Segmentation and Border Identification of Cells in Images of Peripheral Blood Smear Slides

We present an unsupervised blood cell segmentation algorithm for images taken from peripheral blood smear slides. Unlike prior algorithms the method is fast; fully automated; finds all objects—cells, cell groups and cell fragments—that do not intersect the image border; identifies the points interior to each object; finds an accurate one pixel wide border for each object; separates objects that just touch; and has been shown to work with a wide selection of red blood cell morphologies. The full algorithm was tested on two sets of images. In the first set of 47 images, 97.3% of the 2962 image objects were correctly segmented. The second test set—51 images from a different source—contained 5417 objects for which the success rate was 99.0%. The time taken for processing a 2272x1704 image ranged from 4.86 to 11.02 seconds on a Pentium 4, 2.4 GHz machine, depending on the number of objects in the image.