Feature selection methods and predictive models in CT lung cancer radiomics

Area of Science:

• Medical Imaging
• Quantitative Imaging Biomarkers
• Radiomics

Background:

• Radiomics extracts quantitative features from medical images.
• These features reveal tissue characteristics and radiologic phenotypes not visible to clinicians.
• A standard radiomics workflow involves cohort selection, feature extraction, model selection, and validation.

Purpose of the Study:

• To review radiomics investigations in CT lung cancer.
• To provide an overview of commonly used radiomic feature selection and predictive modeling methods.
• To compare limitations and sources of uncertainty in clinical applications of these methods.

Main Methods:

• Literature review of published radiomics studies in CT lung cancer.
• Analysis of radiomic feature extraction techniques.
• Evaluation of feature selection and predictive model methodologies.

Main Results:

• Increasing attention on radiomic feature extraction, standardization, and reproducibility.
• Lack of rigorous evaluation for feature selection methods and predictive models.
• Identification of commonly used methods and their limitations in clinical settings.

Conclusions:

• Awareness of the impact of feature and model selection on radiomics study integrity is crucial.
• Further research is needed to rigorously evaluate selection methods.
• Standardization and validation are key for reliable clinical application of radiomics.