Table 2.
A succinct survey of deep-learning-based histopathological image classification methods. NA indicates either “not available” or “no answer” from the associated authors.
| Year | Ref | Aim | Technique | Dataset | Sample | Training (%) | Testing (%) | Result | Performance | |
|---|---|---|---|---|---|---|---|---|---|---|
| AUC | ACC | |||||||||
| 2016 | Chan and Tuszynski [80] | To predict tumor malignancy in breast cancer | Employed binarization, fractal dimension, SVM | BreaKHis [33] | 7909 | 50 | 50 | ACC of 97.90%, 16.50%, 16.50%, and 25.30% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 39.05% |
| Spanhol et al. [33] | To classify histopathological images | Employed CNN based on AlexNet [81] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 90.0%, 88.4%, 84.6%, and 86.1% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 87.28% | |
| Bayram-oglu et al. [38] | To classify breast cancer histopathology images | Employed single-task CNN and multitask CNN | BreaKHis [33] | 7909 | 70 | 30 | For single-task CNN, ACC of 83.08%, 83.17%, 84.63%, and 82.10%, obtained for 40x, 100x, 200x, and 400x magnification factors, respectively; accordingly, for multitask CNN, ACC of 81.87%, 83.39%, 82.56%, and 80.69% | NA | 82.69% | |
| Abbas [77] | To diagnose breast masses | Applied SURF [82], LBPV [83] | DDSM [84], MIAS [85] | 600 | 40 | 60 | Overall 92%, 84.20%, 91.50%, and 0.91 obtained for sensitivity, specificity, ACC, and AUC, respectively | 0.91 | 91.50% | |
|
| ||||||||||
| 2017 | Song et al. [21] | To classify histopathology images | Employed a model of CNN, Fisher vector [86], SVM | BreaKHis [33], IICBU2008 [87] | 8283 | 70 | 30 | ACC of 94.42%, 89.49%, 87.25%, and 85.62% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 89.19% |
| Wei et al. [22] | To analyze tissue images | Employed a modification of GoogLeNet [88] | BreaKHis [33] | 7909 | 75 | 25 | ACC of 97.46%, 97.43%, 97.73%, and 97.74% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 97.59% | |
| Das et al. [23] | To classify histopathology images | Employed GoogLeNet [88] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 94.82%, 94.38%, 94.67%, and 93.49% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 94.34% | |
| Kahya et al. [89] | To identify features of breast cancer | Employed dimensionality reduction, adaptive sparse SVM | BreaKHis [33] | 7909 | 70 | 30 | ACC of 94.97%, 93.62%, 94.54%, and 94.42% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 94.38% | |
| Song et al. [24] | To classify histopathology images easily | Employed CNN-based Fisher vector [86], SVM | BreaKHis [33] | 7909 | 70 | 30 | ACC of 90.02%, 88.90%, 86.90%, and 86.30% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 88.03% | |
| Gupta and Bhavsar [90] | To classify histopathology images. | Employed an integrated model | BreaKHis [33] | 7909 | 70 | 30 | Average ACC of 88.09% and 88.40% obtained for image and patient levels, respectively | NA | 88.25% | |
| Dhungel et al. [91] | To analyze masses in mammograms | Applied multiscale deep belief nets | INbreast [92] | 410 | 60 | 20 | The best results on the testing set with an ACC got 95% on manual and 91% on the minimal user intervention setup | 0.76 | 91.03% | |
| Spanhol et al. [34] | To classify breast cancer images | Using deep CNN | BreaKHis [33] | 7900 | 70 | 30 | ACC of 84.30%, 84.35%, 85.25% and 82.10% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 83.96% | |
| Han et al. [35] | To study breast cancer multiclassification | Employed class structure based CNN | BreaKHis [33] | 7909 | 50 | 50 | ACC of 95.80%, 96.90%, 96.70%, and 94.9% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 96.08% | |
