Mental health illnesses have long been among the most serious and prevalent public health issues. In particular, depression is a leading cause of disability and significantly increases the risk of suicidal ideation and suicide attempts. According to Statistics Netherlands’ review of 2023, young people’s mental health has not fully recovered since the pandemic. Numerous surveys have shown a rise in mental health issues compared to two years earlier, especially among young women. Specifically, 18 percent of young people in the Netherlands (aged 12 to 24) were considered mentally unhealthy in 2023, up from 11 percent in 2019 and 2020.

With the increasing availability of data on individuals’ mental health, artificial intelligence (AI) and machine learning (ML) techniques are being used to deepen our understanding of mental health conditions and enhance patient care. These technologies are also aiding clinicians and doctors in making more informed clinical decisions.

In recent years, Graph Neural Networks (GNNs) that focus on functional magnetic resonance imaging (fMRI) data have become popular for diagnosing depression in both AI and medical communities. However, existing GNN-based approaches face challenges related to explainability and generalization. Specifically, GNNs are often viewed as black boxes with poor explainability, meaning clinicians do not understand the rationale behind their decisions. Additionally, these trained machine learning models typically exhibit poor generalization, performing well in local hospitals but experiencing significant performance drops when tested on data from other hospitals.

This project aims to advance existing AI approaches from two perspectives. First, we aim to develop a new, explainable GNN to enhance both explainability and generalization. Second, we aim to further improve the detection accuracy by incorporating other modalities of neuroimaging data, such as structural magnetic resonance imaging (sMRI). Thus, creating a multi-modal depression detection system is another key objective.

Researchers: