Validation of computer-assisted, pixel-based analysis of multiple-colour immunofluorescence histology

Developments in immunohistology allow the routine simultaneous use on tissue sections of three monoclonal antibodies, tagged with different fluorochromes. Such staining can identify seven different cell populations and the limiting factor is rapid, reliable and reproducible analysis. Future reliance on computer-assisted analysis of digitised images depends on validation against manual counting, often viewed as the 'gold standard'. In this study images were digitised from sections of normal porcine skin, inflamed skin and tonsil, simultaneously stained with three monoclonal antibodies. Combinations of staining were quantified by four manual counts and by pixel-based area measurement. On individual images, the correlation between automated and manual measurements was poor. Despite this, the concordance between manual and automated measurements in the means and variances of tissues was good, and both techniques identified the same changes in inflamed versus normal tissues. In addition, pixel-based counting permitted statistical analysis of co-localisation of cell types in tissue sections. We conclude that automated counting is acceptable for the assessment of tissues, is faster and provides less opportunity for observer variation than manual counting. We also demonstrate that the technique is applicable where more than three fluorochromes are used such that manual counting becomes essentially impossible.