| Sun and Binder [39] | To assess performance of H&E stain dat. | A comparative study among ResNet-50 [75], CaffeNet [93], and GoogLeNet [88] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 85.75%, 87.03%, and 84.18% obtained for GoogLeNet [88], ResNet-50 [75], and CaffeNet [93], respectively | NA | 85.65% | |
| Kaymak et al. [94] | To organize breast cancer images | Back-Propagation [95] and Radial Basis Neural Networks [96] | 176 images from a hospital | 176 | 65 | 35 | Overall ACC of 59.0% and 70.4% got from Back-Propagation [95] and Radial Basis [96], respectively | NA | 64.70% | |
| Liu et al. [47] | To detect cancer metastases in images | Employed a CNN architecture | Camelyon16 [97] | 110 | 68 | 32 | An AUC of 97.60 (93.60, 100) obtained on par with Camelyon16 [97] test set performance | 0.97 | 95.00% | |
| Zhi et al. [57] | To diagnose breast cancer images | Employed a variation of VGGNet [98] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 91.28%, 91.45%, 88.57%, and 84.58% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 88.97% | |
| Chang et al. [58] | To solve the limited amount of training data | Employed CNN model from Inception [88] family (e.g., Inception V3) | BreaKHis [33] | 4017 | 70 | 30 | ACC of 83.00% for benign class and 89.00% for malignant class. AUC of malignant was 93.00% and AUC of benign was also 93.00% | 0.93 | 86.00% | |
|
| ||||||||||
| 2018 | Jannesari et al. [6] | To classify breast cancer images | Employed variations of Inception [88], ResNet [75] | BreaKHis [33], 6402 images from TMA [99] | 14311 | 85 | 15 | With ResNets ACC of 99.80%, 98.70%, 94.80%, and 96.40% obtained for four cancer types. Inception V2 with fine-tuning all layers got ACC of 94.10% | 0.99 | 96.34% |
| Bardou et al. [7] | To classify breast cancer based on histology images | Employed CNN topology, data augmentation | BreaKHis [33] | 7909 | 70 | 30 | ACC of 98.33%, 97.12%, 97.85%, and 96.15% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 97.36% | |
| Kumar and Rao [9] | To train CNN for using image classification | Employed CNN topology | BreaKHis [33] | 7909 | 70 | 30 | ACC of 85.52%, 83.60%, 84.84%, and 82.67% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 84.16% | |
| Das et al. [11] | To classify breast histopathology images | Employed variation of CNN model | BreaKHis [33] | 7909 | 80 | 20 | ACC of 89.52%, 89.06%, 88.84%, and 87.67% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 88.77% | |
| Nahid et al. [100] | To classify biomedical breast cancer images | Employed Boltzmann machine [101], Tamura et al. features [102] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 88.70%, 85.30%, 88.60%, and 88.40% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 87.75% | |
| Badejo et al. [103] | To classify medical images | Employed local phase quantization, SVM | BreaKHis [33] | 7909 | 70 | 30 | ACC of 91.10%, 90.70%, 86.20%, and 84.30% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 88.08% | |
| Alireza-zadeh et al. [104] | To arrange breast cancer images | Threshold adjacency [105], quadratic analysis [106] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 89.16%, 87.38%, 88.46%, and 86.68% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 87.92% | |
| Du et al. [13] | To distribute breast cancer images | Employed AlexNet [81] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 90.69%, 90.46%, 90.64%, and 90.96% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 90.69% | |
| Gandom-kar et al. [14] | To model CNN for breast cancer image diagnosis | Employed a variation of ResNet [75] (e.g., ResNet152) | BreaKHis [33] | 7786 | 70 | 30 | ACC of 98.60%, 97.90%, 98.30%, and 97.60% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 98.10% | |
| Gupta and Bhavsar [15] | To model CNN for breast cancer image diagnosis | Employed DenseNet [67], XGBoost classifier [107] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 94.71%, 95.92%, 96.76%, and 89.11% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 94.12% | |
