Glioma grading using structural magnetic resonance imaging and molecular data.

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Glioma grading is crucial for treatment and prognosis.
• Current grading relies on invasive methods or less integrated data.
• Need for noninvasive, accurate glioma grading methods.

Purpose of the Study:

• To develop and evaluate a noninvasive glioma grading method using structural MRI and molecular data.
• To compare the performance of MRI-based grading with digital pathology (DP).
• To assess grading performance incorporating molecular alterations (IDH1/2 mutations).

Main Methods:

• Utilized conventional structural MRI with radiomic texture features (dynamic texture analysis, multifractal detrended fluctuation analysis, multiresolution fractal Brownian motion).
• Evaluated on two multicenter MRI datasets: BRATS-2017 (high-grade vs. low-grade) and TCIA (glioblastoma vs. low-grade glioma).
• Compared MRI grading with DP images from TCGA, incorporating molecular data for World Health Organization (WHO) grading criteria.

Main Results:

• Achieved a mean area under the receiver operating characteristic curve (AUC) of 0.88 on the BRATS dataset.
• MRI and DP image classification in TCIA/TCGA yielded mean AUCs of 0.90 and 0.93, respectively.
• The integrated approach demonstrated strong performance in glioma grading.

Conclusions:

• The proposed noninvasive method using structural MRI is effective for glioma grading.
• Combining MRI, DP, and molecular data enhances grading accuracy according to WHO criteria.
• This approach offers a promising alternative for noninvasive tumor assessment.