Multi-stage Deep Layer Aggregation for Brain Tumor Segmentation

AbstractGliomas are among the most aggressive and deadly brain tumors. This paper details proposed Deep Neural Network architecture for tumor segmentation from Magnetic Resonance Images. The consists of a cascade three Layer Aggregation neural networks, where each stage elaborates response using feature maps probabilities previous stage, MRI channels as inputs. neuroimaging data part publicly available Brain Tumor Segmentation (BraTS) 2020 challenge dataset, we evaluated our proposal in BraTS Validation Test sets. In set, experimental results achieved Dice score 0.8858, 0.8297 0.7900, with an Hausdorff Distance 5.32 mm, 22.32 mm 20.44 whole tumor, core enhanced respectively.KeywordsBrain segmentationDeep learningConvolutional NetworksGaussian filters