Recognition and classification of red blood cells using digital holographic microscopy and data clustering with discriminant analysis.

We propose to apply statistical clustering algorithms on a three-dimensional profile of red blood cells (RBCs) obtained through digital holographic microscopy (DHM). We show that two classes of RBCs stored for 14 and 38 days can be effectively classified. Two-dimensional intensity images of these cells are virtually the same. DHM allows for measurement of the RBCs' biconcave profile, resulting in a discriminative dataset. Two statistical clustering algorithms are compared. A model-based clustering approach classifies the pixels of an RBC and recognizes the RBC as either new or old based. The K-means algorithm is applied to the four-dimensional feature vector extracted from the RBC profile.