Table 1.
Current applications of artificial intelligence in breast pathology.
| Applications | Comments |
|---|---|
| Diagnostic applications | |
| Tumor detection | |
| Primary tumor detection: | AI-based algorithms have been developed to detect malignant tumours in the breast and to differentiate it from benign and normal structures. Osareh et al. introduced ML techniques to differentiate between malignant and benign tumours using digitalized images of fine-needle aspiration biopsy samples [73]. Also, algorithms have been developed to provide quantitative measurements of nuclear shape and size, which could be applied across different tumour subtypes [54]. |
| Metastatic deposits detection in lymph nodes: | One of the most important application is detection of metastatic tumour deposits in the lymph nodes. Babak et al., 2017 detected lymph node metastasis in breast cancer patients with a higher diagnostic achievement over 11 pathologists [33] |
| Breast cancer grading | Several algorithms have been developed to assess breast cancer grade. Coutre et al. [50] have used image analysis with DL to predict breast cancer grade. Other algorithms were developed to allow objective enumeration of mitotic figures [26], measurements of nuclear shape and size, and with the automatic detection and segmentation of cell nuclei in histopathology images [74]. |
| Breast cancer subtype | Breast cancer comprises more than 20 histotypes. Coutre et al., used image analysis with DL to detect breast cancer histologic subtypes [50]. |
| Assessment of tumour heterogeneity and tumour microenvironment | AI-based assays to measure tumour intra-tumour and inter-tumour heterogeneity [26,56], identify and quantify non-epithelial cells such as fibroblast, neutrophils, lymphocytes and macrophages [77] and computerized image-based detection and grading of tumour infiltrating lymphocytic (TILs) in HER2+ breast cancer [78] have been developed |
| Receptor status and intrinsic subtype assessment | AI algorithms have been developed to provide quantitative measurements of immunohistochemically stained Ki-67 [52], ER [50], PR and Her2neu images [75]. Xu et al. proposed a novel GAN-based approach to provide a virtual immunohistochemistry staining pattern from the H&E stained WSIs that potentially obviates the need for IHC-based tissue testing [52,76] Coutre et al., used image analysis with DL to predict breast cancer intrinsic subtype [50]. |
| Prognostic Applications | |
| Prognostic significance of tumour morphological features | Morphological features as nuclear shape, texture and architecture can predict risk of recurrence and overall survival. Whitney et al., [54] showed that quantitative features of nuclear shape, texture and architecture independently enable prediction risk of recurrence in patients with ER-positive breast tumours |
| Prognostic significance of different peri-tumoral elements | AI-based assays to measure the arrangement and architecture of different tissue elements such as TILs within the tumour have been developed and demonstrated their value in predicting survival [79] and that the spatial distribution of TILs among tumour cells expression profiling is associated with late recurrence in ER-positive breast cancer [57]. |
| Applications related to predictive values and response to treatment. | ML approached can be used to correlate the expression of certain markers such as cell cycle and proliferation markers [80] or the presence of certain morphological features in the tumour to the response of specific therapy. |