BICoB, page 167-171. ISCA, 2011 M-FISH IMAGES ANALYSIS THROUGH IMPROVED ADAPTIVE FUZZY C-MEANS SEGMENTATION AND SPARSE REPRESENTATION CLASSIFICATION

Image segmentation and classification are two important steps in multicolor fluorescence in-situ hybridization (MFISH) images analysis. In this paper we first developed an improved adaptive fuzzy c-means (IAFCM) algorithm for the segmentation of the DAPI channel of M-FISH images to extract interested regions for the following classification task. Then we employed a sparse representation-based classification (SRC) algorithm for the M-FISH classification. The developed image segmentation and classification methods have been tested on a comprehensive M-FISH database that we established. When comparing with other M-FISH image classifiers such as fuzzy c-means clustering algorithms and adaptive fuzzy c-means clustering algorithms that we proposed earlier, the current SRC method with proper models gave the lowest classification error. In addition, IAFCM improves the classical fuzzy cmeans algorithm (FCM) by using a gain field that models and corrects intensity inhomogeneities caused by microscope imaging system, flairs of targets (chromosomes) and uneven hybridization of DNA. Experiments showed that IAFCM improved performance in segmentation and resulted in better classification, which will contribute to improved diagnosis of genetic diseases and cancers.