OHBM 2023 | 2905 | Symposium | Eloy Geenjaar | Voxelwise dynamical factor learning and multimodal … скачать в хорошем качестве

OHBM 2023 | 2905 | Symposium | Eloy Geenjaar | Voxelwise dynamical factor learning and multimodal …

Title: Voxelwise dynamical factor learning and multimodal fusion in fMRI. Session: Symposia Speaker: Eloy Geenjaar Abstract: A growing body of work in computational neuroscience is utilizing recurrent neural networks to understand the underlying low-dimensional temporal factors in their data. Similar ideas have existed in the neuroimaging community, although often done using matrix factorization techniques, such as independent component analysis (ICA) or principal component analysis (PCA). Moving to newer techniques such as recurrent neural networks allows us to both find non-linear factors, and learn low-dimensional dynamical systems of fMRI with more complex temporal relationships. Given the relative high dimensionality of fMRI data, being able to succinctly and accurately summarize fMRI data is important to aid in a clinically useful understanding of neuropsychiatric disorders, as showcased by the success of linear matrix factorization techniques. Namely, due to its ability to record whole-brain activity, fMRI may be especially poised to help us understand potentially aberrant dynamics in neuropsychiatric disorders. This talk will explain and discuss dynamical factor learning from fMRI data, introduce multimodal fusion, explain the clinical relevance of understanding interactions between a variety of modalities and fMRI data, and provide a perspective on dynamic multimodal fusion.