| Ben-hammou et al. [17] | To study CNN for breast cancer images | Employed Inception V3 [88] module | BreaKHis [33] | 7909 | 70 | 30 | ACC of 87.05%, 82.80%, 85.75%, and 82.70% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 84.58% | |
| Morillo et al. [108] | To label breast cancer images | Employed KAZE features [109] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 86.15%, 80.70%, 77.95%, and 72.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 97.20% | |
| Chattoraj and Vishwakarma [110] | To study breast carcinoma images | Zernike moments [111], entropies of Renyi [112], Yager [113] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 87.7%, 85.8%, 88.0%, and 84.6% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 96.53% | |
| Sharma and Mehra [19] | To analyze behavior of magnification independent breast cancer | Employed models of VGGNet [98] and ResNet [75] (e.g., VGG16, VGG19, and ResNet50) | BreaKHis [33] | 7909 | 90 | 10 | Pretrained VGG16 with logistic regression classifier showed the best performance with 92.60% ACC, 95.65% AUC, and 95.95% ACC precision score for 90%–10% training-testing data splitting | 0.95 | 94.28% | |
| Zheng et al. [114] | To study content-based image retrieval | Employed binarization encoding, Hamming distance [115] | BreaKHis [33] and others | 16309 | 70 | 30 | ACC of 47.00%, 40.00%, 40.00%, and 37.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 41.00% | |
| Cascianelli et al. [20] | To study features extraction from images | Employed dimensionality reduction using CNN | BreaKHis [33] | 7909 | 75 | 25 | ACC of 84.00%, 88.20%, 87.00%, and 80.30% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 84.88% | |
| Mukkamala et al. [116] | To study deep model for feature extraction | Employed PCANet [117] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 96.12%, 97.41%, 90.99%, and 85.85% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 92.59% | |
| Mahraban Nejad et al. [51] | To retrieve breast cancer images | Employed a variation of VGGNet [98], SVM | BreaKHis [33] | 7909 | 98 | 02 | An average ACC of 80.00% was demonstrated from BreaKHis [33] | NA | 80.00% | |
| Rakhlin et al. [118] | To analyze breast cancer images | Several deep neural networks and gradient boosted trees classifier | BACH [78] | 400 | 75 | 25 | For 4-class classification task ACC was 87.2% but for 2-class classification ACC was reported to be 93.8% | 0.97 | 90.50% | |
| Almasni et al. [119] | To detect breast masses | Applied regional deep learning technique | DDSM [84] | 600 | 80 | 20 | Distinguished between benign and malignant lesions with an overall ACC of 97% | 0.96 | 97.00% | |
|
| ||||||||||
| 2019 | Kassani et al. [8] | To use deep learning for binary classification of breast histology images | Usage of VGG19 [98], MobileNet [120], and DenseNet [67] | BreaKHis [33], ICIAR2018 [78], PCam [121], Bioimaging2015 [122] | 8594 | 87 | 13 | Multimodel method got better predictions than single classifiers and other algorithms with ACC of 98.13%, 95.00%, 94.64% and 83.10% obtained for BreaKHis [33], ICIAR2018 [78], PCam [121], and Bioimaging2015 [122], respectively | NA | 92.72% |
| Alom et al. [10] | To classify breast cancer from histopathological images | Inception recurrent residual CNN | BreaKHis [33], Bioimaging2015 [122] | 8158 | 70 | 30 | From BreaKHis [33], ACC of 97.90%, 97.50%, 97.30%, and 97.40%, obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | 0.98 | 97.53% | |
| Nahid and Kong [12] | To classify histopathological breast images | Employed RGB histogram [123] with CNN | BreaKHis [33] | 7909 | 85 | 15 | ACC of 95.00%, 96.60%, 93.500%, and 94.20% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 94.68% | |
