Electrical and Electronic Engineering - Masters by Research Theses
Permanent URI for this collection
Browse
Recent Submissions
Item Hypoxic-ischemic encephalopathy grading using novel EEG signal processing and machine learning techniques(University College Cork, 2025) Twomey, Leah; Popovici, Emanuel; Temko, Andriy; Qualcomm\IEEEPARstart{H}{ypoxic-Ischemic Encephalopathy} (HIE) is a critical brain injury in newborns resulting from oxygen or blood supply deprivation during birth. Traditional diagnosis methods for HIE require specialized expertise from a trained medical professional, and immediate intervention of treatment is required within the first 6 hours post primary hypoxic-ischemic insult. Electroencephalogram (EEG) is the gold standard for analysing brain pathologies for neonates and large windows of this complex signal must be assessed for accurate HIE diagnosis. Using an openly available dataset, this thesis proposes an automated approach for HIE grading, leveraging signal processing techniques and Artificial Intelligence (AI) to provide accurate and timely assessments, with an implementation enabling edge-device deployment for real-world clinical utility. Representing the 1 hour epoch of the EEG signal in the amplitude and frequency domain through Mel Spectorgram representation, the HIE grading task becomes an image recognition problem, where Convolutional Neural Networks have shown high accuracy and efficiency. An initial test accuracy of 84.85\% is achieved. Further enhancement of the signals rhythmic behaviour is necessary to increase the grading potential of the signal, thus the FM/AM sonification algorithm was implemented, transforming the EEG signal into an amplitude and frequency modulated audio signal. This pre-processing is designed to enhance the background rhythmic pattern of the signal, an essential feature used by the clinician for visual inspection. The two dimensional (2D) CNN is designed as a regressor to map the input image to a value on the HIE grading scale to enhance the model's ability to leverage the monotonic relationship between the grades. An optimised rounding function is implemented to define the final clinical grade as a novel post-processing technique. An overall test accuracy of 89.97\% based on a rigourous nested cross-validation evaluation framework is achieved, surpassing the current state-of-the-art by 6\%. The robustness and generalisability of the proposed method is obvious from the weighted F1-score of 0.8985 and a Kappa score of 0.8219. While enhancing the accuracy and speed of HIE diagnosis with this AI-driven approach, the goal is to make it readily available at the point of care. An inference time of 62.7 milliseconds is achieved for the quantized model on the Snapdragon processor highlighting its suitability for real-time, on-device HIE grading.Item Interactive EEG visualisation in Virtual Reality: design and implementation(University College Cork, 2025) Creed, Adam; Popovici, Emanuel; Factor, Andreea; Murphy, David; QualcommElectroencephalography (EEG) has long been a cornerstone in both neuroscience research and clinical diagnostics, providing invaluable insights into the electrical activity of the brain. EEG is the gold standard to detect and analyse brain pathologies such as seizures. An early detection and diagnosis of seizures is key in an effective treatment and decrease in morbidity and mortality. However, neonatal seizures are difficult to detect through clinical signs and even through EEG. The standard method of analysis of neonatal EEG is through visualisation on conventional two dimensional (2D) displays. Despite its importance, interpreting the complex and often nuanced signals produced by EEG remains a significant challenge, particularly when viewed on conventional 2D displays. This limitation can be particularly evident for clinicians who may lack extensive experience in EEG analysis. As the demand for more accurate and accessible diagnostic tools grows, the need for more intuitive methods of data visualisation becomes increasingly apparent. This thesis addresses these challenges by presenting the design, development, and comprehensive evaluation of an innovative neonatal EEG visualisation platform within a Virtual Reality (VR) environment, developed using Unity. The platform reimagines how EEG data can be presented, leveraging the immersive capabilities of VR to offer a fully three-dimensional (3D) space where users can interact with and explore brain activity data in real time. By moving beyond the constraints of 2D screens, this approach provides a more natural, immersive, and intuitive framework for both novice and expert clinicians alike. Recently, another method of analysis through sonification of EEG was proposed to speed up analysis of EEG. This allows users to hear the brain’s electrical activity as a dynamic auditory experience. This novel sonification technique not only provides an additional sensory modality for interpreting EEG data when used complementary with visualisation but also enhances the users’ ability