A mRMRMSRC feature selection method for radiomics approach.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Radiomics extracts quantitative features from medical images, creating high-dimensional datasets.
• Effective feature selection is crucial for accurate prediction and classification in radiomics.
• Glioma subtype classification, like IDH1 status, benefits from advanced image analysis techniques.

Purpose of the Study:

• To propose a novel feature selection criterion for radiomics analysis of glioma.
• To enhance the performance of prediction and classification models using Magnetic Resonance Image (MRI) data.
• To evaluate the proposed method against existing feature selection techniques.

Main Methods:

• A new criterion, minimum Redundancy, Maximum Relevance and Maximum Sparse Representation Coefficient (mRMRMSRC), was developed.
• The criterion considers feature relevance, label relevance influenced by other features, and feature redundancy.
• The method was applied to glioma Isocitrate Dehydrogenase 1 (IDH1) estimation using MRI data.

Main Results:

• The mRMRMSRC criterion demonstrated superior performance compared to traditional methods like SRC, mRMR, F_score, and ReliefF.
• The proposed method achieved an Area Under the ROC Curve (AUC) of 90% for IDH1 estimation.
• This represents a significant improvement over the 77%-89% AUC achieved by state-of-the-art methods.

Conclusions:

• The mRMRMSRC criterion is an effective feature selection method for radiomics analysis of glioma.
• This approach offers improved accuracy in predicting glioma characteristics, such as IDH1 status.
• The findings suggest potential for enhanced diagnostic and prognostic capabilities in neuro-oncology through advanced radiomics.