Chromosome segmentation and classification: an updated review

Karyotyping is a study of chromosomes to identify various chromosomal aberrations related to structure and number. Chromosome image analysis involves challenging issues related to overlapping and touching of chromosomes. Chromosome segmentation and classification generally focus on separating overlapping and touching chromosomes. The analysis methods start from conventional image processing methods to advanced machine learning techniques. These methods are broadly classified into low-level and high-level methods. The low-level methods are thresholding-based approaches, edge detection, feature extraction techniques like active contours and watershed approaches and machine learning for classification. The high-level methods are deep learning algorithms like convolutional neural networks (CNNs), U-Net, autoencoder architectures. These methods help in improving accuracy and automate the process of chromosome segmentation and classification. High-level approaches can handle complexity in chromosome overlaps which provides better segmentation results. The approach learns complicated patterns and structures of chromosome images, which helps in achieving better classification accuracy. The challenges are: (i) working on large and annotated dataset for training deep learning models and (ii) suffer issues with new dataset even in they perform better during training phase. The solution for all these can be a hybrid approach that combines conventional method with modern approaches. This survey gives readers a basic understanding of automated karyotyping and future direction in this domain.