Image Based Data Mining Using Per-voxel Cox Regression.

Area of Science:

• Radiation oncology
• Medical physics
• Data science

Background:

• Image Based Data Mining (IBDM) analyzes radiotherapy data, finding links between dose and outcomes.
• Previous methods often analyzed dose and clinical variables separately.
• Confounding variables were typically addressed post-hoc, limiting direct spatial correlation.

Purpose of the Study:

• To introduce a novel method for integrating confounding variables directly into voxel-wise dose-response analysis.
• To apply this method to a large lung cancer patient cohort.
• To generate 3D hazard maps for clinical variables in radiotherapy.

Main Methods:

• Developed a voxel-based Cox regression to analyze dose-response while accounting for clinical variables simultaneously.
• Applied the method to a large dataset of lung cancer patients.
• Generated 3D maps illustrating the hazard associated with dose in each voxel.

Main Results:

• Confirmed a region of interest at the base of the heart where patients with poor performance status (PS > 1) showed increased sensitivity to incidental radiation dose.
• Demonstrated that the influence of clinical variables, such as patient age, can alter the significance of dose in specific regions.
• Identified that patient age becomes a dominant factor over performance status in certain areas.

Conclusions:

• The novel voxel-wise analysis method effectively integrates clinical confounders into radiotherapy dose-response evaluation.
• Findings highlight the complex interplay between radiation dose, patient factors, and clinical outcomes in lung cancer.
• This approach can refine radiotherapy planning and inform future clinical trial design by generating testable hypotheses.