Differentiation between glioblastoma, brain metastasis and subtypes using radiomics analysis

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Differentiating glioblastoma from brain metastasis is crucial for treatment planning.
• Magnetic resonance imaging (MRI) is essential for brain tumor assessment but can present diagnostic challenges.
• Similar appearances of glioblastoma and solitary brain metastasis on MRI necessitate advanced analytical techniques.

Purpose of the Study:

• To employ radiomics analysis on conventional post-contrast T1-weighted (T1W) MRI to distinguish between glioblastoma and brain metastasis subtypes.
• To assess the efficacy of machine learning algorithms in classifying these brain tumor types.

Main Methods:

• Retrospective analysis of 439 patients (212 glioblastoma, 227 brain metastasis).
• Radiomics features extracted from post-contrast 3D T1W MRI (1.5 and 3.0 T).
• Support-vector machine (SVM) and other machine learning algorithms used for classification after dimension reduction.

Main Results:

• The SVM classifier achieved high accuracy (0.85), sensitivity (0.86), specificity (0.85), and AUC (0.96) for differentiating glioblastoma from brain metastasis in the testing dataset.
• SVM demonstrated strong performance in subtype classification, with varying accuracy, sensitivity, and specificity across glioblastoma, breast, lung, and other brain metastases.

Conclusions:

• Post-contrast T1W MRI combined with radiomics analysis shows high success in differentiating glioblastoma from brain metastasis.
• Additional MRI sequences with different tissue contrasts may be necessary for precise classification of brain metastasis subtypes.