Multimodal Deep Learning for Predicting Treatment Response to Neoadjuvant Chemoimmunotherapy in Esophageal Cancer

This observational study aims to investigate a clinical cohort of patients with locally advanced esophageal cancer undergoing neoadjuvant chemoimmunotherapy. By integrating multimodal clinical data-including demographic characteristics, medical history, imaging studies, pathological findings, and laboratory tests-and employing deep learning algorithms, the study seeks to develop predictive models for the early and accurate assessment of treatment response prior to surgery. Specifically, this study focuses on addressing the following key scientific questions: 1. Can multimodal clinical data be used to construct an accurate model for predicting pathological complete response (pCR) following neoadjuvant therapy? 2. Can deep learning models enable early identification of patients with suboptimal response to neoadjuvant therapy, defined as stable disease (SD) or progressive disease (PD), before surgery?

Sex

All

Inclusion Criteria:

1. Patients with histologically confirmed esophageal cancer based on biopsy results;
2. Patients recommended for neoadjuvant chemoimmunotherapy following multidisciplinary team (MDT) discussion or evaluation by thoracic surgery specialists;
3. Patients who received neoadjuvant chemoimmunotherapy;
4. Patients with complete imaging data before and after neoadjuvant treatment.

Exclusion Criteria:

1. Patients deemed eligible for surgery by the thoracic surgery team but who refused surgical treatment;
2. Patients with missing or poor-quality CT images;
3. Patients with concurrent malignancies other than esophageal cancer;
4. Patients with incomplete clinical data.