A robust and automated cell counting method in quantification of digital breast cancer immunohistochemistry images.

Abstract

Quantitative analysis ofimmunohistochemically stained breast cancer specimens by cell counting is important for prognosis and treatment planning. This paper presents arobust, accurate, and novel method to label immunopositive and immunonegative cells automatically. During preprocessing, we developed anadaptive method to correct thecolour aberration caused by imaging conditions. Next, apixel-level segmentation was performed on preprocessed images using asupport vector machine with aradial basis function kernel in HSV colour space. Thesegmentation result was processed by mathematical morphology operations to correct error-segmented regions and extract themarker for each cell. Validation studies showed that theautomated cell-counting method had divergences varying from -5.05% to 3.99% compared with manual counting by apathologist, indicating considerable agreement ofthepresent automated cell counting method with manual counting. Thus, this method can free pathologists from laborious work and can potentially improve theaccuracy and thereproducibility ofdiagnosis.