Automation of Chromosomal Abnormality Detection ～Applications in dose assessment and genotoxicity testing following radiation overexposure～

Challenge

Chromosomal translocations are known to be an important index in patient follow-up and biological dosimetry after radiation overexposure. When exposed to a relatively lower dose of ionizing radiation, the frequency of chromosomal aberrations is less than 1%, yet it is necessary to observe 5,000 metaphase cells per patient. Even when using high-throughput microscopy systems, conventional methods rely on expert researchers visually inspecting chromosomal images, resulting in significant time and labor requirements for analysis.

LPIXEL’s Solution – AI Technology and Impact

A deep-learning model for chromosomal aberration detection was developed and evaluated using multi-color FISH images. Specimens were prepared from human peripheral blood irradiated with 2.0 Gy of ⁶⁰Co gamma rays, and the AI model was trained on 505 metaphase images containing more than 21,000 chromosomes. The model achieved over 95% accuracy in chromosome detection and up to 75% accuracy and recall in classifying aberrations such as chromosomal translocations. An analysis of 1,000 chromosome images can be completed in under five minutes, demonstrating the potential for a significant improvement in efficiency compared with conventional manual inspection. We will continue to develop a user-friendly analysis system, aiming to contribute to rapid dose assessment following radiation overexposure and to chromosomal aberration research in the field of clinical cytogenetics. This technology can also be applied to chromosomal aberration testing in drug discovery research.