Learning Disentangled Representation for Chromosome Straightening

• DOI: 10.1080/01969722.2023.2296250
• Publication Date: 2023-12-21T09:28:00Z
• Authors: Tao Liu; Yifeng Peng; Ran Chen; Yi Lai; Haoxi Zhang; Edward Szczerbicki
• Keywords: Representation; Feature (linguistics); Feature Learning

Chromosome straightening plays an important role in karyotype analysis. Common methods usually adopt geometric algorithms, which tend to affect the chromosome banding patterns process of straightening, resulting feature changes, loss details, and poor generalization. To solve these problems, this paper proposes a novel method based on disentanglement representation learning. Our consists two main components: Disentanglement Representation Encoder (DRE) Straightening Generator (SG), where DRE discovers disentangles bent content latent space, while SG is used generate straightened images disentangled representations. Leveraging by DRE, our produces reducing keeping unchanged, making without changing its possible. Evaluation results both Frechet Initiation Distance (FID) Downstream Classification Accuracy (DCA) metrics show that achieves good performance.