Nuclear boundary detection algorithm based on a minimax derivative statistic for atypical bronchial squamous...

The thresholding approach in scene segmentation of digitized cell images was found to be unreliable for use with squamous epithelial cells characterized by chromatin condensation within the nucleus. It was observed that the nuclear boundaries of such cells were visually distinguishable. An algorithm was developed for detecting nuclear boundaries in digitized cell images. This algorithm is based on a minimax derivative statistic that has maximum values at nuclear boundaries and low values elsewhere. The statistical properties of the confusion matrix of cellular scene segmentation are outlined. A measurement and a test statistic for scene segmentation errors based on those properties are presented.