Development of machine learning-based open-source tools for processing and analysis of extracellular neuronal signals to facilitate disease monitoring

Tools

Fabietti, M.I. ORCID: 0000-0003-3093-5985, 2022. Development of machine learning-based open-source tools for processing and analysis of extracellular neuronal signals to facilitate disease monitoring. PhD, Nottingham Trent University.

Preview

Text
marcos_fabietti_2023.pdf - Published version
Download (11MB) | Preview

Abstract

Neuronal signals are recordings of the electrical activity of the brain, which allow gaining insight into a diverse range of information. Like other physiological signals, extensive processing and analysis must be carried out in order to extract useful information. In this context, the neuroscience community has developed different open-access tools and pipelines for the different steps involved to facilitate the studies and make more advancements in the field. The aim of the research reported in this thesis is the development of tools and pipelines to facilitate the use of machine learning techniques in chronically recorded invasive signals for early disease detection. This includes the selection of the state-of-the-art for artefact detection and removal, the processing of the signal to feed the models, and lastly a robust machine learning based classifier. The main contributions of this thesis to the application of machine learning in neuronal signal processing include an open-access tool for benchmarking the performance of artefact detection and removal with ML with over 120 articles, the creation of a toolbox with novel methods to detect and remove artefacts from extracellular neuronal signals recorded in the form of local field potentials, a novel channel independent artefact removal method based on the forecasting of normal activity to replace affected segments, an innovative ML pipeline to detect and classify brain states from the processed local field potentials, and lastly finding novel biomarkers from these models and properly assess them against the existing literature.

Item Type:

Thesis

Creators:

Fabietti, M.I.

Contributors: