Table 1.

Summary of AI-based diagnostic algorithms in breast cancer pathology

Year of publication	Reference	Number of slides for training / validation	Pathologists review (training/validation)	Algorithm details	Algorithm endpoints/outputs	Algorithm performance
2017	Yamamoto et al [14]	11661 myoepithelial cells in 22 cases	Three pathologists	Staining > Ilastik > CellProfiler > support vector machines (SVM)	Types of breast tumors, Myoepithelial cells morphology and precise nuclear features	Accuracy 90.9%
2018	Steiner et al [15]	Training: 60-80 Validation:70	Six pathologists	LYNA, inception V3	lymph nodes metastasis detection	Sensitivity (91% vs. 83%, P = 0.02)
2018	Cruz-Roa et a [11]	Training: 349 Validation: 52 Testing: 195	Three expert pathologists	HASHI (High-throughput adaptive sampling for whole-slide histopathology image analysis)	invasive breast cancer detection	Dice coefficient of 76%
2018	Fondón et al [16]	Training: 30 Validation: 70 Testing: 150 + images with artefacts included	Pathologists	SVM (Support Vector Machine) classifier with a quadratic kernel	Breast carcinoma classification on biopsies	accuracy levels ranging from 61.11% to 75.8%
2022	El Agouri et al [17]	328 digital slides from 116 surgical breast specimens	One pathologist, two qualified consultant breast pathologists	CNN, (Resnet50 and Xception)	Breast cancer detection/ diagnosis	accuracy (88%), and sensitivity (95%)
2023	Wang et al [4]	400 WSIs Training: 270 Test: 129	N/A	dual magnification mining network (Two stream network) (SL-Net and PH-Net)	lymph nodes metastasis localization	0.871 FROC score with dual magnification mining network and 0.88 FROC score with high magnification network
2023	Challa et al [18]	a validation cohort with 234 SLNs and a consensus cohort with 102 SLNs)	Three pathologists	Visiopharm Integrator System (VIS) metastasis AI algorithm	Diagnosis of lymph node metastasis	sensitivity of 100%