. 2022 Jul 31;12(8):1850. doi: 10.3390/diagnostics12081850

Table 4.

(a) Overview of included studies that focus on CNN-based deep learning methods for brain tumor classification, with the exception of studies focusing on normal vs. tumorous classification. Datasets, MRI sequences, size of the datasets, and preprocessing methods are summarized. (b) Overview of included studies that focus on CNN-based deep learning methods for brain tumor classification, with the exception of study focusing on normal vs. tumorous classification. Classification tasks, classification architecture, validation methods, and performance metrics are summarized.

(a)
Author and Year	Datasets	MRI Sequences	Size of Dataset		Pre-Processing						Data Augmentation
Author and Year	Datasets	MRI Sequences	Patients	Images	Cropping	Normalization	Resizing	Skull Stripping	Registration ¹	Other	Translation ²	Rotation	Scaling ³	Reflection ⁴	Shearing	Cropping	Other (X = Unspecified)
Özcan et al. [27] 2021	Private dataset	T₂w/FLAIR	104 (50 LGG, 54 HGG)	518	x	x				Conversion to BMP		x	x	x	x
Hao et al. [102] 2021	BraTS 2019	T₁w, ceT₁w, T₂w	335 (259 HGG, 76 LGG)	6700			x	x	x
Tripathi et al. [103] 2021	1. TCGA-GBM, 2. LGG-1p19qDeletion	T₂w	322 (163 HGG, 159 LGG)	7392 (5088 LGG, 2304 HGG)				x			x	x	x	x		x
Ge et al. [40] 2020	BraTS 2017	T₁w, ceT₁w, T₂w, FLAIR	285 (210 HGG, 75 LGG)								x			x
Mzoughi et al. [28] 2020	BraTS 2018	ceT₁w	284 (209 HGG, 75 LGG)			x	x			Contrast enhancement				x
Yang et al. [45] 2018	ClinicalTrials.gov (NCT026226201)	ceT₁w	113 (52 LGG, 61 HGG)							Conversion to BMP		x	x	x			Histogram equalization, adding noise
Zhuge et al. [77] 2020	1.TCIA-LGG, 2. BraTS 2018	T₁w, T₂w, FLAIR, ceT₁w	315 (210 HGG, 105 LGG)			x			x	Clipping, bias field correction		x	x	x
Decuyper et al. [73] 2021	1. TCGA-LGG, 2. TCGA-GBM, 3. TCGA-1p19qDeletion, 4. BraTS 2019. 5. GUH dataset	T₁w, ceT₁w, T₂w, FLAIR	738 (164 from TCGA-GBM, 121 from TCGA-LGG, 141 from 1p19qDeletion, 202 from BraTS 2019, 110 from GUH dataset) (398 GBM vs. 340 LGG)			x		x	x	Interpolation		x		x			Elastic transform
He et al. [78] 2021	1.Dataset from TCIA	FLAIR, ceT₁w	214 (106 HGG, 108 LGG)			x	x		x								x
	2. BraTS 2017	FLAIR, ceT₁w	285 (210 HGG, 75 LGG)			x	x		x								x
Hamdaoui et al. [104] 2021	BraTS 2019	T₁w, ceT₁w, T₂w, FLAIR	285 (210 HGG, 75 LGG)	53,064 (26,532 HGG, 26,532 LGG)	x								x	x
Chikhalikar et al. [105] 2021	BraTS 2015	T₂w, FLAIR	274 (220 HGG, 54 LGG)	521						Contrast enhancement
Ahmad [106] 2019	BraTS 2015	No info shared		124 (99 HGG, 25 LGG)		x
Naser et al. [96] 2020	TCGA-LGG	T₁W, FLAIR, ceT₁w	108 (50 Grade II, 58 Grade III)		x	x	x			Padding	x	x	x	x	x
Allah et al. [44] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x						x		x			PGGAN
Swati et al. [50] 2019	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x
Guan et al. [43] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x			Contrast enhancement		x		x
Deepak et al. [39] 2019	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x
Díaz-Pernas et al. [42] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x											Elastic transform
