A Novel CT Habitat Radiomics Approach for HER2 Status Prediction in Gastric Cancer

Accurate preoperative evaluation of human epidermal growth factor receptor 2 (HER2) status is crucial for guiding personalized treatment in gastric cancer (GC). This study aimed to evaluate the clinical value of CT-based multi-subregion habitat radiomics for noninvasive preoperative prediction of HER2 expression.

This retrospective study enrolled 857 pathologically confirmed GC patients from two medical centers. Patients were stratified into HER2-positive and HER2-negative groups according to postoperative pathology. At Center 1, 657 patients were randomly divided into training (n = 460) and testing (n = 197) cohorts (7:3 ratio), while 200 patients from Center 2 served as an external validation (ex-vad) cohort. Radiomics features were extracted from contrast-enhanced venous-phase CT images, and a radiomics score (Radscore) was subsequently generated. Tumor regions of interest underwent unsupervised clustering to define habitat subregions, forming models based on two, three, and four subregions. The optimal habitat model was selected based on predictive accuracy. A combined model integrating the optimal habitat model and key clinical features was subsequently constructed.

Six models were developed: clinical, Radscore, Habitat2, Habitat3, Habitat4, and combined (Habitat3 plus clinical variables). In the training cohort, area under the receiver operating characteristic (ROC) curve values were 0.68 (95% confidence intervals (CI), 0.62-0.75), 0.78 (0.73-0.83), 0.75 (0.69-0.81), 0.90 (0.87-0.94), 0.78 (0.73-0.84), and 0.94 (0.91-0.97), respectively. The combined model achieved superior predictive performance in both training, test and ex-vad cohorts, followed by Habitat3.

The Habitat3 model robustly predicts HER2 expression in GC. The combined model further improves predictive accuracy.

This study developed CT-based radiomics habitat models for predicting HER2 expression in gastric cancer. Habitat3 achieved optimal performance, and its combination with clinical variables yielded superior efficacy and clinical applicability, providing a robust non-invasive tool for preoperative HER2 assessment in gastric cancer.