ESTRO 2022 - Abstract Book

S764

Abstract book

ESTRO 2022

Conclusion While none of univariate associations identified by van Dijk et al as being statistically significant could be replicated, the addition of maxHU improved SS12m prediction on patients with both SMG intact. These variations, and the limited generalisability of the findings, may be explained by the number of differences in the imaging characteristics of the two studies and subsequent methodological implementation. This highlights the importance of external validation as well as high quality reporting guidelines and standardisation protocols to ensure generalisability, replication and ultimately clinical implementation.

MO-0877 Predicting bilateral involvement in head & neck cancer using an interpretable probabilistic model

R. Ludwig 1 , B. Pouymayou 1 , J. Hoffmann 1 , P. Balermpas 1 , J. Unkelbach 1

1 University Hospital Zurich, Radiation Oncology, Zurich, Switzerland

Purpose or Objective In current clinical practice, the elective clinical target volume (CTV-N) for oropharyngeal head & neck squamous cell carcinoma (OHNSCC) is defined using prevalence-based guidelines [1], which recommend extensive bilateral neck irradiation for most patients. However, the actual patient-specific risk of harboring microscopic metastases in the contralateral neck remains insufficiently quantified. We propose a probabilistic and interpretable model for predicting the personalized risk of involvement in lymph node levels (LNLs) of both sides of the neck depending on T-stage, location of macroscopic metastases, and the primary tumor’s lateralization. This may identify additional patients in whom contralateral neck irradiation can be avoided or reduced, possibly reducing toxicity. Materials and Methods We extracted the lymphatic patterns of progression for 287 OHNSCC patients treated at our institution. The data can be visualized at www.lyprox.org. This was used to train the hidden Markov model (HMM) for predicting nodal involvement we developed and published recently [2]. For this work, the model in [2] was extended to contralateral spread while treating mid-sagittal plane extension of the primary tumor as risk factor. It models the LNLs as binary random variables connected