Ismael et al. [49] 2020	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)	x		x				x	x	x	x	x		Whitening, brightness manipulation
Alhassan et al. [107] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x
Bulla et al. [108] 2020	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x
Ghassemi et al. [109] 2020	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x						x		x
Kakarla et al. [110] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x			Contrast enhancement
Noreen et al. [111] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x
Noreen et al. [112] 2020	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x
Kumar et al. [113] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)								x
Badža et al. [114] 2020	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x					x		x
Alaraimi et al. [115] 2021	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x				x	x	x	x		x
Lo et al. [116] 2019	Dataset from TCIA **	ceT₁w	130 (30 Grade II, 43 Grade III, 57 Grade IV)			x	x			Contrast enhancement	x	x	x	x		x
Kurc et al. [117] 2020	Data from TCGA	ceT₁w, T₂-FLAIR	32 (16 OLI, 16 AST)					x	x	Bias field correction		x				x
Pei et al. [118] 2020	1. CPM-RadPath 2019, 2. BraTS 2019	T₁w, ceT₁w, T₂w, FLAIR	398 (329 from CPM-RadPath 2019, 69 from BraTS 2019)			x		x	x	Noise reduction		x	x			x
Ahammed et al. [72] 2019	Private dataset	T₂w	20	557 (130 Grade I, 169 Grade II, Grade III 103, Grade IV 155)				x		Filtering, enhancement	x	x	x	x
Mohammed et al. [51] 2020	Radiopaedia	No info shared	60 (15 of each class)	1258 (311 EP, 286 normal, 380 MEN, 281 MB)			x			Denoising	x	x	x	x		x
McAvoy et al. [119] 2021	Private dataset	ceT₁w	320 (160 GBM, 160 PCNSL)	3887 (2332 GBM, 1555 PCNSL)		x	x			Random changes to color, noise sampling				x
Gilanie et al. [120] 2021	Private dataset	T₁w, T₂w, FLAIR	180 (50 AST-I, 40 AST-II, 40 AST-III, 50 AST-IV)	30240 (8400 AST-I, 6720 AST-II, 6720 AST-III, 8400 AST-IV)		x				Bias field correction		x
Kulkarni et al. [121] 2021	Private dataset	T₁w, T₂w, FLAIR		200 (100 benign, 100 malignant)						Denoising, contrast enhancement	x	x	x	x	x
Artzi et al. [122] 2021	Private dataset	T₁w, FLAIR, DTI	158 (22 Normal, 63 PA, 57 MB, 16 EP)	731 (110 Normal, 280 PA, 266 MB, 75 EP)	x		x		x	Background removal, bias field correction		x	x	x			Brightness changes
Tariciotti et al. [123] 2022	Private dataset	ceT1w	121 (47 GBM, 37 PCNSL, 37 Metastasis)	3597 (1481 GBM, 1073 PCNSL, 1043 Metastasis))		x	x			Conversion to PNG
Ait et al. [124] 2022	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x
Alanazi et al. [125] 2022	1. Dataset from Kaggle	No info shared		826 Glioma, 822 MEN, 395 no tumor, and 827 PT	x	x	x			Noise removal
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)	x	x	x			Noise removal
Ye et al. [126] 2022	Private dataset	ceT₁w	73			x	x			Image transformation				x			Blurring, ghosting, motion, affining, random elastic deformation
Gaur et al. [127] 2022	MRI dataset by Bhuvaji	No info shared		2296			x			Gaussian noise adding
Guo et al. [128] 2022	CPM-RadPath 2020	T₁w, ceT₁w, T₂w, FLAIR	221 (133 GBM, 54 AST, 34 OLI)					x	x	Bias field correction, Gaussian noise adding		x	x				Random contrast adjusting
