PACE: A Probabilistic Atlas for Normal Tissue Complication Estimation in Radiation Oncology

Area of Science:

• Radiation Oncology
• Medical Physics
• Radiotherapy

Background:

• Traditional Normal Tissue Complication Probability (NTCP) models often lack voxel-based radio-sensitivity data.
• There is a recognized need for advanced NTCP models in radiation oncology to better predict patient outcomes.
• Existing models like Lyman-Kutcher-Burman (LKB) may not fully capture complex radiobiological effects.

Purpose of the Study:

• To introduce and validate a novel formalism, the Probabilistic Atlas for Complication Estimation (PACE), for predicting radiation-induced morbidity (RIM).
• To integrate voxel-based radio-sensitivity and non-dosimetric covariates into NTCP modeling.
• To compare the performance of PACE against classical LKB models.

Main Methods:

• Developed the PACE formalism, adapting classical NTCP model structures but using RIM odds and covariates as inputs instead of dose distributions.
• Validated PACE using *in silico* synthetic data and a clinical dataset of thoracic cancer patients with lung fibrosis.
• Trained and compared LKB models with PACE on the same datasets.

Main Results:

• PACE demonstrated superior predictive accuracy (>0.8) for synthetic outcomes compared to LKB models.
• PACE exhibited significantly higher discrimination and calibration performance on clinical data than LKB models.
• Cross-validation confirmed the enhanced performance of PACE.
• PACE successfully inferred spatial patterns of organ radio-sensitivity, demonstrating its potential as a learning tool.

Conclusions:

• The PACE model represents a significant advancement in NTCP philosophy, incorporating detailed radio-sensitivity information.
• PACE offers improved prediction of radiation-induced morbidity, outperforming traditional LKB models.
• The ability of PACE to map radio-sensitivity opens new avenues for personalized radiotherapy and understanding treatment effects.