Table 1.

This table summarises the methods and datasets adopted by previous studies on breast cancer grading.

References	Methods	Datasets	Result
17	The multi-resolution method that combined the three NGS evaluation criteria and Gaussian model functions	Own Custom dataset	Quantitative results were not available Grading result was similar to the pathologists’ scores but slightly lower in general
18	Spectral clustering with image textural and architecture features	Own Custom dataset	93.3% accuracy with all architecture features
19	Segmentation method that utilised the combination of low-level, high-level, and domain specific information	Own Custom dataset	80.52% accuracy in automated feature extraction set low vs high grades 93.33% accuracy in manual feature extraction set low vs high grades
20	Multi field-of-view (multi-FOV) classifier	Own Custom dataset	AUC values: 0.93 (low vs high grades), 0.72 (low vs intermediate grades), 0.74 (intermediate vs high grades)
34	Grassmann manifold	BreaKHis and Breast Cancer Grading Dataset	95.8% accuracy (overlapping )patch size 8 × 8 strategy
22	Deep learning with manual feature extraction -Cascaded ensemble method with multi-level image features combination (pixel, object, semantic)	Own Custom dataset	92% (low vs high) 77% (low vs intermediate) 76% (intermediate vs high) 69% (overall)
24	Deep learning with automatic feature extraction -Multi-task deep learning method	BreaKHis and Breast Cancer Grading Dataset	93.33% accuracy in manual feature extraction set low vs high grades
37	Deep learning with automatic feature extraction Nuclei aware network (NaNet) that applies more attention into nuclei related features while learning the whole pathological image feature representation	Breast Cancer Grading Dataset with own custom dataset	92% for overall IDC grading
23	Deep learning with automatic feature extraction Entropy-Based Elastic Ensemble of deep convolutional network (CNN) models (3E-Net) for breast cancer grading	BreaKHis and Breast Cancer Grading Dataset	3E-Net (Version A): 96.15% accuracy 3E-Net (Version b): 99.50%,
26	Transfer learning (feature extraction) using ResNetV1-50 and MobileNetV1	BreaKHis and Breast Cancer Grading Dataset	Four Breast Cancer Grade dataset: 97.03% accuracy (ResNet50), 94.42% accuracy (MobileNet) Three Breast Cancer Grade dataset: 92.39% accuracy (ResNet50), 93.48% accuracy (MobileNet)
27	Transfer learning (feature extraction) using VGG16	Databiox	88% validation accuracy 72% test accuracy