Quantitative Physiologic MRI Combined with Feature Engineering for Developing Machine Learning-Based Prediction...

Area of Science:

• Neuroimaging
• Machine Learning
• Oncology

Background:

• Differentiating glioblastomas (GBMs) from single brain metastases (BMs) is crucial for effective treatment planning.
• Accurate early distinction enables timely therapeutic interventions.

Purpose of the Study:

• To leverage diffusion tensor imaging (DTI) and dynamic susceptibility contrast (DSC)-perfusion-weighted imaging (PWI) parameters with machine learning to distinguish GBMs from BMs.
• To evaluate the diagnostic performance of various machine learning classifiers.

Main Methods:

• Collected 3T MRI data (anatomical, DTI, DSC-PWI) from 62 GBM and 26 BM patients.
• Extracted quantitative imaging features (MD, anisotropy, rCBV) from contrast-enhancing and peritumor regions.
• Employed feature engineering and 10 machine learning classifiers, validated with cross-validation and ROC analysis.

Main Results:

• A random forest classifier with ANOVA F-value feature selection achieved the highest performance.
• Achieved an area under the ROC curve of 92.67%, 87.8% accuracy, 73.64% sensitivity, and 97.5% specificity.
• Combined interacting and non-interacting MRI features improved diagnostic capabilities.

Conclusions:

• Machine learning integrating physiological MRI parameters shows high accuracy in differentiating GBMs from BMs.
• This approach holds promise for improving diagnostic accuracy in neuro-oncology.
• The findings support the potential for enhanced treatment strategies based on precise tumor characterization.