Improvement of mitoses counting on whole slide images of breast cancer using artificial intelligence

Abstract

Introduction: Breast cancer (BC) prognosis is largely influenced by histopathological grade, assessed according to the Nottingham modification of Bloom-Richardson (BR). Mitotic count (MC) is a component of histopathological grading but is prone to subjectivity. This study investigated whether mitoses counting in BC using digital whole slide imaging (WSI) compares better to light microscopy (LM) when assisted by artificial intelligence (AI), and to which extent differences in digital MC (AI assisted or not) result in BR grade variations. Methods: Fifty BC patients with paired core biopsies and resections were randomly selected. Component scores for BR grade were abstracted from pathology reports. MC was assessed using LM, WSI and AI. Different modalities (LM-MC, WSI-MC and AI-MC) were analyzed for correlation with scatterplots and linear regression, and for agreement in BR with Cohen’s Kappa. Results: MC modalities strongly correlated in both biopsies and resections: LM-MC and WSI-MC (R2 0.737 and 0.773, respectively), LM-MC and AI-MC (R2 0.545 and 0.706) and WSI-MC and AI-MC (R2 0.692 and 0.760). Agreement in BR between modalities was high in both biopsies and resections: LM-MC and WSI-MC (kappa 0.93 and 0.834, respectively), LM-MC and AI-MC (kappa 0.89 and 0.825), and WSI-MC and AI-MC (kappa 0.96 and 0.732). Conclusion: MC in WSI did not compare better to LM-MC when supported by AI. However, LM-MC and WSI-MC were already strongly correlated, so the expected gain from AI was inherently low. Agreement between different modalities for BR was high. WSI-MC appears as a viable alternative to LM-MC.