Predicting a multi-parametric probability map of active tumor extent using random forests.

Area of Science:

• Neuroimaging
• Machine Learning in Medicine
• Oncology

Background:

• Glioblastoma multiforme (GBM) exhibits high infiltrative growth, complicating accurate tumor margin delineation.
• Traditional anatomic MRI sequences have limitations in defining the full spatial extent of GBM.
• Manual segmentation and classification of multi-parametric MRI data are time-consuming and error-prone.

Purpose of the Study:

• To develop and evaluate a machine learning-based approach for automated tumor segmentation in GBM.
• To integrate multi-parametric MRI data for improved spatial extent determination of brain tumors.
• To provide a preliminary step towards incorporating advanced imaging techniques into clinical practice for primary brain tumors.

Main Methods:

• A machine learning classifier, specifically a random forests classifier, was trained using radiologist-generated labels.
• The classifier performed voxel-wise tissue classification to automatically generate tumor segmentations.
• A leave-one-out cross-validation strategy was employed for model evaluation.
• A simple linear classifier was trained for comparative analysis.

Main Results:

• The random forests classifier demonstrated high accuracy in predicting radiologist-generated tumor segmentations.
• The machine learning approach successfully delineated the spatial extent of glioblastoma multiforme.
• The automated segmentation aligned well with expert manual delineations.

Conclusions:

• Machine learning-based multi-parametric MRI analysis offers a robust and accurate method for glioblastoma segmentation.
• This automated approach can significantly improve the efficiency and reliability of tumor margin assessment.
• The proposed method shows promise for integration into clinical workflows for primary brain tumor management.