Machine Learning-Based Multiomics Prediction Model for Radiation Pneumonitis

Area of Science:

• Oncology
• Medical Physics
• Radiology

Background:

• Radiation pneumonitis (RP) is a common side effect of thoracic radiotherapy.
• Accurate prediction of RP is crucial for optimizing treatment plans and patient outcomes.
• Current prediction models often lack sufficient predictive power.

Purpose of the Study:

• To establish and validate an effective CT-based prediction model for radiation pneumonitis (RP).
• To leverage a multiomics approach combining radiomics and equivalent dose of 2 Gy fractionated radiation (EQD2)-based dosiomics.
• To compare the performance of the novel multiomics model against traditional methods.

Main Methods:

• Retrospective analysis of 91 non-small cell lung cancer patients treated with radiotherapy.
• Extraction of radiomic features from lung-Clinical Target Volume (lung-CTV) and dosiomic features from physical and EQD2-based dose distributions.
• Development and validation of four machine learning models (DVH, radio+DVH, radio+dose, radio+eqdose) using eleven classifiers and fivefold cross-validation.

Main Results:

• The radiomics combined with EQD2-based dosiomics (radio+eqdose) model showed significantly higher training AUC, accuracy, and F1-score compared to DVH, radio+DVH, and radio+dose models (p < 0.05).
• The radio+eqdose model also demonstrated higher average precision and recall, though not statistically significant (p > 0.05).

Conclusions:

• Machine learning-based multiomics incorporating radiomics and EQD2-based dosiomics offers a more efficient and effective method for predicting radiation pneumonitis.
• This advanced approach holds promise for improving personalized radiotherapy in lung cancer patients.
• Further validation in larger cohorts is warranted to solidify clinical applicability.