ESTRO 2024 - Abstract Book

S3099

Physics - Autosegmentation

ESTRO 2024

[6] – Nikolov, S. et al. (2021) ‘Clinically applicable segmentation of head and neck anatomy for radiotherapy: Deep Learning algorithm development and Validation Study’, Journal of Medical Internet Research, 23(7). doi:10.2196/26151.

2399

Digital Poster

Head and neck automatic multi-organ segmentation on dual-energy CT (DECT)

Anh Thu Lê 1 , Killian Sambourg 1 , Roger Sun 1,2 , Rahimeh Rouhi 1 , Eric Deutsch 1,2 , Nathalie Fournier-Bidoz 1,2 , Charlotte Robert 1,2 1 Université Paris-Saclay, Gustave Roussy Institute, Inserm, Molecular Radiotherapy and Therapeutic Innovation, Villejuif, France. 2 Gustave Roussy Cancer Campus, Université Paris-Saclay, Gustave-Roussy, Department of Radiation Oncology, Villejuif, France

Purpose/Objective:

The definition of target volumes and organs at risk (OAR) is a crucial step in the preparation of radiotherapy treatment. However, this stage suffers from high inter-observer variability due to the limited contrast of conventional CT acquisitions. By acquiring images at two complementary voltages, dual-energy CT (DECT) makes it possible to reconstruct images offering contrast-enhanced structures such as virtual monoenergetic images (VMI)[1]. The clinical implementation of this technology could facilitate and harmonize manual segmentation, but also open up new prospects for automatic segmentation [2]. This study was conducted to test in a single-center setting the performance of a commercial automatic segmentation software (Annotate ART-Plan, TheraPanacea, Paris, France) on images derived from the dual energy Siemens Somatom go.Sim scanner (Siemens Healthcare, Forcheim, Germany).

Material/Methods:

In this study, the performance of ART-Plan v1.11.5 was quantitatively evaluated on various DECT-derived images from 74 head and neck patients, who benefited from a DECT acquisition between April 2023 and August 2023 in our center: polyenergetic 80kV DirectDensity (Sm36 filter), polyenergetic 80kV (Br36 filter) and 40keV VMI (Qr40 filter). In our clinical protocol, the 80kV DirectDensity serie is used as reference for dosimetry replacing the standard 120kV acquisition. The 80kV, which is the first voltage of the dual acquisition, and 40keV series are reconstructed as a help for segmentation. DirectDensity (DD) is an algorithm-based feature of Siemens scanners (Somatom models) that allows to use the same calibration curve for scans of any kV setting. VMI was chosen at 40keV as it can be seen in literature [3] as one of the DECT images with the best contrast in head and neck cancer. A preliminary study was carried out on 15 patients to identify the OARs for which the automatic segmentation still needs to be improved. At the time, DirectDensity was also used on the reference images. To this end, two senior doctors had the task to score 57 OAR (A=Perfect contours, B=Minor corrections, C=Major corrections or contour redone completely). A list of 13 head and neck structures was derived from this analysis, including 12 structures with B and C scoring (list given in Table 1) and the brain, which was perfectly segmented (score A) for all patients. The final contours corrected by the doctors on the 80kV DD in the clinical workflow were finally used as ground truths and compared with the automatic contours on all the images with a Dice Similarity Coefficient (DSC).