Monte Carlo Gradient Boosted Trees for Cancer Staging: A Machine Learning Approach

Area of Science:

• Medical imaging analysis
• Machine learning applications in oncology

Background:

• High-dimensional data analysis requires feature selection and classification.
• Gradient Boosted Trees (GBT) and XGBoost offer robust, interpretable classification.
• Radiomics from CT scans provide quantitative imaging biomarkers for cancer analysis.

Purpose of the Study:

• To develop a Monte Carlo Gradient Boosted Trees (MCGBT) model for feature reduction and classification.
• To apply MCGBT to a lung cancer dataset for radiomic feature identification and staging.
• To evaluate the performance of MCGBT in achieving accurate and efficient cancer classification.

Main Methods:

• Implementation of a Monte Carlo Gradient Boosted Trees (MCGBT) model.
• Application of MCGBT to a dataset of 107 radiomic features from lung CT scans.
• Feature reduction to identify a subset of key radiomics for classification.

Main Results:

• A reduced set of 12 radiomic features was identified as significant.
• MCGBT achieved a cancer staging accuracy of 90.3% over 100 independent runs.
• Performance with reduced features was comparable to using the full 107 radiomic features.

Conclusions:

• MCGBT provides an effective method for feature reduction and classification in medical data.
• The identified subset of radiomics enables the development of lean and deployable lung cancer classifiers.
• This approach enhances efficiency and interpretability in radiomic-based cancer analysis.