A Computer Vision Approach to Micro-Nucleus Automatic Detection for Protozoan Parasites Segmentation

Till now, protozoan parasites cause many diseases, for examples, malaria, EHEC infection, shigellosis and amoebiasis etc. The kinds and growing stages of protozoan parasites would lead to different treatments. The most significant characteristic of different growing stages is the number of nucleuses, but partial nucleuses of a cell may be more unclear than the others causing the missing in nucleus detection. This paper presents a novel multiple nucleus detection schemes which are composed from the adaptive protozoan parasite erasure, gamma equalization, Fuzzy C-means clustering algorithm, modified connected component detection method, and circle mask scoring method. For each cell, the proposed scheme first detects the most significant nucleus, and then performs gamma equalization iteratively to extract a correct nucleus. In each iteration, only the remained region would be considered. Iterations are terminated when all parameters of gamma equalization are considered. The adaptive protozoan parasite erasure method is used to erase the boundary of a protozoan parasite by dynamic size mask. The modified connected component detection method labels each connected component more accurately than the traditional method. The iterative gamma equalization performs gamma equalization iteratively by different parameters to enhance the boundaries of nucleuses with different edge intensities. The circular mask scoring method can help estimate the circular degree of objects. The experiment shows that the proposed scheme can detect the nucleuses with indistinct boundaries effectively.