Aamir et al. [129] 2022	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x				Contrast enhancement		x		x
Rizwan et al. [130] 2022	Figshare (Cheng et al., 2017)	ceT₁w	230 (81 MEN, 90 Glioma, 59 PT)	3061 (707 MEN, 1425 Glioma, 929 PT)	x		x			Noise filtering and smoothing							salt-noise/grayscale di stortion
	Dataset from TCIA	T₁w	513 (204 Grade II, 128 Grade III, 181 Grade IV)	70 (32 Grade II, 18 Grade III, 20 Grade IV)	x		x			Noise filtering and smoothing							salt-noise/grayscale di stortion
Nayak et al. [131] 2022	1.daataset from Kaggle, 2. Figshare (Cheng et al., 2017)	ceT₁w	1. No info shared, 2. 233 (as shown in Table 2)	3260 (196 Normal, 3064 (as shown in Table 2))		x				Gaussian blurring, noise removal	x	x	x
Chatterjee et al. [132] 2022	1.BraTS2019, 2. IXI Dataset	ceT₁w	1. 332 (259 HGG, 73 LGG), 2. 259 Normal			x	x	x						x			Affine
Khazaee et al. [133] 2022	BraTS2019	ceT₁w, T₂w, FLAIR	335 (259 HGG, 76 LGG)	26,904 (13,233 HGG, 13,671 LGG)								x		x
Isunuri et al. [134] 2022	Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x
Gu et al. [30] 2021	1. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)			x
	2. REMBRANDT	No info shared	130	110,020			x
Rajini [135] 2019	1. IXI dataset, REMBRANDT, TCGA-GBM, TCGA-LGG	No info shared	600 normal images from IXI dataset, 130 patients from REMBRANDT, 200 patients from TCGA-GBM, 299 patients from TCGA-LGG
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)
Anaraki et al. [136] 2019	1: IXI dataset, REMBRANDT, TCGA-GBM, TCGA-LGG, private dataset	no info of IXI, ceT₁w from REMBRANDT, TCGA-GBM, TCGA-LGG	600 normal images from IXI dataset, 130 patients from REMBRANDT, 199 patients from TCGA-GBM, 299 patients from TCGA-LGG, 60 patients from private dataset			x	x				x	x	x	x
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x				x	x	x	x
Sajjad et al. [100] 2019	1. Radiopaedia	No info shared		121 (36 Grade I, 32 Grade II, 25 Grade III, 28 Grade IV)		x	x			Denoising, bias field correction		x		x	x		Gaussian blurring, sharpening, embossing, skewing
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)		x	x			Denoising, bias field correction		x		x	x		Gaussian blurring, sharpening, embossing, skewing
Wahlang et al. [137] 2020	1. Radiopaedia	FLAIR	11 (2 Metastasis, 6 Glioma, 3 MEN)													x
	2. BraTS 2017	No info shared	20	3100						Median filtering
Tandel et al. [138] 2021	REMBRANDT	T₂w	See 1–4 below	See 1–4 below			x			Converted to RGB		x	x
			130	1. 2156 (1041 normal, 1091 tumorous)
			47	2. 557 (356 AST-II, 201 AST-III)
			21	3. 219 (128 OLI-II, 91 OLI-III)
			112	4. 1115 (484 LGG, 631 HGG)
Xiao et al. [97] 2021	1. Private dataset	No info shared		1109 (495 MT, 614 Normal)			x
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)			x
	3. Brain Tumor Classification (MRI) Dataset from Kaggle	No info shared		3264 (937 MEN, 926 Glioma, 901 PT, 500 Normal)			x
Tandel et al. [24] 2020	REMBRANDT	T₂w	112 (30 AST-II, 17 AST-II, 14 OLI-II, 7 OLI-III, 44 GBM)	See 1–5 below				x				x	x
				1. 2132 (1041 normal, 1091 tumorous)
