Modeling bilateral lymphatic head and neck tumor progression for personalized elective target volume definition

Abstract Objective. Large irradiated volumes are a major contributor to severe side-effects in patients with head and neck cancer undergoing curatively intended radiotherapy. We propose a data-driven approach for defining the elective clinical target volume (CTV-E) on a patient-specific basis, with the potential to reduce irradiated volumes compared to standard guidelines. Approach. We introduce a bilateral Bayesian network (BN), trained on a large cohort, to estimate the patient-specific risk of undetected nodal involvement for both ipsilateral and contralateral lymph node levels (LNLs) I, II, III, and IV, using clinical features, such as patterns of nodal involvement, T-stage, tumor location. By applying risk thresholds, we generated individualized, risk-dependent CTV-E’s for representative patient scenarios and compared the resulting treatment volumes and residual risk to those recommended by standard clinical guidelines. Main results. We computed the risks for a set of representative patient scenarios including (1) N0 (T1 and T2 tumor stage), (2) N+ in ipsilateral LNL II (T1 and T2 tumor stage), (3) N+ in ipsilateral LNL II and III (T1 and T2 tumor stage), and (4) N+ of both ipsilateral and contralateral LNL II (T3 and T4 tumor stage). Depending on the chosen risk threshold, the bilateral BN allowed for reductions in irradiated volumes relative to standard clinical protocols. For every patient scenario considered, the CTV-E’s defined by the applied risk thresholds were associated with a low estimated probability of undetected nodal involvement in any excluded LNL. Significance. We present a data-driven framework for personalized CTV-E definition, encouraging the discussion of more patient-specific elective nodal target volumes, with potential for de-escalation of irradiated elective volumes.