Abstract
Obesity has become a pressing global health concern and is strongly associated with various physical and neurological impairments. Despite the prevalence of traditional interventions, such as diet modifications and surgical procedures, long-term weight regain often occurs in individuals with obesity after weight loss, highlighting the need to better understand the biological mechanisms associated with the condition for the development of more effective interventions.
Furthermore, emerging research evidence suggests that obesity is linked to brain alterations and neurological dysfunctions, underscoring the need for developing innovative brain-based approaches to unravel the complex neural mechanisms underlying this condition and to aid in the development of more effective obesity interventions.
This thesis leverages artificial intelligence (AI) techniques to investigate brain patterns related to obesity in female subjects, with a focus on uncovering intricate neural relationships and reducing bias inherent in traditional statistical methods. Using EEG brain imaging data, this thesis proposes a comprehensive framework to analyze obesity-associated brain activity, offering new insights into its neural underpinnings and advancing methodologies in high-dimensional data analysis.
This thesis comprises three interconnected studies. In Study One, an efficient feature selection method was developed to identify non-redundant and informative features from diverse high-dimensional datasets, including EEG-based brain functional connectivity (FC) data for obesity classification. Study Two applied this feature selection method to uncover key FC patterns associated with obesity, considering variations in metabolic and weight-status, thereby providing a deeper understanding of the neural network alterations linked to the condition. In Study Three, a deep learning (DL)-based approach was introduced to learn robust feature representations directly from resting-state EEG data, with obesity-based EEG data used as the case study. This end-to-end model enabled the identification of intricate EEG patterns related to obesity.
Integrating the findings from these three studies, this thesis identifies several critical brain FC patterns fundamentally associated with obesity, independent of meal consumption, as well as patterns associated with obesity that could be influenced by meal consumption. These insights can inform future clinical efforts in the design of sustainable and personalized obesity treatments.
From a methodological perspective, the models developed in this research are made adaptable to various high-dimensional datasets across different domains, providing tools for future researchers to efficiently select features and extract meaningful EEG-based feature representations. By bridging neuroscience and AI, this work establishes a foundation for understanding the neural mechanisms of obesity and advancing AI-driven investigations in complex biomedical contexts.