				2. 2156 (1041 normal, 484 LGG, 631 HGG)
				3. 2156 (1041 normal, 557 AST, 219 OLI, 339 GBM)
				4. 1115 (356 AST-II, 201 AST-III, 128 OLI-II, 91 OLI-III, 339 GBM) 5. 2156 (1041 normal, 356 AST-II, 201 AST-III, 128 OLI-II, 91 OLI-III, 339 GBM)

Ayadi et al. [98] 2021	1. Radiopaedia	No info shared		121 (36 Grade I, 32 Grade II, 25 Grade III, 28 Grade IV)								x		x			Gaussian blurring, sharpening
	2. Figshare (Cheng et al., 2017)	ceT₁w	233 (as shown in Table 2)	3064 (as shown in Table 2)
	3. REMBRANDT	FLAIR, T₁w, T₂w	130 (47 AST, 21 OLI, 44 GBM, 18 unknown)	See 1–5 below								x		x			Gaussian blurring, sharpening
				1. 2132 (1041 normal, 1091 tumorous) 2. 2156 (1041 normal, 484 LGG, 631 HGG) 3. 2156 (1041 normal, 557 AST, 219 OLI, 339 GBM) 4. 1115 (356 AST-II, 201 AST-III, 128 OLI-II, 91 OLI-III, 339 GBM) 5. 2156 (1041 normal, 356 AST-II, 201 AST-III, 128 OLI-II, 91 OLI-III, 339 GBM)




(b)
Author and Year		Classification Tasks		Model Architecture		Validation				Performance							ACC% ⁵
2 classes
Özcan et al. [27] 2021		LGG (grade II) vs. HGG (grade IV)		Custom CNN model		5-fold CV				SEN = 98.0%, SPE = 96.3%, F1 score = 97.0%, AUC = 0.989							97.1
Hao et al. [102] 2021		LGG vs. HGG		Transfer learning with AlexNet		No info shared				AUC = 82.89%
Tripathi et al. [103] 2021		LGG vs. HGG		Transfer learning with Resnet18		No info shared											95.87
Ge et al. [40] 2020		LGG vs. HGG		Custom CNN model		No info shared				SEN = 84.35%, SPE = 93.65%							90.7
Mzoughi et al. [28] 2020		LGG vs. HGG		Multi-scale 3D CNN		No info shared											96.49
Yang et al. [45] 2018		LGG vs. HGG		Transfer learning with AlexNet, GoogLeNet		5-fold CV				AUC = 0.939							86.7
Zhuge et al. [77] 2020		LGG vs. HGG		Transfer learning with ResNet50		5-fold CV				SEN = 93.5%, SPE = 97.2%							96.3
				3D CNN		5-fold CV				SEN = 94.7%, SPE = 96.8%							97.1
Decuyper et al. [73] 2021		LGG vs. GBM		3D CNN		No info shared				SEN = 90.16%, SPE = 89.80%, AUC = 0.9398							90
He et al. [78] 2021		LGG vs. HGG		Custom CNN model		5-fold CV				TCIA: SEN = 97.14%, SPE = 90.48%, AUC = 0.9349							92.86
										BraTS 2017: SEN = 95.24%, SPE = 92%, AUC = 0.952							94.39
Hamdaoui et al. [104] 2021		LGG vs. HGG		Transfer learning with stacking VGG16, VGG19, MobileNet, InceptionV3, Xception, Inception ResNetV2, DenseNet121		10-fold CV				PRE = 98.67%, F1 score = 98.62%, SEN = 98.33%							98.06
Chikhalikar et al. [105] 2021		LGG vs. HGG		Custom CNN model		No info shared											99.46
Ahmad [106] 2019		LGG vs. HGG		Custom CNN model		No info shared											88
Khazaee et al. [133] 2022		LGG vs. HGG		Transfer learning with EfficientNetB0		CV				PRE = 98.98%, SEN = 98.86%, SPE = 98.79%							98.87%
Naser et al. [96] 2020		LGG (Grade II) vs. LGG (Grade III)		Transfer learning with VGG16		5-fold CV				SEN = 97%, SPE = 98%							95
Kurc et al. [117] 2020	OLI vs. AST		3D CNN		5-fold CV						80
McAvoy et al. [119] 2021		GBM vs. PCNSL		Transfer learning with EfficientNetB4		No info shared				GBM: AUC = 0.94, PCNSL: AUC = 0.95
