Machine learning feature extraction in naturalistic stimuli for human brain mapping using high-density diffuse optical tomography

Studying brain development requires child-friendly imaging modalities and stimulus paradigms. High density diffuse optical tomography provides enhanced image quality over fNIRS and is validated extensively against fMRI in adults. Movie viewing reduces head motion and increases task engagement. Movie features are tracked and correlated with brain activity to map multiple processing pathways in parallel. We propose machine learning methods to extract high-level audiovisual features to avoid the time-consuming, subjective task of manual coding these feature regressors. Using a Faster Region-based Convolutional Neural Network, we achieve high correlation values between manually and automatically generated face regressors and regression coefficient maps.