Related Experiment Video
Updated: Dec 22, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.3K
DeepSeg: deep neural network framework for automatic brain tumor segmentation using magnetic resonance FLAIR images
Ramy A Zeineldin1, Mohamed E Karar2, Jan Coburger3
1Research Group Computer Assisted Medicine (CaMed), Reutlingen University, 72762, Reutlingen, Germany. Ramy.Zeineldin@Reutlingen-University.DE.
Summary
This study introduces DeepSeg, a novel deep learning framework for automated brain tumor segmentation using FLAIR MRI. DeepSeg successfully distinguishes tumor boundaries, improving diagnostic accuracy for aggressive gliomas.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Gliomas are aggressive brain tumors with challenging-to-define boundaries in clinical practice.
- Fluid-attenuated inversion recovery (FLAIR) Magnetic Resonance Imaging (MRI) aids in visualizing tumor infiltration.
Purpose of the Study:
- To propose DeepSeg, a generic deep learning architecture for automated brain lesion detection and segmentation.
- To utilize FLAIR MRI data for enhanced tumor boundary identification.
Main Methods:
- Developed a modular decoupling framework (DeepSeg) with encoder-decoder architecture.
- Employed Convolutional Neural Networks (CNNs) including ResNet, DenseNet, and NASNet within a modified U-Net.
- Tested on the BraTS 2019 challenge dataset with 336 training and 125 validation cases.
Main Results:
- Achieved Dice scores between 0.81 and 0.84.
- Obtained Hausdorff distance scores ranging from 9.8 to 19.7.
- Demonstrated successful online testing and evaluation of the deep learning models.
Conclusions:
- Confirmed the feasibility and comparative performance of DeepSeg for automated brain tumor segmentation.
- Highlighted the potential of various deep learning models within the DeepSeg framework.
- The DeepSeg framework is open-source and publicly available for research and clinical use.
