Hierarchal Bayes model with AlexNet for characterization of M-FISH chromosome images

The analysis of chromosomes is a significant and challenging task for clinical diagnosis and biological research. The technique based on color imaging is a multiplex fluorescent in situ hybridization (M-FISH), which was implemented to ease the exploration of the chromosomes. Thus, in this paper, we propose a novel quasi-Newton-based K-means clustering for the M-FISH image segmentation. Then, we use the expectation–maximization-based hierarchical Bayes model to characterize the M-FISH images. The contextual-based classification and region merging of chromosomal images is made to avoid any misclassification, and we made use of AlexNet, by modifying the activation functions of the sigmoid and softmax layer and for the optimum classification between the autosomal chromosomes and the sex chromosome. Finally, we conducted a performance analysis by measuring accuracy, recall, sensitivity, specificity, PPV, NPV, F-score, kappa, Jaccard, and Dice coefficient and compared with other existing methods and found that our proposed methodology can achieve more percentage of accuracy (6.96%) than the state of the art methods.