Titulo Estágio
Exploring transfer learning techniques for emotional arousal classification in neuroimaging data
Local do Estágio
CISUC-DEI e CIBIT-ICNAS
Enquadramento
Emotions can be characterized as responses to environmental events (stimuli) that facilitate goal-directed behavior and individual adaptation to a changing environment. Typically, these responses involve a complex blend of cognitive, affective, behavioral, and physiological reactions. Emotions hold significant importance in various psychological and psychiatric disorders. To gain a deeper understanding of emotional responses, neuroimaging techniques such as electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI) can be employed to capture the intricate interactions within the central nervous system that underlie our emotional experiences.
Here, we focus specifically on exploring fMRI data, with an emphasis on functional and effective brain connectivity measures derived from it. These metrics offer valuable insights into the functional organization of the human brain and can provide crucial information for understanding emotional processing and the underlying mechanisms of neurological and psychiatric disorders that arise from abnormal emotional experiences. Moreover, leveraging these connectivity metrics holds the potential to develop intelligent systems capable of accurately identifying an individual's emotional state. Such systems have the capacity to pave the way for exploring novel therapeutic approaches tailored to individual emotional needs.
In this context, our objective is to analyze fMRI data and investigate functional and effective brain connectivity measures to enhance our understanding of emotional processing, elucidate the mechanisms underlying emotional disorders, and potentially develop intelligent systems for accurately assessing an individual's emotional state. The outcomes of this research have the potential to contribute significantly to the development of targeted therapeutic interventions that address emotional difficulties on an individualized level.
We will focus on a large fMRI dataset to train the models ([1] human connectome project, 1049 subjects) and then transfer them to the Biohab project dataset (in-house collected data, about 40 participants).
Objetivo
Explore and evaluate transfer learning techniques to improve the predictive capability of functional and effective connectivity measures across different datasets, which include:
- Identify the most relevant functional and effective connectivity measures,
- Develop a solution that can be applied across different datasets,
- Train and “transfer” the models between datasets, aiming for lower complexity models without compromising accuracy.
Plano de Trabalhos - Semestre 1
1. State-of-the-art review:
a. Functional Magnetic Resonance Imaging data
b. Connectivity measures
c. Mental state classification
d. Transfer-learning methods
2. Familiarization with the datasets to be used
3. Identification of the potential connectivity measures and classification/deep-learning models
Plano de Trabalhos - Semestre 2
4. Implementation, training and evaluation of models
5. Analysis of the results
6. Dissertation writing
Condições
Access to a GPU-enabled mainframe for running the models.
Observações
[1] https://www.humanconnectome.org/ Article: "Function in the human connectome: task-fMRI and individual differences in behavior" doi:10.1016/j.neuroimage.2013.05.033 url: https://pubmed.ncbi.nlm.nih.gov/23684877/
Orientador
Marco Simões
msimoes@dei.uc.pt 📩