Using MitPro, pathologists agreed more closely on dividing-cell counts and spent about 55% less time per assessment than they did without AI, according to a retrospective study of 385 digitized slides. The AI software suggests regions where cells are likely dividing and flags candidate mitotic figures, but pathologists choose the region and confirm the count. The team posted its preprint on Oct. 1.
Mitotic figures are cells seen in the process of dividing. Their count is one input to grading several tumor types, but the cells can be sparse and some other structures can look similar. MitPro scans a whole-slide image to suggest regions with likely mitotic activity, then highlights candidate cells inside the selected area.
The retrospective study covered 385 digitized slides from three clinical centers in Lithuania, the United Kingdom and Spain. Thirteen pathologists took part; three readers assessed each slide without AI and again with MitPro after at least a two-week break. The sample included seven tumor types and three scanner models.
Agreement among readers on the counts rose from an intraclass correlation coefficient of 0.589 to 0.949. The coefficient measures how consistently readers' counts align, not whether those counts are correct. Assessment time fell by an average of 151.8 seconds per read, or about 55%, according to the study.
These results support a faster, more repeatable counting workflow, but do not establish greater accuracy: the study measured reproducibility and efficiency without comparing clinical-slide counts with independently established reference counts. The authors say the upward shift in counts and scores cannot be treated as evidence of greater accuracy. A future test against independent reference counts in routine work could show whether the gains hold and how score changes affect grading.
Histofy developed MitPro and partly funded the study. The paper discloses that three authors hold shares in the company, two of them directors.
Sources
- Simon Graham et al., “AI-assisted mitotic counting improves reproducibility and efficiency across multiple tumour types,” arXiv:2610.01813 (posted Oct. 1, 2026).
- Image: “Invasive ductal carcinoma,” by Endolysosome via Wikimedia Commons, CC BY-SA 4.0. Illustrative image, not from the MitPro study; unmodified.
