CNN and Computational Configuration Mathematical Chromosomal Defect Analysis in Medical Images

Abstract

Chromosomal defect analysis plays an important role in current medical care and diagnosis as one of the principal methods in cytogenetics via the processing of a medical picture. There are two main elements of human karyotype analysis: first, chromosomes are separated by digital images of the chromosome metaphase under the microscope. Chromatids are then closely analyzed, compared, organized and categorized. The segmentation and classification operation is tedious, where conventional geometric or mathematical approaches have only limited impact due to low precision, according to this technique. In most cases however, the workflow is still highly supervised and errors are still required by humans. This paper provides an optimised workflow to isolate and automatically identify chromosomes by a combination of many CNN and mathematical optimizations called mCNN GO. Mask R-CNN is investigated to separate the chromosome from chromosome metaphase images and train mCNN GO to identify the sub-images. We apply a new functional approach to synthesize images on the labelled data in order to enhance the efficiency of the segmentation network. Moreover, to ensure accuracy of the results, we create computational algorithms to straighten the genomes before registration. Experimental findings indicate that our methods for automated karyotype analysis are greatly superior to state-of-the-art.