Glioma Grade Classification Using Machine Learning and MRI Radiomics: A Single-Center Prospective Study Comparing...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate glioma grading is crucial for patient prognosis and treatment decisions.
• Machine learning (ML) models can aid in glioma classification using MRI data.
• Evaluating advanced feature extraction techniques can enhance ML model performance.

Purpose of the Study:

• To assess the impact of wavelet-based radiomics features on ML model performance for glioma grade classification.
• To compare classification performance across different MRI sequences and ML models.
• To identify the optimal combination of radiomics features and ML models for accurate glioma grading.

Main Methods:

• Prospective study of 92 glioma patients with histopathological confirmation.
• MRI acquisition using a 1.5T scanner, including DWI, T1-weighted, T2-FLAIR, and T1Gd sequences.
• Radiomics feature extraction from original and wavelet-transformed images, followed by LASSO selection and Random Forest classification.

Main Results:

• Wavelet-based radiomics features significantly improved classification performance for all ML models and MRI sequences.
• The Random Forest model using original and wavelet-transformed eADC map features achieved the highest performance (AUC 0.96, accuracy 0.97).
• The model demonstrated high sensitivity (0.94) and specificity (0.91) in glioma grade classification.

Conclusions:

• Integrating wavelet-transformed radiomics features enhances the accuracy of ML-based glioma grade classification.
• The Random Forest model applied to eADC maps shows strong potential for clinical application.
• This approach offers an accurate, non-invasive, and efficient method for glioma grading.