ESTRO 2024 - Abstract Book

S3163

Physics - Autosegmentation

ESTRO 2024

[2] DOOLAN, Paul J., et al. A clinical evaluation of the performance of five commercial artificial intelligence contouring systems for radiotherapy. Frontiers in Oncology, 2023, vol. 13, p. 1213068.

[3] TAHA, Abdel Aziz; HANBURY, Allan. Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool. BMC medical imaging, 2015, vol. 15, no 1, p. 1-28.

[4] PERA, Óscar, et al. Clinical Validation of Siemens’ Syngo. via Automatic Contouring System. Advances in Radiation Oncology, 2023, vol. 8, no 3, p. 101177.

[5] GUO, Hongbo, et al. The dosimetric impact of deep learning-based auto-segmentation of organs at risk on nasopharyngeal and rectal cancer. Radiation Oncology, 2021, vol. 16, no 1, p. 1-14.

[6] OKTAY, Ozan, et al. Evaluation of deep learning to augment image-guided radiotherapy for head and neck and prostate cancers. JAMA network open, 2020, vol. 3, no 11, p. e2027426-e2027426.

3118

Digital Poster

CNN vs manual segmentation for ES-NSCLC: performance assessment and current pitfalls

Federico Mastroleo 1,2 , Stefania Volpe 1,3 , Francesca Botta 4 , Mattia Zaffaroni 1 , Maria Giulia Vincini 1 , Matteo Ferrante 4 , Lisa Rinaldi 5 , Gaia Piperno 1 , Matthias Guckenberger 6,7 , Roberto Orecchia 8 , Barbara Alicja Jereczek-Fossa 1,3 1 IEO European Institute of Oncology IRCCS, Division of Radiation Oncology, Milan, Italy. 2 University of Piemonte Orientale, Department of Translational Medicine, Novara, Italy. 3 University of Milan, Department of Oncology and Hemato-Oncology, Milan, Italy. 4 IEO European Institute of Oncology IRCCS, Medical Physics Unit, Milan, Italy. 5 IEO European Institute of Oncology IRCCS, Radiation Research Unit, Milan, Italy. 6 University Hospital Zürich, Division of Radiation Oncology, Zurich, Switzerland. 7 University of Zürich, Division of Radiation Oncology, Zurich, Switzerland. 8 IEO European Institute of Oncology IRCCS, Scientific Direction, Milan, Italy

Purpose/Objective:

The increasing interest on automated segmentation has led to the creation of several commercial and non commercial tools. In this work we applied an in-house developed convolutional neural network (U-net) to a dataset of candidates to curative-intent stereotactic body radiotherapy (SBRT) for early-stage non-small cell lung cancer (ES NSCLC). The aims were to: 1) assess the performance of the U-net against a gold standard given by manual segmentation performed by a single Radiation Oncologist, and to 2) analyze which qualitative radiological parameters were associated with a decrease in the U-net performance, if any.

Material/Methods:

Pre-SBRT contrast-enhanced computed tomographies (CTs) were retrospectively retrieved. After the presence of informed consent was verified, the U-net and manual segmentations were compared by the Dice coefficient using 3D Slicer v.5.2.2, with a Dice ≥0.8 being considered as satisfactory. Qualitative radiological parameters included, but were