Recommendations for ICA Denoising of Task-Based Functional MRI Data of Stroke Patients

Independent component analysis (ICA) denoising represents a useful tool in functional magnetic resonance imaging (fMRI) preprocessing pipelines. The most used methods for ICA denoising involve automatic artifact component selection for resting-state fMRI studies on healthy populations. However, these automated methods have important limitations that are magnified if they are to be used in clinical populations with commonly limited sample sizes, like stroke, and in task-based fMRI (tb-fMRI), which requires the execution of a given task. Nevertheless, this imaging modality can offer helpful information about the neural activation patterns post-stroke produced by the execution of a motor task, which could be relevant in clinical assessments. However, stroke populations show higher artifact presence in tb-fMRI studies and increased variability in neural activation patterns, which can complicate the identification of signal components. In this work, recommendations to manually identify features of artifact and signal components in ICA denoising of tb-fMRI stroke data are provided. These suggestions could allow a more reliable analysis and interpretation of tb-fMRI studies in stroke populations.