Next generation image analysis for immunohistochemistry quantitation

Transcription

1 Next generation image analysis for immunohistochemistry quantitation Ben Vainer Department of Pathology, Rigshospitalet University of Copenhagen Medical Center

2 Men are only so good as their technical developments allows them to be. George Orwell Visiopharm s User Group Meeting 2015

3 Inter-laboratory variation Eight North American and European laboratories. At a hypothetical 13.5% cutoff, there are 32.3% cases that Laboratory D would call high Ki67 but Laboratory B would call low Ki67. Our data suggest that even if a common Ki67 cutoff is agreed upon, lack of interlaboratory reproducibility in Ki67 measurements represents a major obstacle to confident use of Ki67 for clinical decisions.

4 Intralaboratory variation Same 50 breast cancer cases (two 1 mm-cores from each) assessed on three consequtive days. Numerical differences.

5 Three pathologists, using their own scoring method Interval < 2 months

6 Overall estimate Hot spots Andre metoder Eyeballing Image analysis Manual counting Laboratory variation - Ki67 in breast cancer Hvor tælles? Vyberg & Røge, NordiQC Rigshospitalet, Københavns Universitet

7 Explanations? Counting method Experience Vyberg & Røge, NordiQC

8 How can we improve the assessment?

9 A: Same pathologist, 2 weeks interval B: Digital assessment, 2 weeks interval Visiopharms software

10 Neuroendocrine tumours in the gastrointestinal tract Cutoff-values in neuroendocrine tumours G1: <2 mitoses per HPF and Ki67 2% G2: 2-20 mitoses per HPF and/or Ki % G3: >20 mitoses per HPF and/or Ki67>20% Mitosis count in 50 HPF (2 mm 2 ) Ki67 proliferation index in hot spots ( cells) ENETS Consensus Guidelines. Neuroendocrinology 2012; 95.

11 Digital photos ImmunoRatio software Original proliferation indices (PI): From pathology reports Re-assessment PI: One qualified pathologist

12 22 cases from Memorial Sloan-Kettering Cancer Center Eyeball-estimate (EE) Manual counting (MC) of 2000 cells (!) Digital image analysis (Aperio s QIA-algorithm) Manual selection of hot-spots Manuel exclusion of stained non-tumour cells

13 HER2-CONNECT in breast cancer Breast Cancer Res Treat 2012; 132: % konkordans med FISH

14 The Rigshospitalet study Holten-Rossing H, Talman MLM, Kristensson M, Vainer B: Optimizing HER2 assessment in breast cancer - application of automated image analysis. Breast Cancer Research and Treatment breast cancer samples TMA with automated dearraying Zeiss Axio Scan Z.1 Visiopharm s cloud-based software

15 Results HER2 IHC assessment Image Analysis Classification 0/ Total Manual reading 0/ Total Agreement 90.5% Cohen s kappa HER2 FISH assessment Image Analysis Classification Negative Equivocal Amplification Total Manual Reading Negative Equivocal Amplification Total Agreement 93.6% Cohen s kappa 0.758

16 Results HER2 IHC assessment Classificatio n Manual reading Image Analysis 0/ Total 0/ Total Agreement 90.5% Cohen s kappa HER2 FISH assessment Image Analysis Classification Negative Equivocal Amplification Total Manual Reading Negative Inconclusive Amplification Total Agreement 93.6% Cohen s kappa % reduced need for reflex testing (FISH)

17 Results HER2 IHC assessment Classificatio n Manual reading Image Analysis 0/ Total 0/ Total Agreement 90.5% Cohen s kappa HER2 FISH assessment Image Analysis Classification Negative Equivocal Amplification Total Manual Reading Negative Equivocal Amplification Total Agreement 93.6% Cohen s kappa % reduceret behov for reflex-testing (FISH) When adjusting the build-in cutoff to optimal values: Equivocal: 1.4% Reduced need for FISH: 90%

18 Other applications Ki67 proliferation index in other tumour types PHH3 as a substitute for mitoses count Other diagnostic, prognostic or predictive markers Unambiguous staining / cell marking Translation to clinically relevant information

19 Malignant melanoma

20 HPV-positive orophangyal cancer

21 Using automated image analysis to prognosticate HPV+ oropharyngeal carcinomas Virtual double staining with serial sections of CK5 and marker of interest (Visiopharm) Ki67 PHH3 PD-L1 (T-cell modulator) Hani Channir, Rigshospitalet, UCPH Ki67 CK5

22 Challenges Visiopharm s User Group Meeting 2015

23 Selection of invasive areas in breast cancer automated elimination of DCIS Torben Steiniche, Aarhus

24 Virtual double staining

25 Alignment and analysis Ki67 CK Analysis: Identification of positive/negative cells. Exclusion of CKnegative cells. Michael Grunkin, Visiopharm

26 Example: Ki67 in breast cancer Ki67 Visiopharm (Novartis Cambridge)

27 Automated detection of hot spots (Visiopharm) Heatmaps 8/20 = 40% Positive cells 4/4 = 100% Visually more dense! Negative cells Thomas Ebstrup, Visiopharm

28 Digital hot spot-detektion example (Visiopharm) Ki67 CK Negative heatmap Positive heatmap Index-heatmap

29

30 Next generation immunoassessment - Why? Minimal inter- and intraobserver variation Same result every time Likelyhood for a given treatment is identical for all patients Increased patient safety Correct and exact algorithms Stable laboratory work-up Possibly adjustment of treatment protocols Integration with existing reporting systems and databases

31 When the winds of change blows, some people build walls, while others build windmills. Chinese proverb