A Patient-Specific Digital Twin for Adaptive Radiotherapy of Non-Small Cell Lung Cancer

Yale University researchers have developed COMPASS (Comprehensive Personalized Assessment System), an AI-driven digital twin architecture that could revolutionize cancer radiotherapy. The system creates patient-specific digital twins that continuously update throughout treatment, analyzing data from each radiation fraction to predict toxicity before symptoms appear.

Technically, COMPASS employs a GRU autoencoder to learn organ-specific latent trajectories from sequential data streams including PET scans, CT images, dosiomics, radiomics, and cumulative biologically equivalent dose kinetics. In their study of 8 non-small cell lung cancer patients undergoing biologically guided radiotherapy, the system analyzed 99 organ fraction observations covering 24 organ trajectories (spinal cord, heart, and esophagus). Despite the small cohort, the intensive temporal phenotyping revealed that increasing risk ratings occurred several fractions before clinical toxicity emerged.

The system's dense BED-driven representation uncovered biologically relevant spatial dose texture characteristics that traditional volume-based dosimetry averages out. This represents a significant advancement beyond current static, population-based NTCP models that overlook dynamic biological trajectories. COMPASS successfully predicted CTCAE grade 1 or higher toxicity using logistic regression classification of the learned latent trajectories.

This research establishes a proof-of-concept for AI-enabled adaptive radiotherapy where treatment decisions are guided by continually updated digital twins tracking each patient's evolving biological response. The early warning window demonstrated by COMPASS could enable clinicians to modify treatment plans proactively, potentially reducing side effects and improving outcomes through truly personalized cancer care.