| Jiang et al. [16] | To use CNN for breast cancer histopathological images | Employed CNN, Squeeze-and-Excitation [124] based ResNet [75] | BreaKHis [33] | 7909 | 70 | 30 | The achieved accuracy between 98.87% and 99.34% for the binary classification as well as between 90.66% and 93.81% for the multiclass classification | 0.99 | 95.67% | |
| Sudharshan et al. [18] | To use instance learning for image sorting | Employed CNN-based multiple instance learning algorithm | BreaKHis [33] | 7909 | 70 | 30 | ACC of 86.59%, 84.98%, 83.47%, and 82.79% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 84.46% | |
| Gupta and Bhavsar [25] | To segment breast cancer images | Employed ResNet [75] for multilayer feature extraction | BreaKHis [33] | 7909 | 70 | 30 | ACC of 88.37%, 90.29%, 90.54%, and 86.11% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 88.82% | |
| Vo et al. [125] | To extract visual features from training images | Combined weak classifiers into a stronger classifier | BreaKHis [33], Bioimaging2015 [122] | 8194 | 87 | 13 | ACC of 95.10%, 96.30%, 96.90%, and 93.80% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 95.56% | |
| Qi et al. [32] | To organize breast cancer images | Employed a CNN network to complete the classification task | BreaKHis [33] | 7909 | 70 | 30 | ACC of 91.48%, 92.20%, 93.01%, and 92.58% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 92.32% | |
| Talo [41] | To detect and classify diseases in images | DenseNet [67], ResNet [75] (e.g., DenseNet161, ResNet50) | KimiaPath24 [126] | 25241 | 80 | 20 | DenseNet161 pretrained and ResNet50 achieved ACC of 97.89% and 98.87% on grayscale and color images, respectively | NA | 98.38% | |
| Li et al. [127] | To detect invading component in cancer images | Convolutional autoencoder-based contrast pattern mining framework | 361 samples of the breast cancer | 361 | 90 | 10 | ACC was taken into account. The overall ACC achieved was 76.00%, whereas 77.70% was presented for F1S | NA | 76.00% | |
| Ragab et al. [44] | To detect breast cancer from images | AlexNet [81] and SVM | DDSM [84], CBIS-DDSM [128] | 2781 | 70 | 30 | The deep CNN presented an ACC of 73.6%, whereas the SVM demonstrated an ACC of 87.2% | 0.88 | 73.60% | |
| Romero et al. [45] | To study cancer images | A modification of Inception module [88] | HASHI [129] | 151465 | 63 | 37 | From deep learning networks, an overall ACC of 89.00% was demonstrated along with F1S of 90.00% | 0.96 | 89.00% | |
| Minh et al. [46] | To diagnose breast cancer images | A modification of ResNet [75] and InceptionV3 [88] | BACH [78] | 400 | 70 | 20 | ACC with 95% for 4 types of cancer classes and ACC with 97.5% for two combined groups of cancer | 0.97 | 96.25% | |
|
| ||||||||||
| 2020 | Stanitsas et al. [130] | To visualize a health system for clinicians | Employed region covariance [131], SVM, multiple instance learning [132] | FABCD [133], BreaKHis [33] | 7949 | 70 | 15 | ACC of 91.27% and 92.00% at the patient and image level, respectively | 0.98 | 91.64% |
| Togacar et al. [26] | To analyze breast cancer images rapidly | Employed a ResNet [75] architecture with attention modules | BreaKHis [33] | 7909 | 80 | 20 | ACC of 97.99%, 97.84%, 98.51%, and 95.88% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 97.56% | |
| Asare et al. [134] | To study breast cancer images | Employed self-training and self-paced learning | BreaKHis [33] | 7909 | 70 | 30 | ACC of 93.58%, 91.04%, 93.38%, and 91.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 92.25% | |
| Gour et al. [28] | To diagnose breast cancer tumors images | Employed a modification of ResNet [75] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 90.69%, 91.12%, 95.36%, and 90.24% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | 0.91 | 92.52% | |
