AI model aims to identify chick sex at day 4 without opening eggs: full analysis

A new Animals paper points to a possible earlier, less invasive route for in-ovo sexing in layer hatcheries. Published April 30, 2026, the study describes a non-destructive method that uses blood-vessel images from 4-day-incubated chicken embryos and an improved MobileViT-V3 model to classify sex. According to the authors, the system reached 92.26% test accuracy and processed images at 97.6 frames per second, suggesting potential compatibility with commercial sorting speeds if the results hold up outside the lab. (studocu.vn)

The work targets a longstanding problem in egg production: male chicks from layer lines do not lay eggs and are typically not economical to raise for meat, which has led to routine culling after hatch. The paper places that issue in a global welfare context, estimating that roughly 7 billion male chicks are culled each year worldwide. It also argues that sexing at day 4 could matter ethically and operationally because it occurs well before later developmental stages that are central to debates over embryonic nociception and pain perception. A 2026 review in the Journal of Applied Poultry Research notes that in-ovo sex determination is viewed as a feasible alternative to day-old chick culling, but also says the exact onset of pain perception in chick embryos remains unsettled, with one recent study finding first nociceptive responses at day 13. (studocu.vn)

Technically, the Huazhong Agricultural University group built on the lightweight MobileViT-V3 architecture and added two modules: a Micro Feature Enhancement module to improve extraction of fine vascular details, and a Multi-Scale Adaptive Attention Fusion module to better filter useful features from noisy images. The authors said the model outperformed comparison architectures including YOLOv12, ShuffleNetV2, ConvNeXt-T, ResNet, and Swin-ViT on their dataset. They also reported stable performance on lower-quality images, including overexposed, dark, bubble-occluded, and blurry samples, with subset accuracies ranging from 89% to 94%. (studocu.vn)

That matters because weak vascular contrast has been a core barrier to very early image-based sexing. Prior reviews of in-ovo sexing methods have noted that blood-vessel distribution approaches are best suited to the day-3 to day-5 window and work best in eggs with good light transmission, while commercial systems have more often advanced through spectroscopy, hormone, or DNA-based approaches. SELEGGT, for example, says its DNA-based commercial platform works from day 8 onward and reports accuracy above 99.5%, while AP reported in December 2024 that Agri Advanced Technologies had begun operating an in-ovo sexing system at Hy-Line’s Iowa hatchery, processing eggs at one of the largest U.S. chick hatcheries. (mdpi.com)

Industry momentum is clearly building, even if this specific paper is still at the research stage. Europe has been a major driver, with continuing policy pressure around male chick culling and wider uptake of in-ovo systems. Reporting in 2025 indicated that in-ovo sexing had reached a meaningful share of the EU laying flock, while U.S. adoption has started to move from pilot stories to commercial installation. That gives studies like this one a more immediate practical backdrop than similar academic work might have had a few years ago. (europarl.europa.eu)

Why it matters: For veterinarians and poultry health professionals, the paper is less about replacing current hatchery workflows tomorrow and more about where the field may be headed. A non-destructive day-4 method could, in principle, reduce incubation waste, avoid embryo sampling, and fit welfare expectations that are pushing hatcheries away from post-hatch culling. But 92% accuracy is not yet trivial in a commercial setting, where even small error rates can affect hatchery economics, replacement pullet supply, and confidence in the system. The authors themselves frame the model as a scalable, low-cost candidate, not a finished commercial platform. (studocu.vn)

There’s also a practical veterinary angle around validation. Performance in a self-constructed dataset is useful, but commercial deployment would require evidence across different genetics, shell pigmentation, lighting setups, incubator environments, and image artifacts. It would also need a clear understanding of downstream handling of identified male eggs, biosecurity implications, and how the technology compares with already marketed systems on throughput, reliability, labor, and welfare outcomes. Those are the issues hatcheries, integrators, and veterinary advisers will likely focus on before any earlier-stage computer-vision method moves from paper to plant. (studocu.vn)

What to watch: Watch for independent validation studies, breed-specific performance data, and any pilot-scale hatchery deployments that test whether day-4 image-based sexing can approach the reliability of established in-ovo platforms while preserving its main advantage: earlier, non-destructive screening. (studocu.vn)