Machine learning model using routine clinical and MRI data predicts neoadjuvant response in HER2-positive breast cancer

A support vector machine model using routine clinical and MRI data achieved 82.1% accuracy and 83.3% AUC for predicting pathological complete response before neoadjuvant chemotherapy in HER2-positive breast cancer.

• Among 70 classifiers tested with nested leave-one-out cross-validation, an SVM with RBF kernel performed best: sensitivity 87.7%, specificity 76.4%, F1-score 83.3%, Cohen's kappa 64.2%.
• The most influential predictors were hormone receptor status (especially PR and ER expression), Ki67, and baseline MRI features.
• The study included 112 patients (57 with pCR, 55 without) and the authors note prospective multicentre external validation and calibration assessment are needed before clinical use.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.