Automatic overlapping chromosome segmentation method based on adversarial learning multi-scale features

Abstract

The invention discloses an automatic overlapping chromosome segmentation method based on adversarial learning multi-scale features. Challenges of human chromosome analysis are overlapping of automaticchromosome segmentation, which hinders medical diagnosis and biomedical research. Therefore, the invention provides an adversarial multi-scale feature learning framework, adopts a nested U-shaped network as a generator, and aims to explore 'optimal' representation of chromosome images by using multi-scale features. Conditional generative adversarial network (cGAN) adversarial learning are used topromote output distribution to be closer to a gold standard image; a least square GAN target is adopted to improve the training stability of the framework; and the continuity optimization is carriedout by using the Lovasz-Softmax loss, so that better performance is obtained. Experimental results show that the method is superior to other traditional algorithms in subjective visual effect and objective evaluation standard.

Description

technical field [0001] The invention belongs to the technical field of biomedical image processing, and more particularly, relates to an automatic segmentation method for overlapping chromosomes based on multi-scale features of adversarial learning. Background technique [0002] Human karyotype analysis is an important work in the medical diagnosis of genetic diseases, which is usually performed in clinical and tumor cytogenetics, such as the detection of genetic abnormalities such as Edwards syndrome and Down syndrome. For this type of diagnosis, cytogeneticists typically identify the disease by looking at chromosome excesses or deletions and structural defects, or by comparing images of the patient's chromosomes to the chromosomal banding patterns of a prototype human. However, these chromosomal analysis processes still require considerable manual effort even after years of expert, cytogeneticist studies. Therefore, it is urgent to develop an effective automatic analysis ...

AI Technical Summary

Problems solved by technology

Although these methods have also achieved good performance, because only a few layers of features are used, the feature representation of chromosome images is somewhat simple, and the potential semantic features are not well captured.

Images