ESTRO 2023 - Abstract Book

S1396

Digital Posters

ESTRO 2023

Conclusion This project has developed an automated method for generating the MidP image using NiftyReg software for deformable image registration. The results show that image quality is improved when using a MidP dataset compared to Ave dataset or phase image, and the median averaging approach is more robust to mis-registration artefacts. The findings also validate the accuracy of the reconstruction of the mid-position to within 0.65 mm. [1] Wolthaus, J W H (2009). Four-dimensional imaging in radiotherapy for lung cancer patients. Nederlands Kanker Instituut.

PO-1686 Establishing a reliable DIR methodology applied to pancreatic recurrence evaluation

S. Poeta 1 , M. Manderlier 2 , C. Bouchart 2 , A. Gulyban 1 , Y. Jourani 1 , N. Reynaert 1

1 Institut Jules Bordet, Medical Physics, Brussels, Belgium; 2 Institut Jules Bordet, Radiation Oncology, Brussels, Belgium

Purpose or Objective The aim of this work is to establish a reliable methodology for pancreatic recurrence (PR) back propagation from recurrence to initial image sets, using deformable image registration (DIR) including evaluation metrics, use of regions of interest (ROI) and define limitations. Materials and Methods Loco-regional PR after isotoxic high-dose stereotactic body radiotherapy (iHD-SBRT) ( Bouchart C. et al TAM 2021 ) was identified in seventeen patients included in this retrospective study. The iHD-SBRT treatment was based on CT and MRI with abdominal compression. An additional oncological resection of the primary tumor was performed in ten patients. At the time of recurrence, CT or MRI was performed for ten and seven patients respectively. Single modality antichronologic image registration was applied (CT <- CT or MR <- MR), for recurrence mapping using MIM (version 7.1.5, MIMvista Inc, Cleaveland OH USA). An initial local rigid registration (RIR) was performed guided by either the great vessels (GVs) or pancreas, followed by automatic DIR. For each patient the pancreas (corresponding residual anatomical segments in case of surgery) and GVs were defined on both image sets, while initial tumour and recurrence only in the corresponding ones. The automatic DIR was verified and adjusted by a physicist and a specialized physician using MIM’s Reg Refine tool ( Johnson et al, JACMP, 2016 ). All delineated structures were propagated to the initial image set using the final DIR. All corresponding structures (initials and deformed) were evaluated using DICE similarity coefficient (DICE) for volumetric assessment, additional to symmetric Hausdorff distance 95% (HD95%) and mean distance to agreement (MDA). Furthermore, Jacobian determinant (JD) was used for DIR plausibility and finally visual inspection as qualitative assessment. GVs provided a consistent guide for DIR adjustments, while pancreas could help small DIR adjustments whenever used for initial RIR. DICE showed good and moderate agreement for GVs and pancreas, respectively. HD95 showed large variation for both structures up to 21mm and 16mm while it remained reasonable for MDA with values up to 3.2mm and 5.7mm for GVs and pancreas, respectively. Mean Jacobian confirmed plausible DIR, with values ranging from 0.2 - 1.3, 0.6 – 1.5 and 0.4 – 1.2, for recurrence, GVs and pancreas, respectively. Recurrence and pancreas had a predisposition for shrinkage while GVs remained stable, reinforcing the role of GVs as a reliable landmark. Visual inspection resulted in clinically satisfactory agreement within the ROI. Results For every case the initial RIR and automatic DIR required manual adjustment.