ICHNO-ECHNO 2022 - Abstract Book

S12

ICHNO-ECHNO 2022

OC-0025 Personalized elective clinical target volume definition in head & neck cancer

R. Ludwig 1 , B. Pouymayou 1 , J. Hoffmann 1 , M. Guckenberger 1 , L. Bauwens 2 , V. Grégoire 2 , J. Unkelbach 1

1 University Hospital Zurich, Radiation Oncology, Zurich, Switzerland; 2 Centre Léon Bérard, Radiation Oncology, Lyon, France Purpose or Objective Head & neck squamous cell carcinomas (HNSCC) spread lymphatically through the neck and form metastases in regional lymph nodes. Modern imaging techniques such as FDG-PET/CT can visualize sufficiently large metastases but fail to detect small occult ones. Therefore, large parts of the neck are electively irradiated as part of the nodal clinical target volume (CTV-N). Here, we report on activities within a newly formed ESTRO working group that aims at developing computational methods to support CTV definition. The goal of this specific project is to improve quantification of the risk of occult disease in lymph node levels (LNL), considering the individual patient's state of tumor progression. Materials and Methods Quantification of the risk of occult metastases relies on three main components. (1) detailed datasets of lymphatic progression patterns that report, on a patient-by-patient basis, the location of lymph node metastases together with primary tumor characteristics such as location, T-category, midsagittal plane extension, and HPV status; (2) sensitivity and specificity of diagnostic modalities to detect lymph node metastases, obtained from comparing pathological and clinical lymph node involvement; and (3) a statistical model that quantitatively describes lymphatic metastatic progression and the diagnostic process. Results To address (1), we developed a web-based platform to host, visualize, and share datasets of lymphatic progression patterns, currently containing two datasets of 287 and 263 oropharyngeal HNSCC patients from independent institutions. Figure 1 illustrates the use of the platform. To address (3), we develop a probabilistic model of metastatic progression based on Hidden Markov Models (Figure 2, upper panel). In this model, each LNL is represented by a hidden binary random variable that indicates if a LNL truly harbors tumor including occult metastases. Each LNL is further associated with a binary observed state, which indicates if the LNL harbors visible macroscopic metastases. Hidden microscopic state and observed macroscopic state are linked via (2), sensitivity and specificity. The parameters of the model are the probabilities for the primary tumor to spread to a LNL and the probabilities to spread from upstream to downstream LNL along the lymphatic drainage, which are learned from (1), datasets of lymph node level involvement. Estimating the risk of occult metastases for a newly diagnosed patient corresponds to calculating marginals of the joint probability distribution of LNL involvement conditioned on the patient's diagnose (Figure 2, bottom panel).