ICHNO-ECHNO 2022 - Abstract Book

S30

ICHNO-ECHNO 2022

Results Forty-two patients were included. The prevalence of occult metastases in PET/CT- and MRI-negative necks was 3.1% and 5.9%, respectively. Negative predictive values in neck node re-staging were 96.9% (PET/CT) and 94.1% (MRI). Conclusion Both PET/CT and MRI showed high negative predictive values for occult neck metastases in re-staging for recurrent laryngeal cancer after radio(chemo)therapy prior to salvage laryngectomy. These imaging modalities could help to avoid unnecessary interventions to the neck and its associated complications in selected patients.

PO-0062 Evaluation of deformable image registration in DWI and T2 MRI for head and neck cancer radiotherapy

M. Naser 1 , K. Wahid 1 , S. Ahmed 1 , V. Salama 1 , C. Dede 1 , R. Lin 2 , B. McDonald 1 , A. Mohamed 1 , D. Thill 3 , N. O’Connell 3 , V. Willcut 3 , J. Christodouleas 3 , C. Fuller 1 1 MD Anderson Cancer Center, Radiation Oncology, Houston, USA; 2 MD Anderson Cancer Center, Biostatistics, Houston, USA; 3 Elekta AB, Elekta AB, Stockholm, Sweden Purpose or Objective MRI-guided adaptive radiotherapy for head and neck cancer (HNC) has the potential to deliver more personalized treatment by acquiring both anatomic and functional MRI scans. T2-weighted MRI can provide excellent soft-tissue visualization for more accurate definition of target volumes and organs at risk (OARs), while imaging biomarkers based on diffusion-weighted imaging (DWI) can be used for treatment plan adaptation based on biological response. Despite the quantitative capabilities of DWI, geometric distortion remains a considerable issue with these images. Therefore, in this study, we systematically investigate the effect of various deformable image registration (DIR) algorithms to co-register DWI and T2 images acquired in the same imaging session. Materials and Methods We compared DIR algorithms from the software packages ADMIRE and Velocity AI, and B-spline and Affine algorithms from 3D Slicer applied to T2-weighted MRI and DWI images in 20 HNC patients by utilizing a workflow implementing radiotherapy structures of interest including the parotid glands, submandibular glands, spinal cord, brainstem, and primary gross tumor volume. Ground truth contours were generated by a physician expert observer on both image sequences. Three additional observers provided segmentations for a subset of five cases for inter-observer variability studies. For each registration algorithm, structures were propagated from T2-weighted images to DWI images. These propagated structures were then compared with ground truth DWI structures using a variety of evaluation metrics, including Dice similarity coefficient (DSC), false-positive DSC, false-negative DSC, surface DSC, 95% Hausdorff distance, and mean surface distance. Wilcoxon signed- rank tests were used to compare registration algorithms with implicitly registered image results.

Results The ADMIRE DIR registration algorithm demonstrated significant gain for the greatest number of structures for the greatest number of metrics compared to implicitly registered images, while Velocity and 3D Slicer algorithms did not demonstrate significant improvement from implicitly registered images for most metrics and structures ( Table 1 ). Interobserver variability analysis revealed no significant difference between observers for most structures and evaluation metrics for all algorithms.