ESTRO 2022 - Abstract Book

S739

Abstract book

ESTRO 2022

Conclusion Two explainability methods were employed to explore which CT regions were the most relevant in HPV status prediction by a 3D DL model. Our study showed a promising classification performance and volumetric overlap between the resulting heatmaps and the GTV pt and GTV ln . These findings contribute to address reliability concerns of DL in diagnostics and bring closer its application in a clinical setting.

PD-0821 Head & neck squamous cell carcinoma: Comparison of multimodal imaging with pathological specimens

E. Terzidis 1 , J. Friborg 1 , A.B. Olin 2 , I.R. Vogelius 1 , G. Lelkaitis 3 , C. Buchwald 4 , H.H. Johanesen 2 , B.M. Fischer 2 , I. Wessel 4 , J.H. Rasmussen 4 1 Rigshospitalet, Department of Oncology, section of Radiotherapy, Copenhagen, Denmark; 2 Rigshospitalet, Department of Clinical Physiology & Nuclear Medicine, Copenhagen, Denmark; 3 Rigshospitalet, Department of Pathology, Copenhagen, Denmark; 4 Rigshospitalet, Department of Otorhinolaryngology, Head & Neck Surgery and Audiology, Copenhagen, Denmark Purpose or Objective Despite development in imaging modalities, precise target definition remains a major challenge in head and neck cancer. In this study, tumor volumes defined from multimodal imaging with PET/MRI prior to surgery are compared with the actual pathological tumor volumes from surgical specimens. The mismatch between the volumes defined from imaging and pathology is estimated and the clinical impact of potential mismatch is evaluated. Materials and Methods Twenty-eight patients with head and neck squamous cell carcinoma were scanned on an integrated PET/MRI system, prior to surgery. Three GTV’s were delineated defined from the MRI (GTV MRI ), the PET (GTV PET ) and one by utilizing both anatomical images and clinical information (GTV ONCO ). Surgical specimens were extracted en bloc and scanned with the same PET/MRI and co-registered to the patient images based on predefined anatomical landmarks. Each specimen was sectioned in blocks which were then sliced (4- µ m thick) and stained with haematoxylin and eosin. All slices were digitalized, scanned and pathological tumor volume was delineated by a trained head and neck pathologist. Pathological tumor areas were interpolated to yield a 3D tumor volume (GTV PATO ), which was imported in Eclipse TPS (Varian Medical Systems, Palo Alto, CA) as a structure. The GTV PATO was compared with the GTV’s defined from imaging. Additionally, the mismatch between GTV’s were assessed by dividing the GTV PATO that is outside the imaging GTV with the whole volume of the corresponding GTV. To validate current practice of a 5 mm CTV expansion, a margin of 5 mm was added to the GTV ONCO and the mismatch was re-assessed. Results For consistency, only T-sites were included in this analysis (N=27). Further thirteen patients were excluded before the analysis, because fragmentation of the specimen during histologic processing or poor interpolation and insufficient number of points to form a structure in Eclipse. For the fourteen patients that were included in the analysis the mean volume of the GTV ONCO was larger than the other GTV’s. In four patients the GTV PATO was larger than the GTV MRI and for one patient the GTV PATO was larger than the GTV PET . The mean mismatch of the GTV PATO when evaluated to the GTV PET , GTV MRI and GTV ONCO was 29.7%, 39.5% and 7.5% respectively (Table 1). However, after the addition of 5mm margin all GTV PATO ’s were encompassed in GTV ONCO .