What question did this study set out to answer?

June 10, 2026Open Access

Radiomic and dosiomic machine learning models for predicting radiation-induced oral mucositis in head and neck cancer: a systematic review

Key Points

This review aims to evaluate machine learning models for predicting radiation-induced oral mucositis in head and neck cancer patients, focusing on the role of imaging biomarkers.
Systematic review following PRISMA 2020 guidelines registered in PROSPERO (CRD420251075245).
Inclusion of studies with machine learning models for oral mucositis prediction using radiomic and dosiomic data.
Risk of bias assessed using the PROBAST tool, with comparisons of models based on predictor types.
Seven studies involving 2,311 patients were included, all using dosimetric predictors, with some incorporating radiomic (n=3) and dosiomic (n=2) features.
Reported discrimination (AUC) ranged from 0.70 to 0.96, with external validations showing reduced performance (AUC ~ 0.65).
High risk of bias noted due to limited sample sizes and calibration issues, indicating models are exploratory.

Abstract

OBJECTIVE: To critically evaluate machine learning (ML) models developed for predicting radiation-induced oral mucositis (OM) in head and neck cancer (HNC) patients, emphasizing the contribution of advanced imaging biomarkers such as radiomic and dosiomic features. MATERIALS AND METHODS: This systematic review followed the PRISMA 2020 guidelines and was registered in PROSPERO (CRD420251075245). The PICOS framework guided the review question and inclusion criteria. Studies developing or validating ML models for OM prediction using radiomic and/or dosiomic data were included. Electronic databases (PubMed, EMBASE, Web of Science, Scopus) were searched up to July 2025. The risk of bias was assessed using the PROBAST tool, and models were compared based on predictor types. RESULTS: Seven studies involving 2,311 patients were included. All models incorporated dosimetric predictors, while some used additional radiomic (n = 3), dosiomic (n = 2), contouromic (n = 1), or oral health indices (n = 1). Reported discrimination (AUC) ranged from 0.70 to 0.96, but external validations showed reduced performance (AUC ~ 0.65). High risk of bias was mainly observed in analysis domains due to limited sample sizes and incomplete calibration. CONCLUSION: ML models for OM prediction remain exploratory and are not yet ready for routine clinical implementation due to poor generalizability and calibration issues.

Bookmark

View Full Paper

Cite This Study

Abreu et al. (Mon,) studied this question.

synapsesocial.com/papers/6a2901536f82f25be989d9f4 https://doi.org/https://doi.org/10.1007/s12094-026-04427-0

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

Bookmark

View Full Paper