Deep neural networks for spatio-temporal analysis of rs-fMRI in ADHD

Deep learning algorithms have been successful in tackling a wide range of problems in different fields, but lack of interpretability and their reliance on large data limits their performance on brain imaging applications. Tailored deep learning solutions need to be developed for the analysis of bo...

Full description

Saved in:
Bibliographic Details
Main Author: Abhay Mallikarjun Somaradhya Aradhya
Other Authors: Mahardhika Pratama
Format: Thesis-Doctor of Philosophy
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/154986
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Deep learning algorithms have been successful in tackling a wide range of problems in different fields, but lack of interpretability and their reliance on large data limits their performance on brain imaging applications. Tailored deep learning solutions need to be developed for the analysis of both the spatial and temporal activities of the brain. In this thesis, novel data driven deep learning approaches have been proposed to increase feature separability, optimize deep learning network topology, improve classification performance and provide interpretability. The proposed approaches have achieved a 10% improvement in classification performance over previous methods on the ADHD200 dataset. Further, the spatio-temporal relationship between the independent brain regions have been analyzed to determine the causal factors of ADHD. The deep learning approaches proposed in this thesis play a vital role in the analysis of spatio-temporal data and can be adapted in other fields with similar challenges.