Public defence in Computational Science, M.Sc. Athanasios Gotsopoulos

Title of the doctoral thesis: Interpretable artificial neural networks for fMRI data classification

Doctoral students: Athanasios Gotsopoulos
Opponent: Prof. Jussi Tohka, University of Eastern Finland, Finland
Custos: Prof. Jouko Lampinen, Aalto University School of Science, Department of Neuroscience and Biomedical Engineering

In recent years, machine learning has demonstrated remarkable success across various domains, achieving even superhuman performance in everyday classification tasks such as image recognition. However the application of machine learning for classification of neuronal activity measured by functional magnetic resonance imaging (fMRI) has been limited. 

In this work, we apply machine learning for classification of fMRI data, investigating neural network architectures that address limitations of conventional fMRI data analysis and enhance interpretability of the classification models in terms of brain regions that contribute to the classification results. Additionally, the study introduces novel neuroscientifically motivated architectures, aiming to enhance the understanding of neural processes. 

We demonstrate machine learning techniques tailored for the analysis of neuroscience data, addressing multivariate activity patterns that characterize complex cognitive tasks. The objective is to encourage further research in employing machine learning as a tool for gaining deeper insights into the complexities of the human brain. 

The results of this work underscore the efficacy of the proposed neural network-based classification methods in discerning task-related activation patterns from fMRI data. Furthermore, the application of importance extraction methods promote transparency of the classification models by highlighting the contribution of the brain regions involved in the task. 

This research is highly relevant to current trends in the field, as it bridges the gap between neuroscience and explainable artificial intelligence. By offering novel perspectives on data analysis and introducing neuroscientifically motivated architectures, this work contributes valuable insights to the evolving landscape of brain research.

Key words: Brain, fMRI, machine learning, classification, neural networks, importance maps

Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/

Contact details: 

Doctoral theses in the School of Science: https://aaltodoc.aalto.fi/handle/123456789/52