Table 5.

Overview of papers using deep learning for brain tumor segmentation.

Study	Method	Proposed Solution and Preprocessing Approach	Softwares/Tools/Languages/ Libraries used for Simulation and Implementation	Evaluation
Xiaomei Zhao et al. [1].	Fully Convolutional Neural Networks (FCNNs)	Integration of FCNNs and CRFs	Tesla K80 GPUs and Intel E5-2620 CPUs	Dice Scores Complete = 0.84 Core Tumor = 0.67 Enhancing = 0.62
Mamta Mittal et al. [12].		Stationary Wavelet Transform (SWT) and the new Growing Convolution Neural Network (GCNN).	Not Mention	Accuracy = 98.6 Precision = 0.9881 Recall = 0.9823
Yan Xu et al. [76].	Deep Convolutional Activation Features(CNNs)	CNN Activations Trained by ImageNet to Extract Features through Feature Selection, Feature Pooling, and Data Augmentation	Not Mention	Accuracy = 84%
Eze Benson et al. [89].	Convolutional Neural Network (CNN)	Singular Hourglass Structure	NVIDIA TITAN X GPU	Coefficient = 92%
Chenhong Zhou et al. [90].	Convolutional Neural Network	OM-Net MC-baseline and OM-Net from multiple aspects to further promote the performance.	Not Mention	Dice Scores Enhancing = 0.8136 Whole Tumor = 0.909 Core Tumor = 0.8651
Geena Kim [92].	2D Fully Convolutional Neural Networks	double convolution layers, inception modules, and dense modules were added to a U-Net to achieve a deep architecture	Not Mention	Dice Scores Enhancing = 0.75 Whole Tumor = 0.88 Core Tumor = 0.73
Yan Hu & Yong Xia [93].	Deep Convolutional Neural Network	3D Deep Neural Network-based Algorithm Cascaded U-Net	NVIDIA GTX 1080	Dice Scores Enhancing = 0..55 Whole Tumor = 0.81 Core Tumor = 0.69
Aparna Natarajan & Sathiyasekar Kumarasamy [94].	Fuzzy Logic with Spiking Neuron Model (FL-SNM)		MATLABR2017	Accuracy = 94.87%
Peter D. Chang [98].	Fully Convolutional Neural Networks	Fully Convolutional Residual Neural Network (FCR-NN)	MATLAB R2016a	Dice Scores Complete = 0.87 Core Tumor = 0.81 Enhancing = 0.72
Fabian Isensee et al. [99].	Convolutional Neural Networks	UNet Architecture	Pascal Titan X GPU	Dice Scores Whole = 90.1 Core Tumor = 90.0 Enhancing = 84.5
Sanjay Kumar et al. [100].	Fully Convolution Neural Networks	UNET Architecture	Not Mention	Accuracy = 89%
Guotai Wang et al. [101].	Convolutional neural networks (CNNs)	Fine-tuning-based Segmentation (BIFSeg)	NVIDIA GPU	Accuracy = 88.11%
Yun Jiang et al. [102].	Convolutional Neural Networks	Statistical Thresholding and Multiscale Convolutional Neural Networks (MSCNN)	Not Mention	Dice Coefficient = 86.6% Predictive Positivity Value (PPV) = 88.6% Sensitivity Coefficient = 85.2%
Dongnan Liu et al. [103].	Deep Convolutional Neural Network (DNN)	3D Large Kernel Anisotropic Network	CBICA’s Image Processing Portal	Dice Scores Whole = 0.86 Core Tumor = 0.81 Enhancing = 0.793
Mina Rezaei et al. [104].	3D Conditional Generative Adversarial Network (cGAN)	Adversarial Network, named Voxel-GAN	Keras library and Tensorflow	Dice Scores Whole = 0.84 Core Tumor = 0.79 Enhancing = 0.63 Dice = 0.83 Hausdorff = 9.3 Precision = 0.81 Recall = 0.78
Haocheng Shen et al. [105].	Fully Convolutional Network (FCN)	Boundary-Aware Fully Convolutional Network	Keras and Theano	Dice Scores Complete = 88.7 Core Tumor = 71.8 Enhancing = 72.5
V. Shreyas and Vinod Pankajakshan [106].	Simple Fully Convolutional Network (FCN)	U-Net	Uadro K4000 GPU	Dice Scores Whole = 0.83 Core Tumor = 0.75 Enhancing = 0.72
Nicholas J et al. [107].	Random Forests	Random Forests with ANTsR	ANTsR Package, CMake Tool, R-code	Dice Scores Complete = 0.87 Core Tumor = 0.78 Enhancing = 0.74
Liya Zhao & Kebin Jia [108].	Convolutional Neural Networks (CNNs)	Multi-Scale CNN Architecture of tumor Recognitionon 2D slice and Multiple Intermediate Layers in CNNs	Not Mention	Dice Accuracy = 0.88%
R. Thillaikkarasi & S. Saravanan [109].	CNN with M-SVM	Novel Deep Learning Algorithm (Kernel-based CNN) with M-SVM	Not Mention	Accuracy = 84%
