Nucleus shape recognition for an automated hematology analyzing system

An automated hematology analyzing system was developed in our lab, intended to replace large, expensive equipments in the most common clinical laboratory test - complete blood cell count (CBC). Based on segmented nucleus images, the nucleus shape recognition plays an essential role in the system for the purpose of cell type differentiation and immature cell classification. A neural-net based shape recognition algorithm is used to classify the cells based on the contour radius and curvature features. Promising classification results have been achieved. This result is also compared with the result of another shape recognition algorithm that adopts the most commonly used shape descriptor, Fourier descriptors, as the features.