Â鶹ÒùÔº

Enhancing chromosome detection in metaphase cell images

ChromTR, a novel framework for chromosome detection in metaphase cell images, represents a significant advancement in the field of cytogenetics. This framework, which integrates semantic feature learning and chromosome class distribution learning, is designed to automate the detection and classification of 24 types of chromosomes in raw metaphase cell images.

This is particularly important for the clinical diagnosis of genetic diseases such as Edwards, Turners, and Down syndromes, where accurate chromosome karyotyping is crucial. The research is in the journal Frontiers of Medicine.

The framework, ChromTR, leverages a CNN backbone to extract features from metaphase cell images, followed by a semantic feature learning network (SFLN) for semantic feature extraction and a segmentation head for chromosome foreground region segmentation. A semantic-aware transformer (SAT) with two parallel encoders and a semantic-aware decoder is then used for visual and semantic feature encoding to locate and separate each chromosome.

An attention mask in the semantic encoder restricts the attention computation within the predicted foreground region, improving detection accuracy. Finally, a CDRM is incorporated to determine the number and class distribution of chromosomes, capitalizing on the fixed pattern of chromosome numbers and classes in metaphase cells.

The study demonstrates ChromTR's superior performance compared to existing methods, including one-stage and two-stage object detection methods, as well as other DETR-based methods, on both R-band and G-band metaphase cell image datasets.

The results show improvements in mean average precision (mAP), precision, recall, and a reduction in the average error rate (AER), indicating that ChromTR can reduce the workload of manual modification and improve the efficiency of chromosome karyotyping.

In a clinical test comparing ChromTR with commercial software Ikaros, used in Ruijin Hospital, ChromTR reduced the need for manual correction by approximately 50% and improved the detection rate of numerically abnormal chromosomes, which is crucial for clinical diagnosis. This highlights the potential of ChromTR to revolutionize the field of cytogenetics by providing a more accurate, efficient, and automated approach to chromosome detection in karyotyping.

The framework's ability to integrate semantic features and class distribution reasoning into a unified detection framework positions it as a powerful tool for improving the accuracy and efficiency of chromosome detection and classification, offering significant benefits for clinical diagnosis and genetic research.