| Li et al. [29] | To grade pathological images | Employed a modification of Xception network [135] | BreaKHis [33], VLAD [136], LSC [137] | 8583 | 60 | 40 | ACC of 95.13%, 95.21%, 94.09%, and 91.42% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 93.96% | |
| Feng et al. [138] | To allocate breast cancer images | Deep neural-network-based manifold preserving autoencoder [139] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 90.12%, 88.89%, 91.57%, and 90.25% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 90.53% | |
| Parvin and Mehedi Hasan [31] | To study CNN models for cancer images | LeNet [140], AlexNet [81], VGGNet [98], ResNet [75], Inception V3 [88] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 89.00%, 92.00%, 94.00% and 90.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | 0.85 | 91.25% | |
| Carvalho et al. [141] | To classify histological breast images | Entropies of Shannon [142], Renyi [112], Tsallis [143] | BreaKHis [33] | 4960 | 70 | 30 | ACC of 95.40%, 94.70%, 97.60%, and 95.50% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | 0.99 | 95.80% | |
| Li et al. [144] | To analyze breast cancer images | Employed global covariance pooling module [145] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 96.00%, 96.16%, 98.01%, and 95.97% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 94.93% | |
| Man et al. [36] | To classify cancer images | Usage of generative adversarial networks, DenseNet [67] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 97.72%, 96.19%, 86.66%, and 85.18% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 91.44% | |
| Kumar et al. [37] | To classify human breast cancer and canine mammary tumors | Employed a framework based on a variant of VGGNet [98] (e.g., VGGNet16) and SVM | BreaKHis [33] and CMTHis [37] | 8261 | 70 | 30 | For BreaKHis [33], ACC of 95.94%, 96.22%, 98.15%, and 94.41% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively; the same for CMTHis [37], ACC of 94.54%, 97.22%, 92.07%, and 82.84% obtained | 0.95 | 96.93% | |
| Kaushal and Singla [40] | To detect cancerous cells in images. | Employed a CNN model of self-training and self-paced learning | Total 50 images of various patients | 50 | 90 | 10 | ACC was taken into account. Estimation of the standard error of mean was approximately 0.81 | NA | 93.10% | |
| Hameed et al. [43] | To use deep learning for classification of breast cancer images | Variants of VGGNet [98] (e.g., fully trained VGG16, fine-tuned VGG16, fully trained VGG19, and fine-tuned VGG19 models) | Breast cancer images: 675 for training and 170 for testing | 845 | 80 | 20 | The ensemble of fine-tuned VGG16 and VGG19 models offered sensitivity of 97.73% for carcinoma class and overall accuracy of 95.29%. It also offered an F1 score of 95.29% | NA | 95.29% | |
| Alantari et al. [48] | To detect breast lesions in digital X-ray mammograms | Adopted three deep CNN models | INbreast [92], DDSM [84] | 1010 | 70 | 20 | In INbreast [92] mean ACC of 89%, 93%, and 95% for CNN, ResNet50, and Inception-ResNet V2, respectively; 95%, 96%, and 98% for DDSM [146] | 0.96 | 94.08% | |
| Zhang et al. [49] | To classify breast mass | ResNet [75], DenseNet [67], VGGNet [98] | CBIS-DDSM [128], INbreast [92] | 3168 | 70 | 30 | Overall ACC of 90.91% and 87.93% obtained from CBIS-DDSM [128] and INbreast [92], respectively | 0.96 | 89.42% | |
| Hassan et al. [59] | To classify breast cancer masses | Modification of AlexNet [22] and GoogLeNet [88] | CBIS-DDSM [128], MIAS [85], INbreast [92], etc | 600 | 75 | 17 | With CBIS-DDSM [128] and INbreast [92] databases, the modified GoogLeNet achieved ACC of 98.46% and 92.5%, respectively | 0.97 | 96.98% | |
|
| ||||||||||
