Brain-computer interfaces (BCIs) are very convenient tools to assess locked-in (LIS) and completely locked-in state (CLIS) patients' hidden states of consciousness. For the time being, there is no ground-truth data in respect to these states for above-mentioned patients. This lack of gold standard makes this problem particularly challenging. In addition to consciousness assessment, BCIs also provide them with a communication device that does not require the presence of motor responses, which they are lacking. Communication plays an important role in the patients' quality of life and prognosis. Significant progress have been made to provide them with EEG-based BCIs in particular. Nonetheless, the majority of existing studies directly dive into the communication part without assessing if the patient is even conscious. Additionally, the few studies that do essentially use evoked brain potentials, mostly the P300, that necessitates the patient's voluntary and active participation to be elicited. Patients are easily fatigued, and would consequently be less successful during the main communication task. Furthermore, when the consciousness states are determined using resting state data, only one or two features were used.
In this thesis, different sets of EEG features are used to assess the consciousness level of CLIS patients using resting-state data. This is done as a preliminary step that needed to be succeeded in order to engage to the next step, communication with the patient. In other words, the 'conversation' is initiated only if the patient is sufficiently conscious. This variety of EEG features is utilised to increase the probability of correctly estimating the patients' consciousness states. Indeed, each of them captures a particular signal attribute, and combining them would allow the collection of different hidden characteristics that could have not been obtained from a single feature. Furthermore, the proposed method should allow to determine if communication shall be initiated at a specific time with the patient.
The EEG features used are frequency-based, complexity related and connectivity metrics. Besides, instead of analysing results from individual channels or specific brain regions, the global activity of the brain is assessed. The estimated consciousness levels are then obtained by applying two different soft-clustering analysis methods, namely Fuzzy c-means (FCM) and Gaussian Mixture Models (GMM), to the individual features and ensembling their results using their average or their product. The proposed approach is first applied to EEG data recorded from patients with unresponsive wakefulness syndrome (UWS) and minimally conscious state (MCS) (patients with disorders of consciousness (DoC)) to evaluate its performance. It is subsequently applied to data from one CLIS patient that is unique in its kind because it contains a time frame during which the experimenters affirmed that he was conscious. Finally, it is used to estimate the levels of consciousness of nine other CLIS patients. The obtained results revealed that the presented approach was able to take into account the variations of the different features and deduce a unique output taking into consideration the individual features contributions. Some of them performed better than others, which is not surprising since each person is different. It was also able to draw very accurate estimations of the level of consciousness under specific conditions.
The approach presented in this thesis provides an additional tool for diagnosis to the medical staff. Furthermore, when implemented online, it would enable to determine the optimal time to engage in communication with CLIS patients. Moreover, it could possibly be used to predict patients' cognitive decline and/or death.
| Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa:de:qucosa:78610 |
| Date | 29 March 2022 |
| Creators | Adama, Volafidy Sophie |
| Contributors | Universität Leipzig |
| Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
| Language | English |
| Detected Language | English |
| Type | info:eu-repo/semantics/acceptedVersion, doc-type:doctoralThesis, info:eu-repo/semantics/doctoralThesis, doc-type:Text |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0606 seconds