to detect patterns and anomalies in the brain’s electrical signals that may be overlooked visually. Therefore, it was added to this platform. Furthermore, the platform integrates an AI-driven seizure detection algorithm. This feature maps detected seizure events onto a 3D brain model, offering clinicians a more spatially accurate representation of potential seizure zones. The ability to visualise these detections in a 3D context is expected to improve clinical decision-making. To validate the effectiveness of this VR-based EEG platform, a series of user evaluations was conducted. Participants, including both clinicians with EEG knowledge and those without prior experience, interacted with the system and provided feedback on its usability and functionality. The results were overwhelmingly positive, with the platform achieving a high System Usability Scale (SUS) score of 83, indicating strong user satisfaction. Participants also completed the NASA Task Load Index (NASA-TLX), with an average score of 36.65, reflecting a low perceived cognitive workload during the interaction. This suggests that the platform not only offers a rich and engaging user experience but also minimises the mental effort required to interpret complex EEG data. In addition to its intuitive user interface, the platform was rigorously tested across a range of hardware configurations, from high-end VR systems to more affordable, lowerend devices. It performed consistently well, demonstrating its potential for widespread adoption in diverse clinical environments, from cutting-edge research facilities to more resource-constrained healthcare settings.Item Low-power real-time seizure monitoring via AI-assisted sonification of neonatal EEG(University College Cork, 2024) Nguyen, Tien Van; Popovici, Emanuel; Temko, Andriy; QualcommSeizures in newborns are most frequently provoked by an acute brain injury or systemic insult, making them predominantly acute symptomatic in nature. Early identification and intervention are essential to reducing the risks of mortality and long-term disabilities. Continuous electroencephalogram (EEG) monitoring remains the gold standard for accurate neonatal seizure detection. This procedure is expensive, time-consuming, and requires a high level of expertise. Recent advancements in artificial intelligence (AI) and machine learning (ML) have shown substantial promise in expediting seizure diagnosis. These algorithms can provide decision assistance with expert-level accuracy and thus are increasingly used for EEG analysis. However, the lack of transparency and explainability in the decision-making process of ML algorithms hinders their widespread adoption in medical settings. Researchers at the Embedded.Systems@UCC Group proposed a novel method that integrates machine learning with sonification, an alternative EEG interpretation approach by converting the signals into sound, to address the black-box problem while simultaneously improving seizure detection performance. The study also found that human listeners achieved better performance using AI-assisted audio analysis compared to using the ML seizure detection algorithm alone. This study presents a real-time processing design that allows the AI-integrated sonification method to operate in parallel with EEG acquisition. The new design eliminates the operational time delay associated with offline processing in the original study. This is particularly beneficial when the diagnosis is time-critical or in a busy environment where medical professionals have a tight schedule. A low-power implementation of the proposed real-time system is also presented. This implementation, featuring a system-on-chip with a deep neural networks accelerator, has an average power consumption of 13mW. It can be powered using a battery or a general USB port, allowing for a compact design, suitable for deployment in space-constrained environments like the neonatal intensive care units (NICU).Item Analog mixed signal IC design for magnetic tracking in surgery: low area clocking for CT∆ΣM ADCs for in-vivo sensing(University College Cork, 2024) Ferro, Alessandro; Cantillon-Murphy, Padraig; O'Hare, DanielThe application for this work is an in-vivo sensor system for capturing magnetic signals used for tracking surgical instruments such as catheters in the body during minimally invasive surgical procedures. With the ascent of image-guided interventions, the precision in determining instrument pose has become paramount. The proposed system leverages low-frequency electromagnetic fields for magnetic tracking, ensuring non-ionising, line-of-sight-free measurements with millimetre-scale accuracy. The miniaturisation demands of electromagnetic sensors, crucial to this endeavour, necessitate dimensions less than 0.5 mm in diameter, with the entire system fitting within an area of 1 mm x 0.5 mm. Continuous Time Delta Sigma (CT∆ΣM) ADCs emerge as a promising solution given their power efficiency and reduced requirements on peripheral circuits. These ADCs boast advantages like built-in anti-alias