Computer Aided Semi-Automated Evaluation of HER2 Immunodetection—A Robust Solution for Supporting the Accuracy of Anti HER2 Therapy

HER2-positive breast cancers usually benefit from anti-HER2 therapy, thus, HER2 evaluation became inevitable for patient selection. HER2-negative (IHC 0, 1+) and strong positive (IHC 3+) cases can easily be interpreted with immunohistochemistry, but equivocal (IHC 2+) cases require further analysis of HER2 gene amplification using in situ hybridization. Our study aimed to validate digital pathology and automated image analysis for unbiased evaluation of HER2 immunostains. We developed an image segmentation algorithm for analyzing HER2-immunostaining (4B5 clone) in tissue microarrays of breast cancers. Two pathologists assessed 309 microscopic regions of at least 100 tumor cells each—representing all HER2 positivity groups—according to international guidelines either semi-quantitatively or by using the MembraneQuant software. Scoring results were statistically correlated with each other and with FISH data, and almost perfect agreement was found (inter-method Cohen’s kappa = 0.872, Spearman-rho = 0.928). When clinical relevance (scoring disagreement that may define erroneous treatment selection) was examined high agreement was found (quadratic weighted kappa = 0.967). Image analysis classified cases with excellent correlation with visual evaluation, therefore, MembraneQuant software proved to be a reliable tool for assessing HER2 immunoreactions and supporting better targeting anti-HER2 therapy. As digital analysis of immunomorphological markers allows permanent archiving, standardization and accurate reviewing of results, it supports quality assurance initiatives in diagnostic pathology—especially of equivocal cases which are hard to interpret.