection algorithms, the duration between incidents may be reduced, hence aiding to prevent patient falls and avoiding injuries. In this paper, we modified the model which was used in our previous work and improved the latency from 50ms to 3 ms. The dataset and input features remain unchanged from our earlier study to ensure comparability between performances of the two models. The individual models for one-second windows running on a PC of nine patients achieved a mean of 90% and 88% of sensitivity and specificities, respectively. The models that were converted and exe-cuted on Google Coral exhibited comparable performance, with a maximum variance of 1% compared to the performance attained on a personal computer.