Researchers at Tokyo Metropolitan University have developed an artificial intelligence–based system that can automatically detect and count sister chromatid exchanges (SCEs) in chromosome images, offering a faster and more objective alternative to traditional manual analysis. Counting SCEs is an important diagnostic tool for identifying genetic disorders such as Bloom syndrome, but conventional methods require skilled personnel, significant time, and are subject to human interpretation, leading to inconsistencies.
DNA is tightly packaged into chromosomes, which duplicate during cell division to form two identical structures known as sister chromatids. Under normal conditions, these chromatids remain unchanged during mitosis. However, when DNA damage occurs, cells attempt to repair the damage by using the intact chromatid as a template. This repair process can result in the exchange of DNA segments between sister chromatids, a phenomenon known as sister chromatid exchange. While SCEs themselves are not harmful, abnormally high numbers indicate genomic instability and are characteristic of disorders such as Bloom syndrome, which is associated with an increased cancer risk.
Traditionally, SCEs are identified by clinicians examining stained chromosomes under a microscope and visually detecting exchanged regions. This approach is labor-intensive, time-consuming, and susceptible to observer bias. To overcome these limitations, the research team developed a machine-learning framework that integrates multiple algorithms: one to identify individual chromosomes, another to detect the presence of SCEs, and a third to cluster and count them automatically.
The system achieved an accuracy of approximately 84.1%, a level considered suitable for practical clinical use. Validation experiments using cells with a disabled BLM gene—mimicking the genetic defect seen in Bloom syndrome—showed that the AI-generated SCE counts closely matched those produced by human experts.
The researchers are now working to further improve the algorithm by training it on large-scale clinical datasets. They believe that fully automated SCE analysis could standardize diagnostics, reduce workload for clinicians, and mark an important step forward in applying artificial intelligence to medical research and genetic disease detection.
