Validation of a Digital Image Analysis Method to Quantify Immunohistochemistry Cytoplasmic Staining in Pituitary Tumors

Abstract

The visual assessment of cytoplasmic staining by immunohistochemistry (IHC) from pathologists shows high variability, is subjective, time-consuming, and provides qualitative or semi-quantitative data. Additionally, pituitary tumor samples are challenging due to their small size, the presence of blood clots, and tissue damage resulting from transsphenoidal surgical extraction, making it difficult to identify individual cells, which is necessary for visual methods. Digital analysis methods based on detecting chromogen pixel intensities improve the detection of weak staining and produce continuous data. The objective was to validate a digital image analysis method to quantify cytoplasmic staining by IHC in pituitary tumors.

Micrographs of pituitary tumor tissue sections, with IHC for FGFR1 using DAB as a chromogen (cytoplasmic staining), were analyzed. The criteria for evaluating IHC staining were established in collaboration with a pathologist, who performed three visual assessments at two-week intervals, classifying biomarker expression as null, weak, moderate, or high. Digital quantification was performed independently by two observers using ImageJ (IHC-Toolbox plugin), generating a file with a histogram with positive pixels. To evaluate the intra-observer variability of the visual method and the inter-observer variability of the digital method, the Kappa-Fleiss test (κ) and the Intraclass Correlation Coefficient (ICC) were carried out, respectively. ANOVA-Tukey was used to verify the capacity of the digital method to distinguish between the categories of the visual method. To establish ranges for the digital method, cutoff points were defined through exploratory analysis and validated with the Youden index (J).

The intra-observer agreement of the visual method was moderate/high (null expression, κ=0.96; weak expression, κ=0.76; intermediate expression, κ=0.65; high expression, κ=0.87). The digital analysis showed good inter-observer agreement (ICC=0.95). The digital method was able to differentiate between the qualitative categories (p<0.05), establishing biomarker expression ranges as null ([0-20), J=0.98); weak ([20-60), J=0.80); moderate ([60-90), J=0.72); and strong (≥90, J=0.76).

Digital quantification of IHC cytoplasmic staining could be a rapid, accurate, and reproducible alternative method. It correlates with the pathologist categorization and allows the conversion of qualitative data into quantitative data.