Deep Learning in Cancer Diagnosis and Prognosis Prediction: A Minireview on Challenges, Recent Trends, and Future Directions

Table 6

Summary of the studies for the classification, segmentation, prediction, and detection of brain tumors.

Publication	Dataset(s)	Task(s)	Method	Performance
[148]	BRATS 2019	Segmentation	3D fully convolutional network-based multipathway architecture	Dice score of 0.89, 0.78, and 0.76 for WT, TC, and ET subregions, respectively
[149]	BRATS 2018 and BRATS 2019	Segmentation	Combination of U-Net encoding and decoding subarchitecture, dilated convolutional feature extracting layers, and a residual module	Dice score of 0.843, 0.897, and 0.906 and 0.798, 0.902, and 0.845 on ET, WT, and TC brain tumor subregions on BRATS 2018 and BRATS 2019 challenges, respectively
[150]	BRATS 2015, BRATS 2017, and BRATS 2018	Classification	VGG16 and VGG19 transfer learning-based CNN models, partial least square covariance matrix, discrete cosine transform, and extreme learning machine	Accuracy of 97.8%, 96.9%, and 92.5% for BRATS 2015, BRATS 2017, and BRATS 2018 datasets, respectively
[151]	BRATS 2019 and 2019 CPM-RadPath	Classification, segmentation, and prediction	Context-aware CNN architecture for segmentation, 3D CNN architecture for classification, and LASSO for prediction	Dice score of 0.821, 0.895, and 0.835 for ET, WT, and TC regions, respectively, on BRATS 2019 for segmentation task, accuracy of 58.6% for survival prediction task on BRATS 2019 dataset, and balanced accuracy of 63.9% on 2019 CPM-RadPath challenge
[152]	BRATS 2015	Segmentation	Resource-efficient CNN model with memory connections and an adaptive dense block	Dice coefficient score of 0.858, 0.816, and 0.818 for WT, TC, and ET subregions
[153]	Custom	Classification	22-layered CNN architecture	Accuracy of 96.56%
[154]	BRATS 2013, BRATS 2014, BRATS 2017, and BRATS 2018	Segmentation and classification	Inception version 3+LBP	Dice score of 0.8373, 0.937, and 0.7994 for TC, WT, and ET subregions on BRATS 2017; dice score of 0.8834, 0.912, and 0.8184 for TC, WT, and ET on BRATS 2018; average accuracy upward of 92% on BRATS 2013, BRATS 2014, BRATS 2017, and BRATS 2018 datasets
[155]	TCGA database	Classification, segmentation, and detection	U-Net-based DL model using skip connections	Accuracy of 99.7% on the classification task, dice score of 0.9573 on the segmentation task, and Jaccard index of 0.86 on the detection task
[156]	TCIA public access repository	Classification	AlexNet, GoogLeNet, ResNet-50, ResNet-101, and SqueezeNet	An accuracy of 99.04% using AlexNet-type architecture
[157]	BRATS 2018	Segmentation and prediction	3D U-Net model	Dice score of 0.7946, 0.9114, and 0.8304 on ET, WT, and TC, accuracy of 32.1%
[158]	BRATS 2013	Segmentation	Ensemble of deep CNN architectures	Dice score of 0.86, 0.86, and 0.88 on WT, TC, and ET
[159]	Custom	Segmentation and classification	U-Net architecture, VGG16 transfer learning architecture, and a fully connected architecture	Dice score of 0.84; accuracy, sensitivity, and specificity of 92% on the binary classification task
[160]	BRATS 2015	Segmentation and classification	Multiscale 3D CNN architecture	Dice score of 0.89, sensitivity of 0.89, and a specificity of 0.90
[161]	BRATS 2013	Segmentation	DL model combining average pooling and max pooling layers along with kernels	Dice score of 0.80, 0.75, and 0.71 on WT, TC, and ET
[162]	Custom	Segmentation and classification	Multiscale Convolutional Neural Network	Dice score of 0.894, 0.779, 0.813, and 0.828 on meningioma, glioma, pituitary tumor, and average and an accuracy of 97.3%