Abstract
INTRODUCTION: We used habitat radiomics as an innovative tumor biomarker to predict the outcome of neoadjuvant therapy for esophageal cancer. METHODS: This was a two-center retrospective clinical study in which pretreatment CT scans of 112 patients with esophageal cancer treated with neoadjuvant chemoimmunotherapy and surgery between November 2020 and July 2023 were retrospectively collected from two institutions. For training (n = 85) and external testing (n = 27), patients from both institutions were allocated. We employed unsupervised methods to delineate distinct heterogeneous regions within the tumor area. RESULTS: To represent the prediction effect of different models, we plotted the AUC curves. The AUCs of the habitat models were 0.909 (0.8418-0.9758, 95% CI) and 0.829 (0.6423-1.0000, 95% CI) in the training and external test cohorts, respectively. The AUCs of the nomogram models were 0.914 (0.8483-0.9801, 95% CI) and 0.849 (0.6752-1.0000, 95% CI) in the training and external test cohorts, respectively. DISCUSSION: The results revealed that the model based on habitat data outperforms traditional radiomic analysis models. In addition, when the model is combined with clinical features, it improves the predictive accuracy of pathological complete response in patients undergoing neoadjuvant chemoimmunotherapy.