Morteza Zabihi: Contributions to Biomedical Signal Analysis Using Nonlinear Dynamics and Machine Learning

Physiological measurements have been widely used for diagnosis and assessment of a variety of abnormalities such as cardiac arrhythmias, brain conditions, and sepsis morbidity and mortality. The early and accurate diagnosis of such anomalies can increase the chances of successful treatments. However, manual interpretation of clinical data is time-consuming and requires skilled personnel, hence costly. Furthermore, with information overload and a rise in the amount of clinical data, often relevant and credible data are overlooked.

Automatic detection and classification of different conditions can assist clinical experts in the diagnosis of a growing number of clinical recordings. The goal of Morteza Zabihi´s dissertation is to design and develop algorithms to help medical experts improve the diagnosis of prevalent medical conditions.

The contributions of the dissertation can be categorized into two parts. In the first part, a systematic approach is proposed for seizure detection using multi-channel electroencephalogram (EEG) recordings. The experimental results yield promising detection accuracy in differentiating seizure from non-seizure events compared to state-of-the-art methods.

In the second part, three assistive medical diagnosis algorithms are designed and implemented as open-source algorithms. The algorithms are (a) detection of heart anomalies using phonocardiogram (PCG) signals and their quality assessment, (b) atrial fibrillation detection using hand-held one-lead Electrocardiogram (ECG) signals, and (c) early prediction of sepsis in intensive care units. The developed solutions were ranked among the top three algorithms in a series of international PhysioNet challenges, and one of them won the 1st place in the challenge.

The doctoral dissertation of MSc Morteza Zabihi in the field of Computing and Electrical Engineering titled Contributions to Biomedical Signal Analysis Using Nonlinear Dynamics and Machine Learning will be publicly examined in the Faculty of Information Technology and Communication Sciences at Tampere University on Friday, 25 September 2020, at 13 o’clock for online public defense. The Opponent will be Professor Abdulnasir Hossen from Sultan Qaboos University, Oman. The Custos will be Professor Moncef Gabbouj from Tampere University.

The dissertation is available online: http://urn.fi/URN:ISBN:978-952-03-1681-5

The dissertation is available online as Microsoft Teams Meeting.