نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Comparison of MRI Radiomics-based Machine Learning and Deep Learning Models for Low-Grade and High-Grade Glioma Differential Diagnosis

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Maryam Bagherpour

Author:
Name 1:	Maryam Bagherpour - Corresponding - Presenter
Affiliation 1:	DEPARTMENT OF RADIOLOGY, FACULTY OF ALLIED MEDICAL SCIENCES, TABRIZ UNIVERSITY OF MEDICAL SCIENCES, TABRIZ, IRAN

Name 2:	Davood Khezerloo
Affiliation 2:	DEPARTMENT OF RADIOLOGY, FACULTY OF ALLIED MEDICAL SCIENCES, TABRIZ UNIVERSITY OF MEDICAL SCIENCES, TABRIZ, IRAN

Name 3:	Ali Ghaffari
Affiliation 3:	Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Gliomas constitute a heterogeneous group of primary brain tumors that differ considerably in their biological behavior and clinical outcomes. Accurate determination of tumor grade plays a critical role in guiding treatment decisions and predicting prognosis. Although histopathological examination remains the gold standard for tumor grading, it requires invasive tissue sampling. MRI provides a noninvasive approach, but conventional visual interpretation may not fully capture the quantitative information contained in imaging data. Radiomics addresses this limitation by extracting quantitative features from MRI, which can be used as inputs to machine learning and deep learning models for glioma grade classification. This study therefore aimed to compare machine learning and deep learning models based on MRI radiomics features for classifying low-grade and high-grade gliomas.

Methods:

This study used a radiomics dataset derived from MRI examinations of 50 patients with histologically confirmed gliomas, obtained from the National Cancer Institute (NCI) public repository The dataset included 25 grade 2 and 50 grade 3 tumors, labeled according to the WHO grading system. The data were divided into training and test sets using an 80:20 stratified ratio. Recursive Feature Elimination (RFE) with 8-fold grouped stratified cross-validation was applied for feature selection while reducing dimensionality and preventing data leakage between samples from the same group. Model hyperparameters were optimized with Optuna using Area Under the Receiver Operating Characteristic Curve (ROC-AUC) and F1-score as optimization criteria. Seven classifiers—Deep Fully Connected Network (DeepFCN), Decision Tree, K-Nearest Neighbors, Multilayer Perceptron, Random Forest, Support Vector Machine Classifier, and XGBoost—were trained and evaluated. Performance was assessed on the independent test set using accuracy, F1-score, and ROC-AUC.

Results:

DeepFCN achieved the highest ROC-AUC in both optimization settings. In the ROC-AUC–optimized setting, DeepFCN achieved an ROC-AUC of 0.947, while MLP showed the highest accuracy and F1-score, both reaching 0.85. SVC also showed competitive performance, with an ROC-AUC of 0.864 and an F1-score of 0.76. In the F1-optimized setting, DeepFCN achieved an ROC-AUC of 0.969, with an accuracy of 0.83 and an F1-score of 0.82. SVC and MLP followed with F1-scores of 0.79 and 0.77, respectively. Decision Tree and K-Nearest Neighbors showed lower classification performance across both optimization settings.

Conclusion:

MRI radiomics combined with machine learning and deep learning models showed potential for distinguishing low-grade from high-grade gliomas. DeepFCN achieved the highest ROC-AUC in both optimization settings, while MLP showed the highest accuracy and F1-score in the ROC-AUC–optimized setting. No single model performed best across all metrics, indicating that classification performance depended on the optimization strategy and evaluation criterion. Further validation is needed before routine clinical application.

Keywords:

Glioma grading, MRI radiomics, Machine learning, Deep learning, Classification