Deep-Learning Model for Real-Time Prediction of Recurrence in Early-Stage Non-Small Cell Lung Cancer: A Multimodal...

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Informatics

Background:

• Current surveillance for early-stage non-small cell lung cancer (NSCLC) lacks personalization based on individual recurrence risk factors.
• Longitudinal monitoring protocols are essential for timely detection of recurrence in NSCLC survivors.

Purpose of the Study:

• To develop and validate a deep-learning model for predicting recurrence in early-stage NSCLC using comprehensive clinical data.
• To create a practical tool for longitudinal monitoring that incorporates individualized risk assessment.

Main Methods:

• A multimodal deep-learning model utilizing transformers was developed for real-time recurrence prediction.
• The model integrated baseline clinical, pathological, and molecular data with longitudinal laboratory and radiologic surveillance data.
• Data from 14,177 patients with stage I-III NSCLC treated with curative intent between 2008-2022 were analyzed.

Main Results:

• The deep-learning model incorporated 64 distinct factors and demonstrated strong predictive performance.
• The area under the curve (AUC) for predicting recurrence within one year was 0.854 across all stages.
• The model achieved a sensitivity of 86.0% and specificity of 71.3%, with stage-specific AUCs ranging from 0.724 to 0.872.

Conclusions:

• A deep-learning model using multimodal data from routine clinical practice can effectively predict relapse in early-stage NSCLC.
• The developed RADAR risk score provides timely, actionable insights for clinicians.
• This tool has the potential to guide risk-adapted surveillance and optimize adjuvant systemic treatment decisions for NSCLC patients.