2022 30th International Conference on Electrical Engineering (ICEE) · 2022 · 23 citations · 14 references
ComputingComputer ScienceOncologyComputer VisionMachine LearningNeuro-oncologyBrain TumorsComputer-aided DiagnosisDeep LearningImage SegmentationMedical Image AnalysisMedical ImagingAutomatic SegmentationComputer-aided Tumor DiagnosisConvolutional Neural Network SegmentationConvolutional Neural NetworksDiceSegmentationTumor Segmentation
Segmentation of brain tumors helps with the diagnosis and treatment tasks. Due to the large number of patients and the high cost of manual segmentation, researchers proposed automatic segmentation methods. The most popular of these are methods based on deep learning and neural networks. In this paper, we implemented an automatic segmentation using U-net++, which is based on deep convolutional neural networks, on two publicly available data sets. Using the U-Net++architecture, the whole tumor (WT), is segmented with an accuracy of 90.37% based on the Dice similarity coefficient (DSC) for the Brats2018 dataset and 89.13% for the Brats2015 dataset. For comparison with the prior works, we implemented an alternative approach using UNet architecture, which segmented WT with an accuracy of 89.21% for the Brats2018 dataset and 89.12% for the Brats2015 dataset. As the results suggest, leveraging U-Net++ for tumor segmentation provides improvement in WT segmentation at a cost of modest increase in runtime.
14
Rethinking the Inception Architecture for Computer Vision
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe et al. · 2016 · 30.9K citations
N4ITK: Improved N3 Bias Correction
Nicholas J. Tustison, Brian Avants, Philip A. Cook et al. · IEEE Transactions on Medical Imaging · 2010 · 6.4K citations
Medicine, Neuroscience, Radiology +16