The impact of MRI-CT registration errors on deep learning-based synthetic CT generation

摘要

Purpose To investigate the impact of image registration on deep learning-based synthetic CT (sCT) generation.Methods Paired MR images and CT scans of the pelvic region of radiotherapy patients were obtained andnon-rigidly registered. After a manual verification of the registrations, the dataset was split into two groupscontaining either well-registered or poorly-registered MR-CT pairs. In three scenarios, a patch-based U-Netdeep learning architecture was trained for sCT generation on (ⅰ) exclusively well-registered data, (ⅱ) mixturesof well-registered and poorly-registered data or on (ⅲ) poorly-registered data only. Furthermore, a failure casewas designed by introducing a single misregistered subject in the training set of six well-registered subjects.Reconstruction quality was assessed using mean absolute error (MAE) in the entire body and specifically inbone and Dice similarity coefficient (DSC) evaluated cortical bone geometric fidelity.Results The model trained on well registered data had an average MAE of 27:6 - 2:6HU on the entirebody contour and 79:1 - 16:1HU on the bone. The average cortical bone DSC was 0:89. When patients withregistration errors were added to the training, MAE's were higher and DSC lower with variations by up to 36HUfor the average MAEbone. The failure mode demonstrated the potential far-reaching consequences of a singlemisregistered subject in the training set with variations by up to 38HU for MAEbone.Conclusion Poor registration quality of the training set had a negative impact on paired, deep learning-basedsCT generation. Notably, as low as one poorly-registered MR-CT pair in the training phase was capable ofdrastically altering a model.