Individualised prediction of HER2 status in colorectal cancer: development and validation of a radiomics prediction model using 18F-fluorodeoxyglucose positron emission tomography/magnetic resonance (18F-FDG PET/MR).

AIM: Accurate, noninvasive preoperative identification of HER2 status in colorectal cancer (CRC) can support patient stratification and treatment planning.

MATERIALS AND

METHODS

This study validated the value of radiomics derived from preoperative 18F-fluorodeoxyglucose positron emission tomography/magnetic resonance (18F-FDG PET/MR) as an imaging surrogate for HER2 assessment in CRC. We retrospectively analysed 98 patients who underwent HER2 testing and PET/MR (27 HER2-positive, 71 HER2-negative). Radiomic features were extracted from PET, T1-weighted (T1WI), and T2-weighted (T2WI) images, and dimensionality reduction identified informative features. Twelve models were constructed by combining three classifiers (logistic regression (LR), random forest (RF), and support vector machine (SVM)) with four feature sets (PET-only, T1WI-only, T2WI-only, and combined) under stratified 5-fold cross-validation.

RESULTS

The LR model with combined features achieved the best performance (area under the curve (AUC) = 0.921; sensitivity (SEN) = 0.813; specificity (SPE) = 0.942; accuracy (ACC) = 0.907; F1-score 0.825). Model interpretability using SHapley Additive exPlanations (SHAP) highlighted biologically plausible image traits contributing to classification.

CONCLUSION

These findings indicate that PET/MR-based radiomics offers a noninvasive and interpretable approach for the preoperative evaluation of HER2 status in CRC patients.