Back to 2026 Abstracts
A Two-step Ct-based Radiomics Model To Predict Benefit From Neoadjuvant Immunotherapy In Patients With Esophageal Squamous Cell Carcinoma
Yingyi Li1, Wei Guo1, Liqiang Shi2, Lan Zhu1, Yuqin Cao1, Chengqiang Li1, Hecheng Li1.
1Shanghai Jiao Tong University School of Medicine Affiliated Ruijin Hospital, Shanghai, China, 2Chongqing University Cancer Hospital, Chongqing, China.
BACKGROUND:Achieving a pathological complete response (pCR) correlates with improved survival. This study aimed to develop an integrated clinical-radiomics approach to identify locally advanced esophageal squamous cell carcinoma (ESCC) patients likely to achieve pCR specifically from the addition ofneoadjuvant immunotherapy.
METHODS:A two-step modeling strategy was employed on a retrospective cohort of 253 ESCC patients. We utilized 155 patients treated with neoadjuvant chemoradiotherapy (nCRT) as the training set and 98 patients receiving neoadjuvant immunochemoradiotherapy (nICRT) as the validation set. From pre treatment thin-slice contrast-enhanced CTs, 851 radiomics features were extracted. First, a logistic regression model (Model 1) predicting pCR to nCRT was developed using features selected via LASSO and stepwise regression. Model 1 was then applied to the nICRT cohort to stratify patients. We specifically identified "immunotherapy beneficiaries" patients predicted as non-pCR by Model 1 (based on nCRT response patterns) but who actually achieved pCR. Finally, to discriminate these beneficiaries from true non-pCR patients, Model 2 was constructed using Ridge Logistic Regression following univariable and correlation analysis. Clinical characteristics were also assessed for integration into this predictive framework.
RESULTS:Using radiomics features alone, Model 1 demonstrated robust performance in the nCRT cohort(AUC=0.790; calibration MAE=0.045). In the nICRT cohort, 30 patients were identified as immunotherapy beneficiaries versus 55 true non-pCRs. The radiomics-based Model 2 effectively discriminated these groups, achieving an AUC of 0.733. Clinical variables are currently being integrated to further refine these predictive models.
CONCLUSIONS:The proposed two-step strategy effectively isolates ESCC patients who derive specific therapeutic benefit from neoadjuvant immunotherapy. By identifying potential pCR candidates who would otherwise fail standard nCRT, this radiomics-based model holds significant promise for guiding personalized treatment and organ preservation strategies.
LEGEND: Fig1 (A) Study Flowchart; (B) ROC Curve and Calibration Curve for Model 1&2.
Back to 2026 Abstracts
