175P - Utility of artificial intelligence (AI) in Ki67 scoring of a breast cancer (BC) patient population

Background

Ki67 is an important BC marker, especially for adjuvant treatment in HR+, HER2- cases. Working groups have provided guidance for Ki67 immunohistochemistry (IHC) BC scoring to limit pathologist’s variability, but no scoring method has been universally accepted. Rapid and reliable image analysis solutions to support scoring have surfaced for the Ki67 assessment. We compared Ki67 scoring with Aiforia® Platform (AI deep learning image analysis), Halo® (image analysis supervised software) and 2 independent pathologists (patho) in a BC population.

Methods

We stained 114 BC tumors for Ki67 on the Dako Omnis. Three methodologies were used to quantify ki67+ tumor cells: 1) A deep learning approach model was trained for BC and the Ki67 clone by Aiforia; 2) Two pathos (Patho A and Patho B) were trained following the International Ki67 Working Group guidelines. Intra-analysis assessment was done for one patho; 3) The random forest classifier from Halo was used to separate the image into tumor, non-tumor and background with patho approval. After cell segmentation, Ki67 positivity was assessed by thresholding. The time needed to complete the analyses was recorded for each method.

Results

Intra-pathologist analysis showed a very high reproducibility (r²=0.95) while matched pair analysis between two patho was lower (r²=0.86) despite following guidelines. Our study shows a high consistency of Ki67 results between AI and the other methods (patho A-AI, r²=0.92; B-AI, r²=0.90; Halo-AI, r²=0.93). The correlation obtained between Halo scoring was not as good, but within an acceptable range (Halo-A, r²=0.79, Halo-B, r²=0.84). The deep learning AI approach was the quickest even including the model training (total time: 2.5 hrs). Pathos time ranged from 22 to 28 hrs without a major gain in analysis time in the second review. Halo took 28 hours including application development, pathologist verification, and analysis.

Conclusions

Overall, the ki67 tumor analysis approaches were quite comparable. AI-based image analysis tools offer valuable assistance in Ki67 scoring and could reduce inter-pathologist variability. These results demonstrate the time benefit of using an AI-driven method for Ki67 analysis in breast cancer.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.