filtering and simplified input buffer requirements. However, their susceptibility to clock jitter, especially in single-bit versions, presents challenges. This research introduces an original MATLAB and Simulink model for the CT∆ΣM. The model initially accounted for clock jitter employing a white noise spectrum. However, subsequent analysis unearthed a profound limitation: while FIR filters within the CT∆ΣM reduced periodic jitter’s impact, they were vulnerable to non-white noise spectrums. Although present in CT∆ΣM literature, this significant observation was previously uncharted for the signal bandwidth applications akin to this work. To address this, the research delves into creating a comprehensive Phase noise spectrum model using MATLAB. This model elucidates the profound limitations imposed on the ADC’s in-band noise by the clock source’s phase noise spectrum, a revelation that reshaped our understanding of the CT∆ΣM’s constraints. Originally, a detailed schematic design for the clock source was undertaken with the primary objective of optimising area and power consumption, relying on a Resistor Capacitor based clock source for the Frequency-Locked Loop (FLL). The prevalent belief was that the typical phase noise spectrum of clock sources would have a negligible influence on the CT∆ΣM ADC. However, post-chip measurement analysis dispelled this notion. This research discovered that the clock source’s phase noise spectrum profoundly affected the CT∆ΣM ADC’s performance, highlighting an overlooked yet critical interplay.Item Development of a wearable system for monitoring people with Parkinson’s at home(University College Cork, 2023) Sica, Marco; O'Flynn, Brendan; Tedesco, Salvatore; AbbVie; Enterprise IrelandParkinson's disease (PD) is a neurodegenerative disorder affecting the central nervous system. Besides impairing motor functions, PD is also characterized by a broad variety of non-motor symptoms, such as mood and cognitive disorders, hallucinations, and sleep disturbances. People with Parkinson’s (PwP) are evaluated using clinical assessments and self-administered diaries and, as a consequence, they receive the necessary pharmacological therapy to alleviate symptoms and enhance sleep quality. Tri-axial accelerometers and gyroscopes might be employed to objectively evaluate Parkinsonians’ condition and help clinicians in making decisions. PwP often have significant abnormalities in blood pressure (BP) due to comorbid age-related cardiovascular disease and orthostatic hypotension, which result in blurred vision, dizziness, syncope, and falls. Frequent BP monitoring may aid in the evaluation of such events and differentiate PD symptoms from those originated by hypotension. A number of commercially available devices designed specifically for PwP include accelerometers and gyroscopes for the estimation of main motor symptoms, gait parameters, and sleep quality; nevertheless, according to the authors' knowledge, neither commercially available systems nor published works include also photoplethysmograph (PPG) and electrocardiogram (ECG) sensors that can be used for Parkinsonians’ cardiovascular monitoring. In this work, the WESAA (Wearable Enabled Symptom Assessment Algorithm) system is introduced as a revolutionary tool for the remote monitoring of PD patients. It is comprised of two devices worn on the wrist and ankle, and its key purpose is to capture accelerations and angular velocities from these body parts, as well as PPG and ECG data. This information may be used off-line to predict common PD motor symptoms (such as tremor, bradykinesia, and dyskinesia), walking speed, sleep-wake cycles, and cuff-less BP measures. The system requirements, market overview, industrial design, hardware and firmware development, user experience, early results of the gathered inertial raw data, and validation of the PPG and ECG signals were looked at in detail in the present work. The created system fulfils all the defined user requirements, and the sensors used yielded results equivalent to gold standard technologies. This thesis also studied the PwP's viewpoints on the WESAA system which is crucial for usability and adherence, examining practical concerns such as size, design, and comfort, as well as emotional consequences, societal impact, and the significance of discretion. In addition, users discussed their data-sharing preferences and how wearable technology could enhance their lives (i.e., the necessity to give feedback, particularly on motor symptoms). The WESAA system thus presents a promising alternative for remote monitoring of PwP since it has the ability to assist physicians in decision-making in terms of medication and treatment, giving them potentially useful information about the motor symptoms and the overall health status of their patients. Future work involves the implementation of off-line solutions for the detection of PD motor symptoms, walking speed, sleep-wake cycles, and cuff-less BP; machine learning algorithms should be adopted and a broader data collection carried out.