Selection of 51 predictors from 13, 782 candidate multimodal features using machine learning improves coronary artery disease prediction.