| 2021 | Li et al. [147] | To use high-resolution info of images | Multiview attention-guided multiple instance detection network | BreaKHis [33], BACH [78], PUIH [148] | 12329 | 70 | 30 | Overall ACC of 94.87%, 91.32%, and 90.45% obtained from BreaKHis [33], BACH [78], and PUIH [148], respectively | 0.99 | 92.21% |
| Wang et al. [27] | To divide breast cancer images | Employed a model of CNN and CapsNet [149] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 92.71%, 94.52%, 94.03%, and 93.54% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 93.70% | |
| Albashish et al. [30] | To analyze VGG16 [98] | Employed a variation of VGGNet [98] | BreaKHis [33] | 7909 | 90 | 10 | ACC of 96%, 95.10%, and 87% obtained for polynomial SVM, Radial Basis SVM, and k-nearest neighbors, respectively | NA | 92.70% | |
| Kundale et al. [150] | To classify breast cancer from histology images | Employed SURF [82], DSIFT [151], linear coding [152] | BreaKHis [33] | 7909 | 70 | 30 | ACC of 93.35%, 93.86%, 93.73%, and 94.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 93.74% | |
| Attallah et al. [153] | To classify breast cancer from histopathological images | Employed several deep learning techniques including autoencoder [139] | BreaKHis [33], ICIAR2018 [78] | 7909 | 70 | 30 | For BreaKHis [33], ACC of 99.03%, 99.53%, 98.08%, and 97.56% got for 40x, 100x, 200x, and 400x magnification factors, respectively; for ICIAR2018 [78], ACC was 97.93% | NA | 98.43% | |
| Burçak et al. [154] | To classify breast cancer histopathological images | Stochastic [155], Nesterov [156], Adaptive [157], RMSprop [158], AdaDelta [159], Adam [160] | BreaKHis [33] | 7909 | 70 | 30 | ACC was taken into account. The overall ACC of 97.00%, 97.00%, 96.00%, and 96.00% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 96.50% | |
| Hirra et al. [161] | To label breast cancer images | Patch-based deep belief network [162] | HASHI [129] | 584 | 52 | 30 | Images from four different data samples achieved an accuracy of 86% | NA | 86.00% | |
| Elmannai et al. [42] | To extract eminent breast cancer image features | A combination of two deep CNNs | BACH [78] | 400 | 60 | 20 | The overall ACC for the subimage classification was 97.29% and for the carcinoma cases the sensitivity achieved was 99.58% | NA | 97.29% | |
| Baker and Abu Qutaish [163] | To segment breast cancer images | Clustering and global thresholding methods | BACH [78] | 400 | 70 | 30 | The maximum ACC obtained from classifiers and neural network using BACH [78] to detect breast cancer | NA | 63.66% | |
| Soumik et al. [60] | To classify breast cancer images | Employed Inception V3 [88] | BreaKHis [33] | 7909 | 80 | 20 | ACC of 99.50%, 98.90%, 98.96% and 98.51% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 98.97% | |
| Brancati et al. [50] | To analyze gigapixel histopathological images | Employed CNN with a compressing path and a learning path | Camelyon16 [164], TUPAC16 [165] | 892 | 68 | 32 | AUC values of 0.698, 0.639, and 0.654 obtained for max-pooling, average pooling, and combined attention maps, respectively | 0.66 | NA | |
| Mahmoud et al. [61] | To classify breast cancer images | Employed transfer learning | Mammography images [166] | 7500 | 80 | 20 | Maximum ACC of 97.80% was claimed by using the given dataset [166]. Sensitivity and specificity were estimated | NA | 94.45% | |
| Munien et al. [62] | To classify breast cancer images | Employed EfficientNet [167] | ICIAR2018 [78] | 400 | 85 | 15 | Overall ACC of 98.33% obtained from ICIAR2018 [78]. Sensitivity was also taken into account | NA | 98.33% | |
| Boumaraf et al. [63] | To analyze breast cancer images | Employed ResNet [75] on ImageNet [168] images | BreaKHis [33] | 7909 | 80 | 20 | ACC of 94.49%, 93.27%, 91.29%, 89.56% obtained for 40x, 100x, 200x, and 400x magnification factors, respectively | NA | 92.15% | |
