An open imaging and AI resource enabling unbiased quantification of extrachromosomal DNA at scale

Quantitative imaging of extrachromosomal DNA (ecDNA) is crucial for understanding cancer heterogeneity and adaptation. However, automated analysis has struggled due to a lack of accessible imaging data, gold standard annotations, and adaptable computational tools. To address this issue, researchers have created an open resource for computational ecDNA imaging. This resource combines 2,986 native-resolution metaphase FISH image sets with manual annotations, standardized benchmarks, and open-source quantification frameworks.

The team used this resource to evaluate various approaches for analyzing ecDNA, including rule-based computer vision, deep-learning-based segmentation, and probabilistic localization. Their comparison revealed a count-dependent underestimation that distorts ecDNA copy-number distributions, highlighting the need for a better method to preserve individual ecDNA signals and quantitative burden.

To tackle this problem, the researchers developed ecCount, a probabilistic localization method specifically designed to maintain individual ecDNA signals and their quantitative burden. In their study, ecCount achieved an impressive object-level F1 score of 0.939 on held-out images, demonstrating minimal count bias. By providing images, annotations, retrainable models, evaluation tools, and guided workflows, this open resource empowers the scientific community to apply, adapt, and improve automated ecDNA quantification across different experimental systems.