Machine learning-based radiomics analysis in predicting RAS mutational status using magnetic resonance imaging.

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Computational Pathology

Background:

• RAS mutations are critical in determining treatment strategies for liver metastases.
• Accurate pre-surgical prediction of RAS mutational status is essential for personalized patient management.
• Current methods for determining RAS status can be invasive and time-consuming.

Purpose of the Study:

• To evaluate the effectiveness of radiomics features derived from magnetic resonance imaging (MRI) with a hepatospecific contrast agent.
• To predict the RAS mutational status of liver metastases in a pre-surgical setting.
• To assess the performance of various machine learning models in this prediction task.

Main Methods:

• Retrospective analysis of patient MRI scans in a pre-surgical setting.
• Extraction of 851 radiomics features using 3D Slicer and PyRadiomics, adhering to IBSI standards.
• Application of SASYNO for sample balancing and statistical analysis including ROC, LRM, and machine learning classifiers (DT, KNN, SVM).

Main Results:

• Seven normalized radiomics features from the arterial phase, 11 from the portal phase, and 12 each from the hepatobiliary phase and T2-W SPACE sequence were identified as robust predictors.
• A linear regression model (LRM) combining 12 normalized features from the hepatobiliary phase achieved 99% accuracy, 97% sensitivity, and 100% specificity.
• No significant accuracy improvement was observed with tested classifiers (DT, KNN, SVM) compared to LRM.

Conclusions:

• A normalized approach to MRI radiomics analysis is effective for predicting RAS mutational status.
• This non-invasive method holds promise for improving pre-surgical assessment of liver metastases.
• Further validation is warranted to integrate this technique into clinical practice.