ESTRO 2024 - Abstract Book

S4440

Physics - Machine learning models and clinical applications

ESTRO 2024

Using the DIVE-map, the number of corrected voxels was reduced by at least 50% in more than half of the 30 fractions (16 for rectum, 18 for bladder), while keeping the same dosimetric quality as the plan with manually corrected organs (differences in V60Gy below +2%).

In conlusion, the method successfully reduced corrections in auto-segmented organs without impairing plan dosimetric quality.

Keywords: auto-segmentation, dose prediction, quality

References:

[1] Nguyen, D., et al., Three-dimensional radiotherapy dose prediction on head and neck cancer patients with a hierarchically densely connected u-net deep learning architecture. arXiv preprint arXiv:1805.10397, 2018. [2] Huet ‐ Dastarac, M., et al., Patient selection for proton therapy using Normal Tissue Complication Probability with deep learning dose prediction for oropharyngeal cancer. Med Phys, 2023. 50(10): p. 6201-6214. [3] Archambault, Y., et al., Making on-line adaptive radiotherapy possible using artificial intelligence and machine learning for efficient daily re-planning. Med Phys Intl J, 2020. 8.

822

Digital Poster

Deep learning-based optimization of field geometry for total marrow irradiation delivered with VMAT

Nicola Lambri 1,2 , Giorgio Longari 3 , Monica Bianchi 1,4 , Andrea Bresolin 1 , Simone Buzzi 1,4 , Damiano Dei 1,2 , Pasqualina Gallo 1 , Francesco La Fauci 1 , Francesca Lobefalo 1 , Lucia Paganini 1 , Sara Parabicoli 1,4 , Marco Pelizzoli 1,4 , Giacomo Reggiori 1,2 , Stefano Tomatis 1 , Caterina Zaccone 1,4 , Marta Scorsetti 1,2 , Daniele Loiacono 3 , Pietro Mancosu 1 1 IRCCS Humanitas Research Hospital, Radiotherapy and Radiosurgery Department, Milan, Italy. 2 Humanitas University, Department of Biomedical Sciences, Milan, Italy. 3 Politecnico di Milano, Dipartimento di Elettronica, Informazione e Bioingegneria, Milan, Italy. 4 Università degli Studi di Milano, Dipartimento di Fisica “Aldo Pontremoli”, Milan, Italy

Purpose/Objective:

Total marrow irradiation (TMI) and total marrow lymphoid irradiation (TMLI) are RT treatments used to selectively target the bone marrow and lymph nodes in conditioning regimens for allogeneic hematopoietic stem cell transplantation in acute leukemia [1]. To cover the large PTV of TMI/TMLI with VMAT, a complex field geometry - i.e., isocenters positions and jaws apertures – is required. Typically, five isocenters and ten overlapping fields are needed for the upper body, while, for obese patients, two specific isocenters are placed on the arms. Therefore, the creation of a field geometry is clinically challenging and is performed by a medical physicist (MP) with high level of expertise in TMI/TMLI. To address this, we developed convolutional neural networks (CNNs) for automatically generating the field geometry of TMI/TMLI.