-
SPS
IEEE Members: $11.00
Non-members: $15.00Length: 00:02:05
Chromosome karyotyping plays a vital role in birth defect diagnosis and biomedical research, in which segmentation is an indispensable task. Current state-of-the-art segmentation methods suffer from bottlenecks due to the small size and polymorphism of the chromosomes, and the overlap between chromosomes is a critical obstacle as well. This work proposes an algorithm for preprocessing datasets and a segmentation network termed RC-Mask for chromosome segmentation. The RC-Mask extends current instance segmentation models by adding a branch for predicting count in parallel with the existing branches, moreover, we add orientation information to both the object detection and semantic segmentation branches. Without tricks, RC-Mask achieves state-of-the-art segmentation accuracy on the original chromosome dataset(63.2%mAP), which outperforms all existing methods in chromosome automatic segmentation. Code is available at GitHub where repository name is RC-Mask.