Multi-modal deep-learning model for real-time prediction of recurrence in early-stage esophageal cancer: A multi-modal approach

BackgroundThe surveillance protocol for early-stage esophageal cancer is not contingent upon individualized risk factors for recurrence. This study aimed to develop a deep-learning model using comprehensive data from clinical practice for practical longitudinal monitoring.MethodsA multi-modal deep-learning model employing transformers was developed for real-time recurrence prediction using clinical, pathological, and molecular data at baseline, alongside longitudinal laboratory, and radiologic data during surveillance. Patients with histologically confirmed esophageal cancer (stage I-III) undergoing curative intent surgery between January 2008 and September 2022 were included. The collected data was divided into training and validation in a 5:5 ratio. The primary outcome focused on predicting likelihood of recurrence within one year from the monitoring point. This study demonstrated the timely provision of risk scores (RADAR score), determined thresholds, and the corresponding Area Under the Curve (AUC).ResultsA total of 1,486 patients were enrolled (655 with stage I, 457 with stage II, and 374 with stage III). The training and validation sets comprised 743 and 743 patients, respectively. The model incorporated 9 clinical-pathological-molecular factors at baseline, alongside longitudinal laboratory, and radiologic text data. Radiologic data of 11,933 was used (mean 8.0 chest CT scan per patient) during surveillance. Baseline RADAR score wsa 0.257 (standard deviation [Std] 0.132) in stage I, 0.468 (Std 0.116) in stage II, and 0.684 (Std 0.107) in stage III. The AUC for predicting relapse within one year from that monitoring point was 0.795 across all stages with the sensitivity of 77.2% and specificity of 70.1% (AUC=0.774 in stage I, AUC=0.697 in stage II, and AUC=0.683 in stage III).ConclusionsThis pilot study introduces a deep-learning model utilizing multi-modal data from routine clinical practice for predicting relapse in early-stage esophageal cancer. It demonstrates the timely provision of risk score, RADAR score, to clinicians for recurrence prediction, potentially guiding risk- adapted surveillance strategies and aggressive adjuvant systemic treatment.Legal entity responsible for the studyThe authors.FundingHas not received any funding.DisclosureAll authors have declared no conflicts of interest.