Table 2.
Summary of AI-based predictive 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 |
|---|---|---|---|---|---|---|
| 2020 | Li et al [13] | 153 invasive breast carcinomas | pathologists and molecular pathologists | Visiopharm HER2-CONNECT APP | Pathological NAC response prediction | HER2 DIA connectivity has the strongest association to predict PCR |
| 2021 | Bodén et al [19] | 200 analyzed areas containing 200 tumor cells | three experienced breast pathologists | Human in the loop + DIA | Ki 67 proliferatio assessment | visual estimation (eyeballing) performed significantly worse than DIA alone and DIA with human-in-the-loop corrections (P < 0.05) |
| 2022 | Shafi et al [20] | 97 invasive breast carcinomas, 73 biopsies, 24 resections | Two Pathologists | Visiopharm automated ER (DIA) algorithm | Estrogen receptor IHC assessment | Concordance (91/97, 93.8%) |
| 2023 | Abele et al [10] | 204 slides | 10 participant pathologist form 8 sites | Mindpeak Breast Ki-67 RoI and Mindpeak ER/PR RoI | quantifying Ki-67, estrogen receptor (ER), and progesterone receptor (PR) in breast cancer |
Agreement rates: 95.8% of Ki-67 cases and 93.2% of ER/PR cases Krippendorff’s α, 0.72 |
| 2023 | Shen et al [9] |
Training:207 Test: 103 |
pathologists | CNN analysis using the ResNeXt model, SVM and RF analysis, and t-SNE analysis | NAC response | 95.15% accuracy |
| 2023 | Huang et al [21] | 62 HER2-positive breast cancer (HER2 +) and 64 triple-negative breast cancer (TNBC) | two pathologists | deep neural network (DeepLabV3) | NAC response | HER2 + AUC = 0.8975; TNBC AUC = 0.7674 |
| 2023 | Aswolinskiy et al [22] |
Training: 721 patients Validation: 126 patients |
Two pathologists,research assistants | Mitosis-Detection CNN & Segmentation CNN | NAC response | AUC between 0.66 and 0.88 |
| 2023 | Saednia [23] | training:144 patients with 9430 annotated tumor beds validation 63 patients with 3574 annotated tumor beds | Board-certified breast pathologists | CoAtNet & ViT models | NAC response | AUCs of 0.79, 0.81, and 0.84 and F1-scores of 86%, 87%, and 89%, respectively |