Title:	Using Machine Learning and Artificial Intelligent to Understand the Control of Fixational Eye Movements in Health and Disease
DNr:	Berzelius-2024-63
Project Type:	LiU Berzelius
Principal Investigator:	Yiting Wang <yiting.wang@ki.se>
Affiliation:	Karolinska Institutet
Duration:	2024-04-11 – 2024-11-01
Classification:	10203
Homepage:	https://ki.se/cns/marianne-bernadotte-centrum-mbc
Keywords:

Abstract

Even when we attempt to maintain a steady gaze on a single point, we make small eye movements beyond our awareness commonly known as fixational eye movements. Research have shown that such small and involuntary movements are closely linked to various aspects of visual perception, attention and cognition, and thus potentially carry important information about brain function and neurological disease. However, we are still far from understanding the dynamic nature of fixational eye movements and how small-amplitude eye movements may be affected by neurological conditions. Moreover, there is essentially a complete lack of studies investigating fixational eye movements using state-of-the-art modeling techniques based on machine learning and predictive modeling. Such techniques, we believe, will be key to finding structure and clinically useful patterns in the complex and irregular eye movement trajectories produced during visual fixation. In this project, we will use deep learning and reinforcement learning to explore the nature and clinical relevance of these types of eye movements. More specifically, supervised machine learning methods will be used to predict early-stage Parkinson’s disease on the basis of discriminative patterns in fixational eye movements. What's more, reinforcement learning is planned to be applied to simulate human fixational eye movements and capture important properties relating to their spatial and temporal dynamics. Deep learning models will be utilized in both supervised learning and reinforcement learning.