Related Experiment Video
Updated: May 10, 2026
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
Factors predicting MRI glioma segmentation accuracy in deep learning models: a systematic review and meta-analysis
Leonardo Di Cosmo1, Filippo Emanuele Colella1, Paweł Łajczak2
1Department of Biomedical Sciences, Humanitas University, Pieve Emanuele, Milan, Italy.
Summary
Deep learning models for glioma segmentation show high accuracy, especially with 3D and multiparametric MRI inputs. Factors like dataset and model architecture influence performance, but no single factor explains all variability in automated glioma segmentation.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Accurate glioma segmentation is crucial for diagnosis and treatment.
- Manual segmentation is time-consuming and variable.
- Deep learning (DL) models offer automated segmentation but performance varies.
Purpose of the Study:
- To identify factors influencing the performance of DL models for glioma segmentation.
- To understand the heterogeneity in DL model performance for preoperative glioma segmentation.
Main Methods:
- Systematic literature review following PRISMA guidelines.
- Meta-regression analysis of 36 DL models for glioma segmentation.
- Extracted data on model and patient characteristics to assess segmentation accuracy (Dice Similarity Coefficient).
Main Results:
- Whole tumor segmentation achieved high accuracy (DSC 0.860).
- 3D and multiparametric MRI inputs improved performance.
- Models trained on BraTS datasets showed higher performance; publication year was the only independent predictor of improved accuracy.
Conclusions:
- 3D architectures and multiparametric MRI inputs are key for DL-based glioma segmentation.
- No single factor fully explains performance variability.
- Further multivariable analyses are needed to optimize DL models for clinical practice.
