ABSTRACT The aim of this study is to propose the cancer grading metric of MAR (Mitosis Area Rate) with AI based histopathological image assessment. Proposed MAR metric is defined as the ratio of the mitosis pixels area to the total image area in the histopathological images. The mitosis pixels area is defined as total pixels area of all mitoses in the patches of histopathological image by CNN that distinguishes true mitosis from false positives. The total image area is defined as the total pixels area of histopathological image. HSV color‐space segmentation in pixel patches of annotated H&E‐stained histopathological images from ICPR12, MiDeSeC, and MIDOG21 datasets is used for the calculation of the mitosis pixels area. MAR is used to categorize mitosis activity into low, moderate, or high levels. The existing Ki‐67 index is used to validate the MAR metric. Ki‐67 is defined as the ratio of the number of mitoses to the total number of cells. The proposed AI‐based MAR metric improves accuracy and consistency in assessing tumor proliferation in histopathological images. Performance was evaluated using ROC analysis and Cohen's kappa. Obtained results showed that proposed MAR metric gives a better assessment than state of art. The proposed metric of MAR aligns well with the Ki‐67 index and improves diagnostic consistency. MAR is the first metric in the literature that uses mitosis pixels area as a quantitative approach in automated histopathology.
Nemati et al. (Fri,) studied this question.