ESTRO 2024 - Abstract Book

S12 ESTRO 2024 of the neck's lymph drainage region are electively irradiated due to the risk of occult lymph node metastases. Current guidelines are mainly based on the prevalence of lymph node involvement in each lymph node level (LNL) for a given primary tumor location. The presentation will report on efforts to reduce and personalize the definition of the CTV-N based on the individual patient's state of tumor progression at the time of diagnosis, including the primary tumor's T-stage, location, and lateralization, as well as the location of clinically detected lymph node metastases. Three main aspects of the problem will be discussed. 1) We built a large multi-institutional database of 2756 HNSCC patients to quantify patterns of lymph node involvement per LNL depending on primary tumor characteristics. The web-based platform LyProX.org was developed to make the data publicly available and to allow interactive exploration and visualization of the data. 2) Based on the data, a statistical model is developed to estimate the individual patient's probability of occult lymph node metastases in each clinically negative LNL. The statistical model is based on Hidden Markov Models (HMM). Each LNL is described by a random variable that indicates the true involvement of the LNL including occult metastases. Lymphatic metastatic tumor progression is described via the transition matrix of the HMM, which is in turn parameterized via a directed graph representing the lymphatic drainage. The model parameters learned from the data are the probabilities of the tumor to spread to a LNL and between LNLs. 3) A phase II clinical trial on personalized volume-deescalated elective nodal irradiation in oropharyngeal SCC is currently in preparation. The primary endpoint of the study is the rate of N-site recurrences in unirradiated LNLs. The CTV-N will be defined such that the estimated risk of occult lymph node involvement in all unirradiated LNLs combined is <10%. Main features of the trial are that a) ipsilateral level IV is not irradiated if level III is clinically negative, b) contralateral levels III/IV are not electively irradiated unless the upstream levels II/III are involved, c) unilateral irradiation is performed in patients with lateralized tumors and clinically negative contralateral neck, d) level I and V are irradiated in fewer patients compared to current guidelines. Invited Speaker

3293

Automatic delineation: How should it be used and possible dangers or drawbacks

Ditte Sloth Møller 1,2 , Anne Ivalu Sander Holm 1 , Lise Bech Jellesmark Thorsen 1

1 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark. 2 Aarhus University, Department of Clinical Medicine, Aarhus, Denmark

Abstract:

The use of high-quality AI segmentation algorithms results in a higher degree of agreement among individual delineators and better adherence to delineation guidelines. It also saves time for the individual delineator, freeing up time for other tasks in increasingly strained healthcare systems. However, such automation comes with new risks and potential drawbacks. Algorithms may perform worse than anticipated, especially if they are applied in patient groups or settings they were not developed for. With the introduction of AI segmentation in the daily clinic, a sense of false security could develop, resulting in inadequate quality and correction of AI delineations.