[Assessment of PD-L1 expression using the neural network analysis algorithm in non-small cell lung carcinoma biopsy specimens].

Neural network analysis of digital copies of histological micropreparations is one of the methods used to standardize quantitative continuous data. PD-L1 (22C3) biomarker expression in metastatic non-small cell lung carcinomas without mutations in the EGFR, ALK, and ROS1 genes serves as an indication for the use of pembrolizumab for the first-line therapy.
Objective: To quantify PD-L1 biomarker expression in non-small cell lung carcinomas using the neural network analysis of digital copies of histological micropreparations.
Material and Methods: Immunohistochemical study of PD-L1 (22C3) expression was performed on 96 non-small cell lung carcinoma biopsy specimens. The digital copies of histological micropreparations were processed by the QuPath software neural network analysis module.
Results: The neural network analysis module segmented tumor, stroma, and artifacts in the micropreparations, showing a sufficient level of agreement with a visual assessment. Digital image analysis quantified stained tumor cells in the high PD-L1 expression group and showed 96% agreement rate versus visual assessment. However, the group of tumors without PD-L1 expression versus visual assessment showed a low (58%) agreement rate.
Conclusion: The neural network analysis algorithm is applicable to the study of digital copies of histological micropreparations containing tumor, stroma, and artifacts. The algorithm allows for quantitative immunohistochemical assessment of PD-L1 expression in tumor cells. The algorithm can quantify the immunohistochemically detected expression of PD-L1 in tumor cells.