ESTRO 2023 - Abstract Book

S1849

Digital Posters

ESTRO 2023

The ADC for different stages and the tumour cell density within the treatment was calculated with the script. As seen in Fig. 2 the hypoxic areas (poorly vascularised areas) of the tumour show lower response to the treatment at all the fractions. The inhomogeneous response of the tumour due to the effects of poor oxygenation can be observed. The use of such tools can be very useful in order to design a voxel-wise personalised treatment. Conclusion We have developed a software in a widely used environment (Eclipse-ARIA), which allows the integration of relevant biological information in functional images for the prediction of tumour response in a routine workflow.

SINFONIA project funded from the Euratom research and training programme 2019-2020 under grant agreement No 945196

PO-2071 Semi-automatic DWI segmentations can be used for response assessment in nCRT in esophageal cancer

R. den Boer 1 , K. Ng Wei Siang 2 , M. Yuen 1 , A. Borggreve 1 , I. Defize 1 , R. van Hillegersberg 3 , J. Ruurda 3 , S. Mook 1 , G. Meijer 1

1 University Medical Center Utrecht, Radiation oncology, Utrecht, The Netherlands; 2 University Medical Center Groningen, Radiation oncology, Groningen, The Netherlands; 3 University Medical Center Utrecht, Surgery, Utrecht, The Netherlands Purpose or Objective Changes in apparent diffusion coefficient (ADC) signal over the course of neoadjuvant chemoradiotherapy (nCRT) have proven to be predictive for pathologic outcome in esophageal cancer patients. However, manual tumor segmentation on high b-value diffusion weighted MRI (DWI) scans is labor intensive and often subject to interpretations especially at later intervals where signal intensities might have dropped. This study presents the feasibility of a semi-automated workflow to assess the predictive value of a series of DWI scans during nCRT in esophageal cancer patients. Materials and Methods Twenty patients diagnosed with esophageal cancer planned to undergo nCRT and esophagectomy underwent one baseline scan and five weekly DW-MRI scans during nCRT. A Patch2Self algorithm was used to denoise the DWI-MRI images. For each scan the ADCmean was assessed based on a manual and semi-automatic segmentation of the b-500 image. Manual segmentations were performed for each separate scan by well-trained experts. For the semi-automatic segmentation, a well-established gradient-based method was used (PET Edge, MIM Software Inc) which only required a single seed placement in the center of the tumor. This segmentation was then non-rigidly propagated to the ADC maps of the follow-up scans based on the grey values of the successive b-0 images, as these images had the highest signal-to-noise ratio and most profound anatomical contrast. For each patient the average increase of ADCmean per week was assessed by applying a linear fit through all ADCmean data points for the separate workflows. Bland-Altman analysis was done to evaluate the agreement between the two segmentation methods. The predictive value of the weekly ADCmean increase for obtaining a pCR was assessed by the Area Under the Receiver Operating Characteristic (AUC).

Results