Epithelial atypia refers to cells' deviation from their normal morphology. In breast cancer (BC), atypia is assessed by multiple nuclear changes. These changes include nuclear size and shape, pleomorphism, nuclear membrane irregularities, chromatin pattern, and nucleolar prominence. To determine atypical cells, they are compared to normal epithelial cells in the terminal duct lobular units, lymphocytes, and red blood cells that serve as reference cells.
Atypia, which is interchangeably but less accurately referred to as pleomorphism, is a crucial element of the Nottingham grading system, the most potent prognostic indicator in BC. However, atypia is subject to individual interpretation. The assessment of its features is relatively subjective, and the relative importance of each component remains undefined. A significant discordance in assessing atypia has been reported, which could potentially affect the accuracy of overall tumour grading, clinical risk assessment, adjuvant therapies and surgical management. This pressing issue underscores the critical need for objective tools like artificial intelligence (AI) in atypia assessment.
AI can be developed to automatically measure certain cell characteristics, revolutionizing the diagnosis and categorisation of atypia. This advancement could significantly improve diagnostic accuracy, precision, and efficiency. As an example, AI-powered automated cell detection has the ability to generate precise cell measurements such as size, shape, and staining density, thereby providing efficient and accurate results. These parameters are instrumental in distinguishing between benign and atypical lesions and determining the classification of both in-situ and invasive BC grades. Additionally, AI facilitates the establishment of diagnostic criteria for poorly defined tumours like pleomorphic lobular carcinoma. Another remarkable development in AI application is mitosis scoring, a critical aspect of histological grade that is typically time-consuming and often results in weak concordance between pathologists.
Applying AI to assess atypia and tumour grading is a promising tool for creating clear, standardised, and well-defined diagnostic guidelines.
 

Immune checkpoint blockade (ICB) has improved treatment success in triple negative breast cancer (TNBC), but it is still unclear as to which tumours will respond to treatment. As ICB blocks specific cell-cell interactions, we investigated whether the arrangement of cells in the TME could predict ICB response and explored how ICB remodels the tumour microenvironment during the course of treatment. We used imaging mass cytometry to profile the in situ expression of 43 proteins in tumours from patients in a randomised trial of neoadjuvant ICB, sampled at three timepoints (baseline biopsy, n = 243; early on-treatment biopsy, n = 207; post-treatment surgical excision, n = 210). At baseline, proliferating CD8+TCF1+T cells and MHCII+ cancer cells, and cancer–immune interactions with B cells and granzyme B+ T cells were significant predictors of combined chemotherapy-immunotherapy response. Immunotherapy resistance early on-treatment was characterised by CD15+ expression on cancer cells. Response was best predicted by combining tissue features before and on-treatment, showing a potential role for early biopsies in guiding adaptive therapy.