Comparing supervised and semi-supervised machine learning approaches in NTCP modeling to predict complications in head and neck cancer patients

Background and purposeHead neck cancer (HNC) patients treated with radiotherapy often suffer from radiation-induced toxicities. Normal Tissue Complication Probability (NTCP) modeling can be used to determine the probability develop these toxicities based on patient, tumor, treatment dose characteristics. Since currently NTCP models are developed using supervised methods that discard unlabeled patient data, we assessed whether addition of data by semi-supervised would gain predictive performance.Materials methodsThe method self-training was compared regression without prior multiple imputation chained equation (MICE). The were for most common toxicity outcomes in HNC patients, xerostomia (dry mouth) dysphagia (difficulty swallowing), measured at six months after treatment, a development cohort 750 patients. externally validated validation 395 Model performance discrimination calibration.ResultsMICE did not improve terms or calibration external current models. In addition, relative different change upon decrease amount (labeled) available model development. Models ridge outperformed logistic outcome.ConclusionSince there no apparent MICE, still preferred