A clinical decision-support framework to differentiate radiation necrosis from tumor progression in brain metastases

Differentiating radiation necrosis (RN) from tumor progression (TP) after stereotactic radiotherapy (SRT) in brain metastases (BMs) is a clinically consequential problem, as conventional MRI frequently fails to distinguish between them. Misclassification can lead to inappropriate treatment decisions or delayed therapy. The objective of this study was to develop a clinically interpretable, data-driven model that integrates lesion growth dynamics with routinely available clinical variables to improve discrimination between RN and TP.

We retrospectively analyzed 175 BMs from 6 institutions. Lesion volumes were extracted from three consecutive contrast-enhanced T1-weighted MRI, and growth dynamics were quantified by estimating the growth exponent β. Clinical and treatment-related variables were systematically evaluated, and a multivariable predictive model was trained on a development cohort (n = 131) and validated on an external cohort (n = 44).

The final model combined β, primary tumor histology, and SRT modality. In the development cohort, the model demonstrated strong discriminative performance (AUC = 0.887). External validation confirmed generalizability, achieving an overall accuracy of 0.75, with high specificity (0.85) and positive predictive value (0.92) for RN. Incorrect classifications were largely confined to an intermediate-probability zone, while predictions at low and high probability extremes were highly reliable. The model was translated into a freely accessible, web-based tool to facilitate clinical decision-making.

By integrating lesion growth dynamics with routine clinical variables, this probability-based framework supports clinically meaningful differentiation between RN and TP. Its ability to explicitly represent diagnostic uncertainty, together with external validation, highlights its potential utility as a decision-support tool in the management of BMs.