Wu Deng et al. [110].	Convolutional Neural Network	Dense Micro-block Difference Feature (DMDF) and Fisher vector Encoding Non-quantifiable local feature FCNN and Fine Feature Fusion Model	GPU NVIDIA GeForce GTX1070, Ubuntu 16.04 LST 64-Bit operating System	Accuracy = 90.98%
Tony C. W. Mok et al. [111].	Generative Adversarial Networks	Novel automatic data augmentation Coarse-to-Fine Generator to capture the Manifold, Coarse-to-Fine Boundary-Aware Generator CB-GANs	Nvidia GTX1080 Ti GPU	Dice Scores Complete = 0.84 Core Tumor = 0.63 Enhancing = 0.57
Anshika Sharma et al. [112].	Neural Network	Differential Evolution algorithm Embedded with OTSU method Hybridization of Differential Evolution(DE) and OTSU	MATLABR2012a	Accuracy = 94.73%
Zhe Xiao et al. [113].		Coarse-to-Fine and ’Stacked Auto-Encoder’ (SAE). Stacked Denoising Auto Encoder SDAE	Not Mention	Accuracy = 98.04%
Adel Kermi et al. [114].	2D Deep Convolutional Neural Networks (DNNs)	Weighted Cross-Entropy (WCE) and Generalized Dice Loss (GDL) U-net	intel Xeon E5-2650 CPU@ 2.00 GHz (64 GB) and NVIDIA Quadro 4000–448 Core CUDA (2 GB) GPU.	Dice Scores Whole = 0.86 Core Tumor = 0.80 Enhancing = 0.78
Hongdou Yao et al. [115].		Cascaded FCN	GTX 1080Ti GPU	Dice Scores Whole = 0.86 Core Tumor = 0.73 Enhancing = 0.63
Lutao Dai et al. [116].	Deep Convolution Neural Networks	Integration of modified U-Net and its domain-adapted version (DAU-Net).	XGBoost	Dice Scores Whole = 0.91 Core Tumor = 0.85 Enhancing = 0.80
Eric Carver et al. [117].	U-net Neural Network		XGBboost	Dice Scores Whole = 0.88 Core Tumor = 0.76 Enhancing = 0.71
Guotai Wang et al. [118].	Convolutional Neural Networks	Cascade Fully Convolutional Neural Network with multiple layers of Anisotropic and dilated Convolution Filters	NVIDIA TITAN X GPU	Dice Scores Whole = 0.83 Core Tumor = 0.90 Enhancing = 0.78
Sara Sedlar [119].	Convolutional Neural Network (CNN	Multi-Path Convolutional Neural Network (CNN)	nVidia’s GeForce GTX 980 Ti (6 GB) GPU and Intel Core i7-6700K CPU @ 4.00 GHz (32 GB).	Dice Scores Whole = 0.84 Core Tumor = 0.69 Enhancing = 0.60
Zoltan Kap et al. [120].		Decision Trees and Random Forest technique	Not Mention	Dice score = 80.1% Sensitivity = 83.1% Specificity = 98.6%
G. Anand Kumar & P. V. Sridevi [121].	3D Convolutional Neural Network (3DCNN)	EGLCM Feature Extraction to Assess, Evaluate and Produce accurate predictions and detailed segmentation maps.	MATLABR2017a	Not Mention
Hao Dong et al. [122].	Fully Convolutional Networks	U-Net based Deep Convolutional Networks	NVIDIA Titan X (Pascal)	Dice Scores Complete = 0.86 Core Tumor = 0.86 Enhancing = 0.65
David Gering et al. [123].	Convolution Neural Network	Multi-Plane Reformat (MPR)	TensorFlow and Neural Networking API Keras	Dice Scores Active= 0.76 Core Tumor = 0.86 Whole = 0.89
Reza Pourreza et al. [124].	Deeply-Supervised Neural Network	Holistically-Nested Edge Detection (HED) Network	Caffe library Python and NVIDIA Titan Xp graphic card	Dice Scores Whole = 0.86 Core Tumor = 0.60 Enhancing = 0.69
Samya AMIRI [140].		Random forest (RF) based Learning Transfer to SVM RF-SVM cascaded	MATLAB	Mean Dice index Secore = 72.0%
Guotai Wang et al. [141].	Deep Convolutional Neural Networks (CNNs)	3D Unet, Cascaded Network of WNet, TNet and ENet	NVIDIA TITAN X GPU	Dice Scores Whole = 90.21 Core Tumor = 85.83 Enhancing = 79.72
Mikael Agn et al. [142].	Gaussian Mixture Model Combined with a Spatial Atlas-based Tissue Prior Generative Model	Convolutional Restricted Boltzmann Machines (cRBMs)	MATLAB 2014b.	Dice Scores Complete = 87 Core Tumor = 82 Enhancing = 70
Xiangmao Kong et al. [134].	U-Net	Novel Hybrid Pyramid U-Net (HPU-Net) Model for Pixel-Level Prediction	NVIDIA Titan X GPU	Dice Scores Complete = 0.90 Core Tumor = 0.71 Enhancing = 0.78 Predictive Positivity Value (PPV) Complete = 0.91 Core Tumor = 0.87 Enhancing = 0.93 Sensitivity Complete = 0.96 Core Tumor = 0.79 Enhancing = 0.67
Richard McKinley et al. [143].	Convolutional Neural Network (CNN)	Densenet and DeepSCAN	Not Mention	Dice Scores
Pawel Mlynarskia et al. [144].		Deep Learning Fully-Annotated and Weakly-Annotated	TensorFlow	Accuracy = 85.67%