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.