Application of a convolutional neural network to the quality control of MRI defacing

Daniel J. Delbarre, Luis A. Santos, Habib Ganjgahi, Neil R. Horner, Aaron McCoy, Henrik Westerberg, Dieter A. Häring, Thomas E. Nichols, Ann-Marie Mallon

Abstract: Highlights•A deep learning model was developed to verify whether MRI defacing was successful.•Our CNN ensures that facial features are removed and that the brain is left intact.•The CNN has high accuracy (92%) when strict thresholds are applied.•The model correctly identified scans with defacing errors in 94% of cases.•When incorporated into our pipeline, the model reduces QC time by over one third.