Abstract
BACKGROUND: Breast cancer is by far the most frequent cancer among women. The proliferative index, Ki-67, is more and more taken into consideration for treatment decisions. However, the reliability of the established Ki-67 scoring is limited. Digital pathology is currently suggested to be a potential solution to Ki 67 assessment problems. METHODS: This is a retrospective and prospective study including 100 patients diagnosed with invasive breast cancer. Three senior pathologists have been asked to estimate the Ki-67 proliferative index for each of the 100 cases by examining the whole glass slides on optical microscope and providing a continuous score then a categorical score ('high' and 'low' Ki 67 index) using once 14%, once 20% as threshold indicative of high Ki67 status. Finally, a digital quantitative assessment of Ki67 was performed. RESULTS: A high inter-observer agreement was found when using optical microscopy for Ki 67 assessment, with correlation coefficient (CC) estimated at 0.878 (p value < 0.01). The overall agreement between manual and automated evaluation of Ki 67 was only substantial (CC estimated at 0.745 (p value < 0.01)). When using categorical scores, the inter-observers concordance was substantial using both cutoff points with kappa value estimated at 0.796 ([0.696-0.925] while using 14% as a cut off point and at 0.766 ([0.672-0.938] while using 20% as a cutoff point (p value < 0). The inter-observers agreement was better while using 14% as cutoff point. Agreement between manual and automated assessment of Ki 67 indices using both cutoff points was only substantial (Kappa estimated at 0.623, p value < 0.01). In comparison to automated assessment of Ki 67 index, while using 14% as a cutoff point, the overall tendency of all observers was to overestimate the Ki 67 values but to underestimate the proliferation index while using 20% as a cutoff point. CONCLUSION: Automated assessment of Ki 67 value would appear to be comparable to visual Ki 67 assessment on optical microscopy. Such study would help define the role of digital pathology as a potential easy-to use tool for a robust and standardized fully automated Ki 67 scoring.