ChroSegNet: An Attention-Based Model for Chromosome Segmentation with Enhanced Processing

In modern medical diagnosis, the karyotype analysis for human chromosome is clinically significant diagnosis and treatment of genetic diseases. such an analysis, it critically important to segment banded chromosomes. Chromosome segmentation, however, technically challenging due variable features, complex background noise, uneven image quality images. Owing these technical challenges, existing deep-learning-based algorithms would have severe overfitting problems are ineffective in segmentation task. this paper, we propose a novel model with our enhanced processing, namely ChroSegNet. First, develop processing techniques realize quantity enhancement data, leading dataset subsequent network training. Second, “ChroSegNet" based on U-Net. According characteristics not only improved baseline structure but also incorporate hybrid attention module ChroSegNet, which can extract key feature information location chromosome. Finally, evaluated ChroSegNet obtained MPA 93.31% F1-score 92.99%. Experimental results show that outperforms representative models performance has lightweight structure. We believe proposed highly promising future applications measurement diagnosis.