Enhanced Karyotyping Through Deep Learning-Assisted Segmentation and Classification of Chromosomal Cells

Abstract

The escalating prevalence of genetic disorders in humans underscores the criticality of early detection. Current clinical methodologies, such as cellular activity assessment and chromosomal examination, are widely employed for early-stage prediction of these disorders. However, traditional approaches to chromosomal cell (CC) examination are intricate and labor-intensive. This study proposes a novel deep learning framework (DLF) to overcome these challenges by precisely and efficiently segmenting and classifying CCs. As part of this scheme, images are collected and resized, the DLF is trained using the selected images, segmentation and deep feature extraction of the CC are performed. The proposed methodology involves image collection and resizing, training the DLF with chosen images, CC segmentation and feature extraction and multiclass classification and performance verification. This work implements the VGG-UNet plan to examine the chosen CC images collected from the Biomedical Imaging Laboratory (BioImLab). An experimental inquiry is being carried out on 5474 images and the achieved findings are addressed. The findings of this research suggest that the presented work help to provide a detection accuracy of >98% with the K-Nearest Neighbor supported classifier. This research can be expanded in the future to detect the genetic disorder based on the information obtained from the CC.