Abstract
Automatic segmentation of brain tumors from medical images is important for clinical assessment and treatment planning of brain tumors. Recent years have seen an increasing use of convolutional neural networks (CNNs) for this task, but most of them use either 2D networks with relatively low memory requirement while ignoring 3D context, or 3D networks exploiting 3D features while with large memory consumption. In addition, existing methods rarely provide uncertainty information associated with the segmentation result. We propose a cascade of CNNs to segment brain tumors with hierarchical subregions from multi-modal Magnetic Resonance images (MRI), and introduce a 2.5D network that is a trade-off between memory consumption, model complexity and receptive field. In addition, we employ test-time augmentation to achieve improved segmentation accuracy, which also provides voxel-wise and structure-wise uncertainty information of the segmentation result. Experiments with BraTS 2017 dataset showed that our cascaded framework with 2.5D CNNs was one of the top performing methods (second-rank) for the BraTS challenge. We also validated our method with BraTS 2018 dataset and found that test-time augmentation improves brain tumor segmentation accuracy and that the resulting uncertainty information can indicate potential mis-segmentations and help to improve segmentation accuracy.
Highlights
In adults, gliomas are the most common primary brain tumors
Unlike model-based uncertainty obtained by test-time dropout (Gal and Ghahramani, 2016; Jungo et al, 2017, 2018), we investigate image-based uncertainty obtained by test-time augmentation that has previously been mainly used for improving segmentation accuracy (Matsunaga et al, 2017; Radosavovic et al, 2018)
This paper is a combination and an extension of our previous works on brain tumor segmentation (Wang et al, 2017, 2018a), where we proposed a cascade of convolutional neural networks (CNNs) for sequential segmentation of brain tumor and the subregions from multi-modal Magnetic Resonance images (MRI), which decomposes the complex task of multi-class segmentation into three simpler binary segmentation tasks
Summary
Gliomas are the most common primary brain tumors They begin in the brain’s glial cells and are typically categorized into different grades: High-Grade Gliomas (HGG) grow rapidly and are more malignant, while Low-Grade Gliomas (LGG) are slower growing tumors with a better patient prognosis (Louis et al, 2016). T2 and FLAIR images mostly highlight the whole tumor region (including infiltrative edema), and T1 and T1ce images give a better contrast for the tumor core region (not including infiltrative edema) (Menze et al, 2015). These different sequences providing complementary information can be combined for the analysis of different subregions of brain tumors
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