| Saber et al. [64] | To detect breast cancer | Employed transfer learning technique | MIAS [85] | 322 | 80 | 20 | Overall ACC, PRS, F1S, and AUC of 98.96%, 97.35%, 97.66%, and 0.995, respectively, got from MIAS [85] | 0.995 | 98.96% | |
|
| ||||||||||
| 2022 | Ameh Joseph et al. [169] | To classify breast cancer images | Employed handcrafted features and dense layer | BreaKHis [33] | 7909 | 90 | 10 | ACC of 97.87% for 40x, 97.60% for 100x, 96.10% for 200x, and 96.84% for 400x demonstrated from BreaKHis [33] | NA | 97.08% |
| Reshma et al. [52] | To detect breast cancer | Employed probabilistic transition rules with CNN | BreaKHis [33] | 7909 | 90 | 10 | ACC, PRS, RES, F1S, and GMN of 89.13%, 86.23%, 81.47%, 85.38%, and 85.17% demonstrated from BreaKHis [33] | NA | 89.13% | |
| Huang et al. [53] | To detect nuclei on breast cancer | Employed mask-region-based CNN | H&E images of patients | 537 | 80 | 20 | PRS, RES, and F1S of 91.28%, 87.68%, and 89.44% demonstrated from the used dataset | NA | 95.00% | |
| Chhipa et al. [170] | To learn efficient representations | Employed magnification prior contrastive similarity | BreaKHis [33] | 7909 | 70 | 30 | Maximum mean ACC of 97.04% and 97.81% were got from patient and image levels, respectively using BreaKHis [33] | NA | 97.42% | |
| Zou et al. [171] | To classify breast cancer images | Employed channel attention module with nondimensionality reduction | BreaKHis [33], BACH [78] | 8309 | 90 | 10 | Average ACC, PRS, RES, and F1S of 97.75%, 95.19%, 97.30%, and 96.30% obtained from BreaKHis [33], respectively. ACC of 85% got from BACH [78] | NA | 91.37% | |
| Liu et al. [172] | To classify breast cancer images | Employed autoencoder and Siamese framework | BreaKHis [33] | 7909 | 80 | 20 | Average ACC, PRS, RES, F1S, and RTM of 96.97%, 96.47%, 99.15%, 97.82%, and 335 seconds obtained from BreaKHis [33], respectively | NA | 96.97% | |
| Jayandhi et al. [54] | To diagnose breast cancer | Employed VGG [98] and SVM | MIAS [85] | 322 | 80 | 20 | Maximum ACC of 98.67% obtained from MIAS [85]. Sensitivity and specificity were also calculated | NA | 98.67% | |
| Sharma and Kumar [55] | To classify breast cancer images | Employed Xception [135] and SVM | BreaKHis [33] | 2000 | 75 | 25 | Average ACC, PRS, RES, F1S, and AUC of 95.58%, 95%, 95%, 95%, and 0.98 obtained from BreaKHis [33], respectively | 0.98 | 95.58% | |
| Zerouaoui and Idri [56] | To classify breast cancer images | Employed multilayer perceptron, DenseNet201 [67] | BreaKHis [33] and others | NA | 80 | 20 | ACC of 92.61%, 92%, 93.93%, and 91.73% on four magnification factors of BreaKHis [33] | NA | 93.85% | |
| Soltane et al. [65] | To classify breast cancer images | Employed ResNet [75] | 323 colored lymphoma images | 323 | 50 | 50 | A total of 27 misclassifications for 323 samples were claimed. PRS, RES, F1S, and Kappa score were estimated | NA | 91.6% | |
| Naik et al. [173] | To analyze breast cancer images | Employed random forest, k-nearest neighbors, SVM | 699 whole-slide images | 699 | 80 | 20 | Random forest algorithm achieved better result for classifying benign and malignant images from 190 testing samples | NA | 98.2% | |
| Chattopadhyay et al. [174] | To classify breast cancer images | Employed dense residual dual-shuffle attention network | BreaKHis [33] | 7909 | 80 | 20 | Average ACC, PRS, RES, and F1S of 96.10%, 96.03%, 96.08%, and 96.02%, respectively, obtained from four different magnification levels of BreaKHis [33] | NA | 96.10% | |
| Alruwaili and Gouda [66] | To detect breast cancer | Employed the principle of transfer learning, ResNet [75] | MIAS [85] | 322 | 80 | 20 | Best results for ACC, PRS, RES, F1S, and AUC of 89.5%, 89.5%, 90%, and 89.5% obtained from MIAS [85], respectively | NA | 89.5% | |