CT-based radiomics nomogram to predict response of advanced adenocarcinoma of esophagogastric junction to neoadjuvant chemotherapy.

objectiveTo establish and validate a CT-based radiomics model to predict the response of adenocarcinoma of the esophagogastric junction (AEG) to neoadjuvant chemotherapy (NAC).

methods259 consecutive AEG patients, receiving 3 cycles of NAC with docetaxel, oxaliplatin and S-1, were retrospectively retrieved from two centers. Patients from center 1 were randomly divided into training (n = 139) and internal validation (n = 60) cohorts. Patients from center 2 were assigned to the external validation cohort (n = 60). In the training cohort, tumour-region-based radiomics features were selected, and a radiomics model was established to differentiate between patients with disease control and those with disease progression. Clinical factors were selected to develop a clinical model, and were incorporated with radiomics features to develop a radiomics-clinical model. Models' predictive performance and calibration ability were assessed with the area under the ROC curve (AUC) and calibration curve analysis, respectively. Decision curve analysis was used to evaluate the net clinical benefit of the models.

resultsThe radiomics model was developed with 9 core radiomics features, the clinical model was established by incorporating gross tumor volume, cT stage and Siewert type, and the clinical-radiomics model was established and plotted the nomogram. The clinical-radiomics model obtained better performance than the clinical or radiomics model (AUCs: 0.903 vs. 0.824 or 0.823, 0.899 vs. 0.813 or 0.800, and 0.895 vs. 0.804 or 0.719) in training, internal validation and external validation sets, respectively. The clinical-radiomics model showed the best calibration ability and the highest net benefit.

conclusionThe clinical-radiomics model can well predict the response of AEG to NAC. CRITICAL RELEVANCE STATEMENT: We provided a radiomics-clinical model to well predict the response of adenocarcinoma of the esophagogastric junction to neoadjuvant chemotherapy, which can help select appropriate patients to undergo chemotherapy, avoiding inappropriate patients from enduring toxic side-effects due to chemotherapy and delaying other treatments. KEY POINTS: A radiomics model for adenocarcinoma of the esophagogastric junction can predict the response to neoadjuvant chemotherapy. A combined model integrating clinical and radiomics features can improve predictive performance. The combined model is helpful for clinicians to develop individualized treatment regimens.