A coarse-to-fine segmentation methodology based on deep networks for automated analysis of Cryptosporidium parasite from fluorescence microscopic images
YANG et al. / Medical Optical Imaging and Virtual Microscopy Image Analysis 2022 (MOVI) Workshop in Conjunction with MICCAI.
In this paper, we present a deep learning-based framework for automated analysis and diagnosis of Cryptosporidium parvum from fluorescence microscopic images. First, a coarse segmentation is applied to roughly delimit the contours either of individual parasites or of grouped ones in the form of a single object from original images. Subsequently, a classifier will be applied to identify grouped parasites which are separated from each other by applying a fine segmentation. Our coarse-to-fine segmentation methodology achieves high accuracy on our generated dataset (over 3,000 parasites) and permit to improve the performance of direct segmentation approaches.