Purpose: In head and neck cancer, diffusion weighted MRI (DWI) can predict response early during treatment. Treatment-induced changes and DWI-specific artifacts hinder an accurate registration between apparent diffusion coefficient (ADC) maps. The aim of the study was to develop a registration tool which calculates and visualizes regional changes in ADC.
Methods: Twenty patients with stage IV HNC treated with primary radiotherapy received an MRI including DWI before and early during treatment. Markers were manually placed at anatomical landmarks on the different modalities at both time points. A registration method, consisting of a fully automatic rigid and nonrigid registration and two semi-automatic thin-plate spline (TPS) warps was developed and applied to the image sets. After each registration step the mean registration errors were calculated and ΔADC was compared between good and poor responders.
Results: Adding the TPS warps significantly reduced the registration error (in mm, 6.3 ± 6.2 vs 3.2 ± 3.3 mm, p<0.001). After the marker based registration the median ΔADC in poor responders was significantly lower than in good responders (7% vs. 21%; p<0.001).
Conclusions: This registration method allowed for a significant reduction of the mean registration error. Furthermore the voxel-wise calculation of the ΔADC early during radiotherapy allowed for a visualization of the regional differences of ΔADC within the tumor.
Keywords: Diffusion weighted MRI; Head and neck cancer; Nonrigid registration.
Copyright © 2013 Elsevier Ireland Ltd. All rights reserved.