Radiomics Features for Radiation Pneumonitis Prediction in Thoracic Radiotherapy

610 lung cancer patient records from five multicenter cohorts were used to develop a federated learning model for predicting radiation pneumonitis. The dataset includes radiomics features extracted from pretreatment CT scans and was published by Zirui Zhu in April 2026. It specifically stratifies patients by immunotherapy exposure to adapt prediction models.

Use Cases

• Predict radiation pneumonitis risk from radiomics features extracted from whole-lung CT scans using federated learning models.
• Compare model performance and feature importance (via SHAP analysis) between immunotherapy-treated and non-immunotherapy patient cohorts.
• Fine-tune a baseline model trained on non-immunotherapy patient data to adapt it for immunotherapy-treated patients using sequential transfer learning.
• Validate prediction model robustness through independent external validation across multiple clinical centers.

Strengths

• 610 patient records provide a multicenter sample for model development and validation.
• Model performance is quantitatively reported with AUC metrics (e.g., 0.77 for baseline, 0.75 for adapted model).
• Explicit comparison of immunotherapy and non-immunotherapy cohorts addresses a contemporary clinical gap.

Limitations

• Dataset scope is limited to 404.8 KB, indicating it contains summary or feature data, not the raw medical images.
• Retrospective data collection may introduce selection bias inherent to observational studies.
• The specific radiomics features and patient-level data are not detailed in the provided description.

Provenance

Source

figshare, authored by Zirui Zhu.

Collection Method

Retrospective collection of multicenter cohorts; radiomics features extracted from pretreatment planning CT scans.

Time Range

null

Freshness

Data reflects a study published in April 2026, indicating recent analysis.

Geography

Data sourced from five multicenter cohorts; specific locations not provided.