Dimensionality reduction in Stroke Patients Neuroimaging Data

The understanding of the stroke lesions’ consequences is limited, and it relies mostly on behavioral reports and mere descriptive correlational information from neuroimaging techniques. Using functional magnetic resonance imaging, it has been shown that the functional synchronization between distinct regions of the brain, referred to as functional connectivity, is disrupted by stroke. Nevertheless, this kind of data is usually large and high-dimensional. Extracting useful information from the vast amount of information afforded by brain networks remains a great challenge so, in order to make sense of this complex neural data, we propose a dimensionality reduction approach. By using autoencoders, we found a low-dimensional representation encoding the fMRI data which preserves the signature features known to be present in stroke patients. We enhanced patients’ diagnostics, severity classification, and embedded properties by analyzing their latent representations. Furthermore, we showed how low-dimensional representation increased the accuracy of recovery prediction.