Kulkarni et al. [121] 2021		Benign vs. Malignant		Transfer learning with AlexNet		5-fold CV				PRE = 93.7%, RE = 100%, F1 score = 96.77%							96.55
				Transfer learning with VGG16		5-fold CV				PRE = 55%, RE = 50%, F1 score = 52.38%							50
				Transfer learning with ResNet18		5-fold CV				PRE = 78.94%, RE = 83.33%, F1 score = 81.07%							82.5
				Transfer learning with ResNet50		5-fold CV				PRE = 95%, RE = 55.88%, F1 score = 70.36%							60
				Transfer learning with GoogLeNet		5-fold CV				PRE = 75%, RE = 100%, F1 score = 85.71%							87.5
Wahlang et al. [137] 2020		HGG vs. LGG		AlexNet		No info shared											62
				U-Net		No info shared											60
Xiao et al. [97] 2021		MT vs. Normal		Transfer learning with ResNet50		3-fold, 5-fold, 10-fold CV				AUC = 0.9530							98.2
Alanazi et al. [125] 2022		Normal vs. Tumorous		Custom CNN		No info shared											95.75%
Tandel et al. [138] 2021		1. Normal vs. Tumorous		DL-MajVot (AlexNet, VGG16, ResNet18, GoogleNet, ResNet50)		5-fold CV				SEN = 96.76%, SPE = 96.43%, AUC = 0.966							96.51
		2. AST-II vs. AST-III		DL-MajVot (AlexNet, VGG16, ResNet18, GoogleNet, ResNet50)		5-fold CV				SEN = 94.63%, SPE = 99.44%, AUC = 0.9704							97.7
		3. OLI-II vs. OLI-III		DL-MajVot (AlexNet, VGG16, ResNet18, GoogleNet, ResNet50)		5-fold CV				SEN = 100%, SPE = 100%, AUC = 1							100
		4. LGG vs. HGG		DL-MajVot (AlexNet, VGG16, ResNet18, GoogleNet, ResNet50)		5-fold CV				SEN = 98.33%, SPE = 98.57%, AUC = 0.9845							98.43
Tandel et al. [24] 2020		Normal vs. Tumorous		Transfer learning with AlexNet		Multiple CV (K2, K5, K10)				RE = 100%, PRE = 100%, F1 score = 100%							100
Ayadi et al. [98] 2021		Normal vs. Tumorous		Custom CNN model		5-fold CV											100
Ye et al. [126] 2022		Germinoma vs. Glioma		Transfer learning with ResNet18		5-fold CV				AUC = 0.88							81%
3 classes
Allah et al. [44] 2021		MEN vs. Glioma vs. PT		PGGAN-augmentation VGG19		No info shared											98.54
Swati et al. [50] 2019		MEN vs. Glioma vs. PT		Transfer learning with VGG19		5-fold CV				SEN = 94.25%, SPE = 94.69%, PRE = 89.52%, F1 score = 91.73%							94.82
Guan et al. [43] 2021		MEN vs. Glioma vs. PT		EfficientNet		5-fold CV											98.04
Deepak et al. [39] 2019		MEN vs. Glioma vs. PT		Transfer learning with GoogleNet		5-fold CV											98
Díaz-Pernas et al. [42] 2021		MEN vs. Glioma vs. PT		Multiscale CNN		5-fold CV											97.3
Ismael et al. [49] 2020		MEN vs. Glioma vs. PT		Residual networks		5-fold CV				PRE = 99.0%, RE = 99.0%, F1 score = 99.0%							99
Alhassan et al. [107] 2021		MEN vs. Glioma vs. PT		Custom CNN model		k-fold CV				PRE = 99.6%, RE = 98.6%, F1 score = 99.0%							98.6
Bulla et al. [108] 2020		MEN vs. Glioma vs. PT		Transfer learning with InceptionV3 CNN model		holdout validation, 10-fold CV, stratified 10-fold CV, group 10-fold CV				Under group 10-fold CV: PRE = 97.57%, RE = 99.47%, F1 score = 98.40%, AUC = 0.995							99.82
Ghassemi et al. [109] 2020		MEN vs. Glioma vs. PT		CNN-GAN		5-fold CV				PRE = 95.29%, SEN = 94.91%, SPE = 97.69%, F1 score = 95.10%							95.6
Kakarla et al. [110] 2021		MEN vs. Glioma vs. PT		Custom CNN model		5-fold CV				PRE = 97.41%, RE = 97.42%							97.42
