|
| Publication | Type of data | Methods | Performance |
|
| [92] | Custom dataset of ultrasound images | Pretrained CNNs | AUC of 0.90 (nonmalignant vs. malignant), AUC of 0.88 (benign vs. malignant) |
| [93] | 1 custom+2 publically available datasets | DL-based approach using a matching layer | (custom dataset), AUCs around 0.89 (publically available datasets) |
| [94] | Custom dataset of ultrasound images | Inception v3, VGG19 | Robust and efficient classification performances |
| [95] | Custom dataset of ultrasound images | Training from scratch, pretrained VGG16, fine-tuning approach | 0.97 accuracy, 0.98 AUC using fine-tuning approach |
| [96] | Custom dataset of breast MRI images | Cross-modal transfer learning approach | Overall accuracy of 0.93 using cross-modal approach |
| [97] | Custom dataset of breast MRI images | DL-based method | High sensitivity in the range of 93-100% |
| [98] | Custom dataset of multiparametric MRI images | Pretrained CNN architectures | |
| [99] | BreakHis, Breast Cancer Classification Challenge 2015 | Inception recurrent residual CNN model | 100% for the binary and multiclass (Breast Cancer Classification Challenge 2015 dataset) |
| [100] | BreakHis | Single-task CNN, multitask CNN | Patient score of 83.72% for binary classification using single-task CNN |
| [101] | 2015 bioimaging breast histology classification challenge, BreakHis dataset | Progressive DL-based models | Recognition rate of 96.4% and 99.5% on multiclass and binary classification tasks on 2015 bioimaging breast histology classification challenge |
| [102] | BreakHis dataset, PatchCamelyon dataset, 2015 Bioimaging challenge dataset, 2018 ICIAR dataset | VGG19, MobileNet, DenseNet | Accuracy of 98.13% on BreakHis dataset |
| [103] | BreakHis dataset | DL and hierarchical classification approach | Accuracy of 95.48% on the multiclass classification task |
| [104] | BreakHis dataset | Integrated DL model | 98.51% classification success on the multiclass classification task |
| [105] | BreakHis dataset | DenseNet and Xception architectures | 99% and 92% accuracy on binary and multiclass classification tasks |
| [106] | BreakHis dataset | DL-based model | Mean recognition rate of for binary classification |
| [107] | BreakHis dataset | DL-based model | Accuracy of for multiclass classification |
| [108] | BreakHis dataset | Bag of words, locality-constrained linear coding, CNNs | For CNN model accuracies between 96.15% and 98.33% for the binary classification and 83.31% and 88.23% for the multiclass classification |
| [109] | BreakHis dataset | Combination of 4 residual networks | Correct classification rate of 96.25% for 8-class categorization |
| [110] | BreakHis dataset | End-to-end model based on FCN and bidirectional LSTM | Accuracy of on binary classification task |
| [111] | BreakHis dataset | ResNet-18, ResNet-50, and AlexNet | Image-level accuracy of 96.88% for binary classification |
| [112] | BreakHis dataset | Weakly supervised learning framework | Classification rate of up to 92.1% for binary classification |
| [38] | 2015 Bioimaging challenge dataset | CNN models | Accuracies of 77.8% for four classes and 83.3% for carcinoma/noncarcinoma were achieved |
| [113] | ICIAR 2018 Grand Challenge | Pretrained ResNet-50, Inception v3, and VGG16 architectures | Accuracies of 87.2% for multiclass, 93.8% for binary classification tasks |
| [114] | 2015 Bioimaging challenge database | Clustering algorithm and ResNet-50 architecture | 88.89% accuracy on the overall test set for multiclass classification |
| [115] | BreakHis dataset | VGG16, VGG19, and ResNet-50 architectures | 92.60% accuracy |
| [116] | Custom dataset | CNN, KNN, Inception v3, SVM, and ANN models | Accuracy of 97% using ANN algorithm for binary classification |
| [117] | Custom dataset | CNN and ANN models | Accuracy of for the binary classification task using a VGG model, accuracy of for the multiclass classification task |
| [118] | BreakHis dataset | VGG16, VGG19, ResNet-50 architectures | Accuracy of 93.25% for multiclass classification task |
| [119] | BreakHis dataset | DL-based model | -score of 90.3 |
| [120] | BreakHis dataset | Deep second-order pooling network | Accuracy of 97.92% for binary classification |
| [121] | BreakHis+custom datasets | Pretrained CNN architectures (GoogLeNet, VGGNet, and ResNet) | Accuracy of 97.67% for binary classification |
| [122] | Custom dataset | Transformer models | Precision of 0.976 for relation recognition |
| [123] | Australian Breast Cancer Tissue Bank, TCGA dataset | Deep neural network | AUC on TCGA of 0.861, AUC on Australian Breast Cancer Tissue Bank was 0.905 |
| [124] | INbreast database | DL models | |
|
