Author response to Cunha et al.

Area of Science:

• Oncology
• Radiology
• Biostatistics

Background:

• The clinical utility of immunotherapy in rare cancers is hindered by a lack of predictive biomarkers.
• Radiomics offers a non-invasive method to extract quantitative imaging features for predictive modeling.
• This work addresses the need for robust radiomics models to predict immunotherapy response in rare cancers.

Discussion:

• Justification for employing Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection and eXtreme Gradient Boosting (XGBoost) for classification in radiomics models.
• Detailed explanation of the rigorous methodology, including Leave-One-Out Cross-Validation (LOOCV) and 10-fold cross-validation, to ensure model generalizability and prevent overfitting.
• Assessment of multicollinearity among selected radiomic features to maintain model interpretability and reliability.

Key Insights:

• Radiomics analysis, utilizing LASSO and XGBoost, demonstrates potential in predicting pembrolizumab response in advanced rare cancers.
• Careful model validation techniques, including cross-validation and multicollinearity checks, are essential for reliable predictive biomarkers.
• The study provides a methodological framework for developing and validating radiomics signatures for immunotherapy response in rare malignancies.

Outlook:

• Further validation in larger, independent cohorts is necessary to translate these findings into clinical practice for rare cancer patients.
• Integration of radiomics with other data modalities may enhance predictive accuracy for immunotherapy response.
• Development of standardized radiomics protocols is crucial for consistent application across different institutions and imaging hardware.