Predicting tumor recurrence site after reirradiation in head and neck cancer: a retrospective external validation of a published [18F]-FDG PET radiomic signature.

Abstract

Purpose: This study evaluates the efficacy of a previously published [18F]-FDG PET radiomic signature in predicting locoregional failure locations post-reirradiation in head and neck cancer (HNC) patients, using an independent cohort from a different institution.

Materials and methods: Among the 66 patients reirradiated for recurrent HNC at Massachusetts General Hospital between 2012 and 2022, 31 underwent pre-reirradiation PET, constituting the external cohort for this analysis. These patients were characterized using the same radiomic features as the original model (Intensity_histogram_min, Kurtosis, Correlation, and Contrast), projected as a supplementary individual onto the published first principal component, and assigned to one of two groups using the published cutoff. The cutoff was then optimized for the external cohort to determine the loss of performance due to technical or population shifts.

Results: Among the 31 patients, 22 experienced a second locoregional failure, distributed between 12 "in-field" and 10 "outside" recurrences. With the original cutoff, the model achieved a BA of 70% and a positive predictive value (PPV) of 86% for detecting "in-field" recurrences. After recalibrating the cutoff, the model achieved a BA of 78% and a PPV of 89%, close to the 84.5% BA obtained in the original article.

Conclusion: The study validates the ability of the previously established PET radiomic signature to predict "in-field" relapses following reRT with a high PPV. These results support the potential of PET radiomics in identifying patients who may benefit from "in-field" dose escalation in reRT schemes. The model is freely available through the user-friendly LIFEx software.