Noreen et al. [111] 2021		MEN vs. Glioma vs. PT		Transfer learning with Inception-v3		K-fold CV											93.31
				Transfer learning with Inception model		K-fold CV											91.63
Noreen et al. [112] 2020		MEN vs. Glioma vs. PT		Transfer learning with Inception-v3		No info shared											99.34
				Transfer learning with DensNet201		No info shared											99.51
Kumar et al. [113] 2021		MEN vs. Glioma vs. PT		Transfer learning with ResNet50		5-fold CV				PRE = 97.20%, RE = 97.20%, F1 score = 97.20%
Badža et al. [114] 2020		MEN vs. Glioma vs. PT		Custom CNN model		10-fold CV				PRE = 95.79%, RE = 96.51%, F1 score = 96.11%							96.56
Ait et al. [124] 2022		MEN vs. Glioma vs. PT		Custom CNN		No info shared				PRE = 98.3%, SEN = 98.6%, F1 score = 98.6%							98.70%
Alanazi et al. [125] 2022		MEN vs. Glioma vs. PT		Custom CNN		No info shared											96.90%
Gaur et al. [127] 2022		MEN vs. Glioma vs. PT		Custom CNN		k-fold CV											94.64%
Aamir et al. [129] 2022		MEN vs. Glioma vs. PT		Custom CNN		5-fold CV											98.95%
Rizwan et al. [130] 2022		MEN vs. Glioma vs. PT		Custom CNN		No info shared											99.8%
Isunuri et al. [134] 2022		MEN vs. Glioma vs. PT		Custom CNN		5-fold CV				PRE = 97.33%, SEN = 97.19%, F1 score = 97.26%							97.52%
Alaraimi et al. [115] 2021		MEN vs. Glioma vs. PT		Transfer learning with AlexNet		No info shared				AUC = 0.976							94.4
				Transfer learning with VGG16		No info shared				AUC = 0.981							100
				Transfer learning with GoogLeNet		No info shared				AUC = 0.986							98.5
Lo et al. [116] 2019		Grade II vs. Grade III vs. Grade IV		Transfer learning with AlexNet		10-fold CV											97.9
Pei et al. [118] 2020		GBM vs. AST vs. OLI		3D CNN		No info shared											74.9
Gu et al. [30] 2021		1. MEN vs. Glioma vs. PT		Custom CNN model		5-fold CV				SEN = 94.64%, PRE = 94.61%, F1 score = 94.70%							96.39
		2. GBM vs. AST vs. OLI		Custom CNN model		5-fold CV				SEN = 93.66%, PRE = 95.12%, F1 score = 94.05%							97.37
Rajini [135] 2019		MEN vs. Glioma vs. PT		Custom CNN model		5-fold CV											98.16
Anaraki et al. [136] 2019		MEN vs. Glioma vs. PT		Custom CNN model		5-fold CV											94.2
Sajjad et al. [100] 2019		MEN vs. Glioma vs. PT		Transfer learning with VGG19		No info shared				SEN = 88.41%, SPE = 96.12%							94.58
Wahlang et al. [137] 2020		Metastasis vs. Glioma vs. MEN		Lenet		No info shared											48
				AlexNet		No info shared											75
Xiao et al. [97] 2021		MEN vs. Glioma vs. PT		Transfer learning with ResNet50		3-fold, 5-fold, 10-fold CV											98.02
Tandel et al. [24] 2020		Normal vs. LGG vs. HGG		Transfer learning with AlexNet		Multiple CV (K2, K5, K10)				RE = 94.85%, PRE = 94.75%, F1 score = 94.8%							95.97
Chatterjee et al. [132] 2022		Normal vs. HGG vs. LGG		Transfer learning with ResNet		3-fold CV				F1 score = 93.45%							96.84%
Ayadi et al. [98] 2021		1. Normal vs. LGG vs. HGG		Custom CNN model		5-fold CV											95
		2. MEN vs. Glioma vs. PT		Custom CNN model		5-fold CV											94.74
