An image analysis-based approach for automated counting of cancer cell nuclei in tissue sections

Semiquantitative evaluation and manual cell counting are the commonly used procedures to assess positive staining of molecular markers in tissue sections. Manual counting is also a laborious task in which consistent objectivity is difficult to achieve. Recently, image analysis has been explored, but the studies reported were limited to histological images acquired at high magnification and containing uniformly stained cells.

The analyzed material consisted of histological sections from different squamous cell cancers that had stained for proliferation using Ki-67 and cyclin A detection. The first step of the method was based on detecting the overall number of cells irrespective to their stain, using second-order edge detection methodology. Then proliferating cells were located using principal component analysis (PCA) of the color image, combined with histogram thresholding.

The algorithms' performances were validated on tissue section images encountered in routine clinical practice by comparison with objective measures of performance and manual cell identification. The algorithms correlated closely with manual counting of all cells (r(2) = 0.96-0.97) and stained cells (4-7% cell count error).

Cell counting in complex large-scale histological images could be applied in routine practice using edge and color information. The proposed technique provides several benefits, such as speed of analysis, consistency, and automation. Moreover, it is faster than human observation and could replace the laborious task of manual cell counting.