Classification of the glioma grading using radiomics analysis

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Glioma grading is crucial for patient prognosis and survival.
• Accurate grading aids in treatment planning and management.
• Current grading methods can be subjective and time-consuming.

Purpose of the Study:

• To apply a radiomics approach using machine learning classifiers for glioma grading.
• To evaluate the efficacy of different machine learning models in classifying glioma grades.
• To identify radiomics features predictive of glioma grade.

Main Methods:

• Utilized 285 multi-modal MRI scans (T1, T1-contrast, T2, FLAIR) from the Brain Tumor Segmentation 2017 Challenge.
• Extracted 468 radiomics features from segmented tumor regions (enhancing, non-enhancing, necrosis, edema).
• Employed Minimum Redundancy Maximum Relevance for feature selection and logistic regression, support vector machines, and random forest for classification with five-fold cross-validation.

Main Results:

• Five significant radiomics features were selected for classification.
• The machine learning models achieved an average Area Under the Curve (AUC) of 0.9400 in training cohorts.
• The models demonstrated an average AUC of 0.9030 in test cohorts, with Random Forest achieving the highest AUC (0.9213).

Conclusions:

• Radiomics combined with machine learning and feature selection accurately determines glioma grade.
• This approach shows potential for developing high-throughput computer-aided diagnosis systems for gliomas.
• The findings support the integration of radiomics in clinical neuro-oncology for improved diagnostic accuracy.