Guo et al. [128] 2022		GBM vs. AST vs. OLI		Custom CNN		3-fold CV				SEN = 0.772, SPE = 93.0%, AUC = 0.902							87.8%
Rizwan et al. [130] 2022		Grade I vs. Grade II vs. Grade III		Custom CNN		No info shared											97.14%
Tariciotti et al. [123] 2022		Metastasis vs. GBM vs. PCNSL		Resnet101		Hold-out				PRE = 91.88%, SEN = 90.84%, SPE = 96.34%, F1 score = 91.0%, AUC = 0.92							94.72%
4 classes
Ahammed et al. [72] 2019		Grade I vs. Grade II vs. Grade III vs. Grade IV		VGG19		No info shared				PRE = 94.71%, SEN = 92.72%, SPE = 98.13%, F1 score = 93.71%							98.25
Mohammed et al. [51] 2020		EP vs. MEN vs. MB vs. Normal		Custom CNN model		No info shared				SEN = 96%, PRE = 100%							96
Gilanie et al. [120] 2021		AST-I vs. AST-II vs. AST-III vs. AST-IV		Custom CNN model		No info shared											96.56
Artzi et al. [122] 2021		Normal vs. PA vs. MB vs. EP		Custom CNN model		5-fold CV											88
Nayak et al. [131] 2022		Normal vs. MEN vs. Glioma vs. PT		Transfer learning with EfficientNet		No info shared				PRE = 98.75%, F1 score = 98.75%							98.78%
Rajini [135] 2019		Normal vs. Grade II vs. Grade III vs. Grade IV		Custom CNN model		5-fold CV											96.77
Anaraki et al. [136] 2019		Normal vs. Grade II vs. Grade III vs. Grade IV		Custom CNN model		5-fold CV
Sajjad et al. [100] 2019		Grade I vs. Grade II vs. Grade III vs. Grade IV		Transfer learning with VGG19		No info shared											90.67
Xiao et al. [97] 2021		MEN vs. Glioma vs. PT vs. Normal		Transfer learning with ResNet50		3-fold, 5-fold, 10-fold CV				PRE = 97.43%, RE = 97.67%, SPE = 99.24%, F1 score = 97.55%							97.7
Tandel et al. [24] 2020		Normal vs. AST vs. OLI vs. GBM		Transfer learning with AlexNet		Multiple CV (K2, K5, K10)				RE = 94.17%, PRE = 95.41%, F1 score = 94.78%							96.56
Ayadi et al. [98] 2021		1. normal vs. AST vs. OLI vs. GBM		Custom CNN model		5-fold CV											94.41
		2. Grade I vs. Grade II vs. Grade III vs. Grade IV		Custom CNN model		5-fold CV											93.71
5 classes
Tandel et al. [24] 2020		AST-II vs. AST-III vs. OLI-II vs. OLI-III vs. GBM-IV		Transfer learning with AlexNet		Multiple CV (K2, K5, K10)				RE = 84.4%, PRE = 89.57%, F1 score = 86.89%							87.14
Ayadi et al. [98] 2021		AST-II vs. AST-III vs. OLI-II vs. OLI-III vs. GBM		Custom CNN model		5-fold CV											86.08
6 classes
Tandel et al. [24] 2020		Normal vs. AST-II vs. AST-III vs. OLI-II vs. OLI-III vs. GBM-IV		Transfer learning with AlexNet		Multiple CV (K2, K5, K10)				RE = 91.51%, PRE = 92.46%, F1 score = 91.97%							93.74
Ayadi et al. [98] 2021		normal vs. AST-II vs. AST-III vs. OLI-II vs. OLI-III vs. GBM		Custom CNN model		5-fold CV											92.09

Notes: ¹ Rigid registration unless otherwise notes; ² translation also referred to as shifting; ³ scaling also referred to as zooming; ⁴ reflection also referred to as flipping or mirroring; ** The Cancer Imaging Archive, https://www.cancerimagingarchive.net/ (accessed on 27 July 2022). ⁵ Referring to overall accuracy, mean accuracy, or highest accuracy depending on the information provided by the paper or the highest accuracy when multiple models are used.