Please use this identifier to cite or link to this item:
|Title:||Detecting passive fatigue with brain-computer interface in drowsy driving and stroke rehabilitation||Authors:||Foong, Ruyi||Keywords:||Engineering::Computer science and engineering::Computer applications||Issue Date:||2019||Source:||Foong, R. (2019). Detecting passive fatigue with brain-computer interface in drowsy driving and stroke rehabilitation. Doctoral thesis, Nanyang Technological University, Singapore||Abstract:||A recent topic of interest for electroencephalography(EEG)-based brain-computer interface (BCI) is mental state monitoring. Mental states such as fatigue, frustration and attention have been shown to affect healthy users’ BCI performance. However, ground truth of mental states is often difficult to ascertain. Instead, surrogate measures such as ratings or reaction times have been used. Additionally, mental state monitoring has yet to be applied in the medical domain. Currently, a handful of studies have reported the mental state of fatigue in motor imagery (MI)-BCI stroke rehabilitation only via subject ratings, but not EEG data. Hence, there is a need to develop algorithms that can detect mental states despite a lack of ground truth. In particular, this thesis focuses on detecting the mental state of passive fatigue induced by monotony in healthy subjects in drowsy driving and in stroke survivors using MI-BCI stroke rehabilitation. A two-step iterative cross-subject negative-unlabeled (NU) learning algorithm is proposed to address the lack of ground truth of passive fatigue in drowsy driving studies. In the first step, subjects’ alert and unlabeled driving EEG data are iteratively used to label each subject’s most fatigued driving block. The second step then used the alert and newly-labeled fatigued blocks to compute subjects’ fatigue score. A drowsy driving experiment is conducted to verify effectiveness of the NU learning algorithm. Dry EEG MuseTM headband data is collected from 29 healthy subjects. The results showed that the algorithm converged in 7 iterations and yielded a high mean accuracy of 93.77% ± 8.15% in detecting fatigue in a cross-subject manner. The results also showed that the fatigue score is significantly negatively correlated with relative EEG beta band power, an indicator of passive fatigue. This suggested that high fatigue score is associated with low beta band power and therefore low brain arousal. Hence the proposed algorithm was able to quantify passive fatigue well. In the medical domain, a clinical trial is conducted with Neurostyle Brain Exercise Therapy Towards Enhanced Recovery (nBETTER), an EEG-based MI-BCI with only visual feedback. Subjects’ Fugl-Meyer Motor Assessment (FMA) score is measured to assess clinical efficacy. Also, BCI performance is correlated with relative EEG beta band power and the proposed NU learning algorithm is applied as well to investigate the presence of fatigue. The results showed significant FMA score gains, but are comparable to a retrospective control group. The results also showed a significant positive correlation of BCI performance with beta band power. Additionally, the proposed NU learning algorithm yielded fatigue scores that are non-significantly negatively correlation with BCI performance. Together, these suggested that fatigue might be present, resulting in the poorer BCI performance and comparable efficacy to the control group. This postulated a disadvantage of MI-BCI systems in inducing fatigue due to the monotony of performing MI repetitively. These studies have presented the potential of BCI to play a fatigue monitoring role, by presenting a cross-subject algorithm to detect and quantify fatigue from alert data in a drowsy driving setting, and finding the possibility of fatigue during a MI-BCI intervention.||URI:||https://hdl.handle.net/10356/107596
|Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.