Determination of freezing of gait with wearable sensors in patients with Parkinson's disease
Küçük Resim Yok
Tarih
2020
Yazarlar
Eliiyi, Uğur
Keskinoğlu, Pembe
Kahraman, Turhan
Özkurt, Ahmet
Yürdem, Betül
Duran, G.
Genç, Aslı
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Wiley
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Objective: The aim was to determine freezing of gait (FOG) with wearable sensors in patients with Parkinson’s disease (PD).
Background: PD is a neurodegenerative disorder leading to deficits in automatic motor performance. FOG is a major mobility problem for patients with PD, can be accompanied by postural instability and subsequent falls. Accurate and automatic FOG detection are essential for long-term symptom monitoring or preventing FOG via cueing. Although some studies have investigated the use of wearable sensors to detect FOG, conducted mostly with participants who were mainly in early stages of PD, there is no firm consensus regarding appropriate methodologies.
Method: This study had a diagnostic accuracy design. Multi-segmental acceleration data was obtained from 12 patients with PD performing standardized tasks, and clinical assessment of FOG was performed by an experienced neurologist in real time and from video recordings. Three-axis wireless accelerometers were attached to patients’ ankles, waist and wrists. Trials were performed during the drug-free period, at least 12 hours after taking medication. The standardized tasks included standing from a chair, walking, 180o and 360o turnings, and passing a doorway. Trials were repeated at least 3 times, and up to 5 times if there was no FOG event. Sensor signals were processed as numerical data.
Results: The mean age of patients was 64 years, 58% of them were males, and mean of disease duration was 10 years. Modified Hoehn and Yahr scale scores ranged from 2.5 to 4. The data matrix dimension is 103,261*18. Random forest (RF), artificial neural network (ANN), and decision tree (DT) methods, which are among the supervised learning algorithms, were used to predict FOG in the presence of misleading tremors on this big data. Algorithms’ performances were AUCRF/ANN/DT= 0.985 / 0.962 / 0.765 and sensitivityRF/ANN/DT= 97.1% / 94.3% / 93.8%.
Conclusion: The predictions of ANN and RF were much better, while the sensitivity of DT was close to other methods. FOG detection is important to prevent it before occurring and decrease its effects. In this study, it has been shown that FOG can be detected by using the proposed algorithms with data collected from wearable sensors in patients with PD, even who are in late stages of PD.
Açıklama
Movement-Disorder-Society (MDS) International Virtual Congress -- SEP 12-16, 2020 -- ELECTR NETWORK
