Potential Pitfalls in Diagnostic Digital Image Analysis: Experience with Ki-67 and PHH3 in Gastrointestinal Neuroendocrine Tumors

Abstract

Gastrointestinal neuroendocrine tumors, or GI-NETs are a highly diverse group of tumors derived from neuroendocrine cells of the GI tract. In GI-NET, a spectrum of histological and molecular parameters exists to predict prognosis and survival. Immunohistochemistry for Ki67, a nuclear antigen that is present in all but the G0 phase of the cell cycle with specificity for proliferating cells, can be used to determine a tumors proliferation index. With this in mind, grading of gastrointestinal neuroendocrine tumors is critical for prognosis and can impact clinical decision making. Recently, digital image analysis (DIA) has been shown in studies to be a superior and less time-consuming alternative to the manual scoring of Ki-67 in breast cancer, secondary to its theoretical diagnostic reproducibility. In DIA, the correct identification of tumor cells and non-tumor is paramount to avoid over or under calculation of biomarker expression. Additionally, DIA requires a pathologist to manually outline a tumor in large tissue areas of hematoxylin and eosin (H&E) sections, which is impractical. The findings in our study showed that ventana virtuoso software computer analyzed Ki-67 only correlated well with Neuroendocrine carcinomas while manual analysis of mitotic index and Ki67 were found to be gold standard. The performance of DIA in our study was plagued by software issues. In future, the advent of new digital imaging technologies such as virtual dual staining will hopefully improve diagnostic accuracy and reproducibility across different DIA platforms. Ultimately, determination of therapeutic strategies should be guided by an amalgamation of clinicopathologic characteristics not limited to mitotic index and Ki-67. As well, A visual check of the results should always be performed and correlated with other findings.