Abstract 48P
Background
Assessment of immunohistochemical (IHC) Ki-67 expression plays a crucial role in breast cancer diagnostics for many therapy decisions. The International Ki-67 in Breast Cancer Working Group (IKWG) has recently proposed a new global scoring method on whole tissue sections to improve accuracy, involving assessment of up to four representative tumor areas with negligible, low, medium and strong proliferation levels. As precise manual assessment is time-consuming, artificial intelligence (AI) support is necessary to make this method practicable for clinical routine.
Methods
We developed an AI software for automated Ki-67 scoring according to the IKWG protocol. Seven pathologists assessed Ki-67 IHC status in 2 x 214 readings of whole-slide images (WSI) derived from 72 breast cancer specimens, four scanning hardware types and two clones. Pathologists selected representative areas manually, scored them with assistance of a CE-cleared AI software for region-of-interest (ROI) analysis and concluded global scores (weighted/unweighted). After a 2-week washout period, the same pathologists were presented with the same slides together with results by a fully automatic AI (AI-only): 1) selection of regions, 2) score and weight per region, and 3) two global scores (weighted/unweighted). Being able to adjust the AI-suggested results, pathologists concluded the two global scores (AI+path).
Results
For weighted global scoring, interrater-agreements between a) pathologist and AI+path, b) pathologist and AI-only, and c) AI-only and AI+path were 93.9%, 92.5%, and 96.7%, respectively (for unweighted scores, 91.1%/92.1%/96.3%). With AI-assistance, pathologists scored significantly faster compared to manual scoring (median time per WSI: 229 vs. 306 sec).
Conclusions
Used as a diagnostics assistance tool, the presented AI system showed high agreement with pathologist scores, while reducing assessment time. Likewise, also fully automatic AI use showed high agreement with human scoring. Overall, this demonstrates that the investigated AI system is suitable to enable safe implementation of the global scoring protocol in clinical routine, ultimately leading to higher accuracy and reproducibility in Ki-67 IHC scoring.
Legal entity responsible for the study
Mindpeak GmbH.
Funding
Mindpeak GmbH.
Disclosure
R. Erber: Financial Interests, Personal, Invited Speaker: AstraZeneca, Novartis, Daiichi Sankyo , Roche, Pfizer, Eisai; Financial Interests, Personal, Expert Testimony: Mindpeak GmbH; Financial Interests, Personal, Other: Mindpeak GmbH, Diaceutics, Veracyte; Financial Interests, Personal, Research Grant: Gilead; Financial Interests, Institutional, Funding: Cepheid, Roche, Biocartis; Financial Interests, Personal and Institutional, Funding: NanoString Technologies; Financial Interests, Personal and Institutional, Other: BioNTech. P. Frey, K. Daifalla, S. Springenberg: Financial Interests, Personal, Full or part-time Employment: Mindpeak GmbH. N. Abele: Financial Interests, Personal, Other: Mindpeak GmbH; Financial Interests, Personal, Advisory Board: Mindpeak GmbH. M. Päpper, T. Lang: Financial Interests, Personal, Full or part-time Employment: Mindpeak GmbH; Financial Interests, Personal, Stocks/Shares: Mindpeak GmbH. A. Hartmann: Financial Interests, Institutional, Funding: AstraZeneca, Roche, Janssen-Cilag, NanoString Technologies, Biocartis, ZytoVision, Novartis, Cepheid, Mindpeak, Gilead, palleos healthcare, BioNTech; Financial Interests, Personal, Advisory Board: BMS, MSD, Roche, Cepheid; Financial Interests, Personal, Invited Speaker: Qiagen, Agilent, Diaceutics, Lilly, AstraZeneca, Boehringer Ingelheim, AbbVie, Janssen-Cilag, Pfizer, Ipsen. All other authors have declared no conflicts of interest.