Thesis title: Advanced signal processing techniques to support diagnosis and prognosis of patients with Disorders of Consciousness
Disorder of Consciousness (DoC) is a clinical condition characterized by alterations in arousal and/or awareness that affects people who survive an acquired brain injury. Depending on the level of consciousness compromission, DoC can be categorized into different states, including coma (in the acute phase), Unresponsive Wakefulness Syndrome (UWS), and Minimally Conscious State (MCS). Coma is defined as a state of absence of both arousal and awareness. The UWS is a condition in which patients demonstrate arousal but remain unresponsive, showing only reflex movements, and no awareness of themselves or their surroundings. The MCS is characterized by the presence of discernible non-reflexive behavior, manifested by the recovery of arousal in conjunction with the eye-tracking ability and command following, but inconsistent reproducibility of this behavior.
Clinical scales are regarded as the gold standard for the diagnosis of DoC, but they are strongly dependent on behavioral markers, which makes them susceptible to alteration by a multitude of factors. This results in a misdiagnosis rate of approximately 40%, with evident ethical and clinical implications for patients' prognoses, treatments, and end-of-life decisions.
A substantial body of literature has been dedicated to investigating brain activity through electroencephalography (EEG) which has proven to be a valuable support for the clinical diagnosis and prognosis of patients with DoC. Quantitative EEG analysis utilizes mathematical processing and computational techniques to derive numerical measures from various domains, such as the spectral and the connectivity domains, non-linear dynamics, and event-related responses. EEG-based measures have been extracted and investigated in a variety of conditions, including the resting state and the execution of passive or active cognitive tasks, thus demonstrating their pivotal role in the extraction of prognostic and diagnostic biomarkers. EEG features calculated at rest range from spectral and connectivity measures to microstates and nonlinear measures (i.e., measures of complexity). With respect to the cognitive tasks examined by the extant literature, auditory stimulation has been the most investigated, as it represents the most straightforward method to elicit event-related potentials (ERP) that disclose late stages of information processing that can be related to consciousness levels. Moreover, among cognitive tasks, active paradigms entail the execution of a command and have been demonstrated to be particularly efficient in the disclosure of covert awareness; in this context, the modulation of sensorimotor rhythms has been shown to have significant potential.
However, not all EEG-based measures have been equally investigated, nor have they all demonstrated an ease of integration into clinical practice. Numerous studies have yielded incoherent and non-conclusive results, often due to the limited extension of patients' cohorts. Furthermore, the majority of these studies are limited to a single time point, overlooking the temporal variability in patients’ responsiveness.
The novel frontiers in the research field of EEG-derived quantitative measures are represented by the optimization of the signal processing techniques and the methods of feature extraction. Indeed, the classical EEG techniques used to process signals derived from healthy subjects are not effective when applied to DoC patients, as their signals present a series of nonidealities (e.g., inter-trial variability, prolonged latencies, responsiveness fluctuations) that, in conjunction with the presence of artifacts, compromise the quality of the data. Consequently, the application of traditional signal processing techniques becomes ineffective, and the implementation of complex computational models is hindered by the limited size of the available datasets. Therefore, there is a fundamental necessity to implement signal processing pipelines suitable to manage the specific issues of the patients under study.
Within this framework, the main objective of the present thesis is to explore the EEG-based quantitative measures that can efficiently and effectively support the clinical diagnosis and prognosis of patients with DoC using advanced signal processing techniques tailored to address the specific issues presented by the EEG signal of DoC patients. This objective is pursued by exploring two different temporal perspectives, the former consisting in a single acquisition of the order of minutes, the latter represented by a monitoring protocol that samples the entire day. The two temporal perspectives encompassed the extraction of both EEG spectral and connectivity characteristics at two different levels of engagement, i.e., rest and responses to cognitive tasks.
The present thesis is divided into three sections. The first section examines quantitative measures extracted in the resting state condition at a single time point to disclose the potential of spectral and connectivity features in distinguishing between UWS and MCS patients and in predicting their 3-month outcome.
The second section involves a single time point analysis, as the previous section, but focused on the characterization of the P300 ERP. The first chapter of this section (Chapter 2) addresses the signal processing techniques used to enhance the potential of P300 in supporting DoC diagnosis by managing the inter-trial variability of the ERP response, acknowledged as the latency jitter phenomenon. Meanwhile, Chapter 3 examines the relationship between latency jitter and the clinical condition, aiming to ascertain the potential of latency jitter in supporting DoC diagnosis.
The third section analyzes the evolution of EEG characteristics over time with the aim of characterizing fluctuations in responsiveness in MCS patients. Chapter 4 is dedicated to the analysis of the variability in EEG characteristics extracted from the spectral and connectivity domains in the resting state condition, while Chapter 5 characterizes the fluctuations in the modulation of sensorimotor rhythms during the execution of a motor command task.