Tumour proportion scoring of programmed death-ligand 1 positive cells using digital image analysis

H. Thirstrup, A. Schønau, M. Vyberg
Visiopharm A/S, Hoersholm, Denmark

Objective: To investigate the diagnostic capability of digital image analysis for assessing programmed death-ligand 1 (PD-L1) protein expression in histological specimens of non-small cell lung cancer (NSCLC).
Method: Digital images of histological specimens stained with PD-L1 IHC 22C3 pharmDx were imported into a database and tumour regions-of-interest (ROIs) were outlined. PD-L1 positive membranes and cell nuclei were identified using a series of polynomial filtering methods. The total number of tumour cells were quantified and a Tumour Proportion Score (TPS) was calculated from the number of PD-L1 positive and negative tumour cells.
Results: The PD-L1 image analysis application consistently identified negative tumour cells and cells showing complete or partial membrane staining at any intensity on samples ranging from 0 to at least 50 % TPS assessment by pathologists. Out of 10 tissue microarray cores stained by 6 different laboratories, 83.33 % of the cases were in agreement with manual scoring. One core manually scored as negative was consistently found to have more than 1 % PD-L1 staining, causing 10 % of the errors.
Conclusion: Quantitative digital pathology was shown to accurately quantify and score PD-L1 stained tumour cells in NSCLC tissue samples. The automation and large-scale analysis potential of the application could become a powerful tool for pathologists