Predicting cortical ROIs via joint modeling of anatomical and connectional profiles

Our predictive models via joint modeling of anatomical and connectional features can estimate functional ROIs with high accuracy in comparison with other methods. Validation results by using fMRI-derived functional ROIs as benchmarks. Benchmark (white bubbles) and predicted ROIs (color-coded as the same as (f)) of five functional networks are illustrated on one subject respectively selected from Dataset 1 and Dataset 2 (a-e). For visualization convenience, the predicted ROIs obtained via joint model without PCA model (joint in short), via anatomical features (T1 in short) and via connectional features (DTI in short) and benchmark ROIs are shown in the first row (a1-e1). The predicted ROIs obtained via joint model with the coordinate PCA constraint (joint with constraint in short) and benchmark ROIs are shown in the second row (a2-e2). Comparison of the average prediction errors (mm) on the five networks is illustrated in (f).