Classification of M-FISH images using fuzzy C-means clustering algorithm and normalization approaches

Abstract

Multiplex or multicolor fluorescence in situ hybridization (M-FISH) imaging is a recently developed cytogenetic technique for cancer diagnosis and research on genetic disorders. By simultaneously viewing the multiple-labeled specimens in different color channels, M-FISH imaging facilitates the detection of subtle chromosomal aberrations. This color karyotyping technique largely depends on the accurate pixel classification. We propose a Fuzzy c-means (FCM) clustering based approach for this problem. The 24 classes of chromosomes form different cluster centers and a pixel is assigned to each individual cluster according to its nearest distance to the center. The FCM has the advantage of locating the clusters more accurately by assuming a membership degree from 0 to 1. The classification was tested on an M-FISH database, demonstrating improved accuracy over other approaches such as the k-Means based clustering. We also discuss image normalization issues such as background subtraction, multispectral channel image registration and dimension reduction, which can lead to improved accuracy of pixel classification. The improved classification accuracy will help the cytogenetician to identify DNA rearrangement more precisely.