Summary
KAYRA is an end-to-end AI karyotyping system designed to operate within clinical cytogenetic laboratory constraints, presented by Attila Pintér and colleagues (arXiv:2604.26869). The system is built as a containerized microservice pipeline that combines an EfficientNet-B5 + U-Net semantic segmenter, a Mask R-CNN (ResNet-50 + FPN) instance detector, and a ResNet-18 classifier, orchestrated via a cascaded ROI-narrowing strategy so each downstream model focuses on chromosome-bearing regions. The same container images can be deployed either as a cloud service or as an on-premise installation, supporting clinical environments where patient data cannot leave the premises. In a pilot clinical evaluation covering 459 chromosomes from 10 metaphase spreads, KAYRA achieved 98.91% segmentation accuracy (vs. 78.21%/40.52%), 89.1% classification accuracy (vs. 86.9%/54.5%), and 89.76% rotation accuracy (vs. 94.55%/78.43%) compared with two commercial reference systems. It outperformed a legacy density-threshold reference on all three metrics (p < 0.0001) and a modern AI-based reference on segmentation (p < 0.0001). The system reaches TRL 6 maturity and includes a human-in-the-loop expert-review workflow required for diagnostic practice.
Paper Overview
Field: Computer Vision
Authors: Attila Pintér, Javier Rico, Attila Répai, Jalal Al-Afandi, Adrienn Éva Borsy, András Kozma, Hajnalka Andrikovics, György Cserey
Published: 2026-04-29
arXiv: 2604.26869
Summary
We present KAYRA, an end-to-end karyotyping system operating within the operational constraints of clinical cytogenetics laboratories. KAYRA is architected as a containerized microservice pipeline whose ML stack combines an EfficientNet-B5 + U-Net semantic segmenter, a Mask R-CNN (ResNet-50 + FPN) instance detector, and a ResNet-18 classifier, orchestrated through a cascaded ROI-narrowing strategy so that each downstream model focuses on chromosome-bearing regions.
The same container images can be deployed both as a cloud service and as an on-premise installation, supporting clinical environments where patient data must not leave the premises.
Pilot Clinical Evaluation
In a pilot evaluation on 459 chromosomes from 10 metaphase spreads, compared with two commercial reference karyotyping systems:
- Segmentation accuracy: 98.91% (vs. 78.21% / 40.52%)
- Classification accuracy: 89.1% (vs. 86.9% / 54.5%)
- Rotation accuracy: 89.76% (vs. 94.55% / 78.43%)
KAYRA outperforms the legacy density-threshold reference on all three metrics (p < 0.0001) and outperforms the modern AI-assisted reference on segmentation (p < 0.0001).
Maturity and Deployment
The system reaches TRL 6 maturity and integrates a human-in-the-loop expert-review workflow required for diagnostic cytogenetics practice. The core argument of the paper: a multi-model cytogenetics AI service can be packaged as a microservice architecture enabling flexible cloud or on-premise deployment while delivering strong empirical performance in pilot clinical evaluation.
---
*Auto-collected on 2026-05-04*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177619239