A research group, led by Representative Director Nao Nitta and Director/CSO Keisuke Goda (who also serves as a Professor at the University of Tokyo) of CYBO, alongside Director Tomohiro Chiba and former Director Yuko Sugiyama of the Cytology Department at the Cancer Institute Hospital of JFCR, has announced the development of the world's first clinical-grade autonomous digital cytology system. It was developed in collaboration with the University of Tsukuba, Juntendo University, and the Kaetsu Comprehensive Health Development Center.
It is expected that its widespread use will improve the standard of medical care and contribute to pathology. The system is equipped with "whole-slide edge tomography" and AI analysis software for cervical cancer, which has been trained on the diagnostic techniques of cytotechnologists and specialist physicians. The testing performance of the AI analysis software was confirmed to be as high as 0.9 ROC-AUC compared to conventional evaluations by experts. The results were published in Nature on February 18.
The introduction of DX and AI in the medical industry, including its use with CT and MRI scans, is advancing rapidly. Until now, it has been difficult to achieve this due to issues such as the lack of information on cytology samples with 3D structures when using a single focus, and the enormous amount of data when acquiring images with multiple focal points.
To achieve objective and highly reproducible testing, CYBO developed "whole-slide edge tomography" technology. This technology can capture high-definition 3D images, including data at different depths, of a cytological sample consisting of 10,000 to 1 million cells on a single slide at practical speeds. By processing images in real-time, the data can be saved in a format suitable for digital observation and AI analysis. Specifically, the system applies edge computing, placing servers near the data source for distributed processing.
The entire slide sample is scanned in 3D and digitally imaged. By arranging video compression technology (HEVC) for cells, 3D image processing and 3D image compression are performed simultaneously on a computer. Through this technology, data ranging from several hundred GB to several TB per sample is compressed to approximately 4 GB without sacrificing image quality. This allows 3D cytological samples to be digitized at high definition and practical speeds and then stored or transferred in a format optimized for AI analysis.
The AI is trained on approximately 200,000 pieces of training data, including representative cases based on evaluations by skilled detail examiners and cytologists at the Cancer Institute. Through this a model capable of high-definition automatic cell classification has been realized. In addition, the cluster of morphological differentiation (CMD) was introduced for estimation of cell classification.
By converting the AI's confidence levels into numerical vectors, the CMD obtained for each individual cell enables the analysis of entire cell populations. CMD is equivalent to the Cluster of Differentiation used in flow cytometry. It allows for an overview analysis of the entire cell population through visualizations such as scatter plots, histograms, and UMAP, which also helps prevent AI from becoming a "black box."
The performance of the developed system and AI analysis was verified using a total of 1,124 cervical liquid-based cytology samples collected from four facilities: the Cancer Institute Hospital of JFCR, University of Tsukuba Hospital, Juntendo University Urayasu Hospital, and the Kaetsu Comprehensive Health Development Center.
The results showed that the AI model maintained consistently high performance, with an AUC of 0.86-0.91 for detection of LSIL-positive cases and 0.89-0.97 for detection of HSIL-positive cases. The LSIL cell counts per slide also confirmed that the AI calculated cell counts corresponded well with clinical risk indicators.
CYBO has already begun sales of "CYBO Scan," which is equipped with "whole-slide edge tomography," and is currently responding to inquiries for quotes.
Journal Information
Publication: Nature
Title: Clinical-grade autonomous cytopathology through whole-slide edge tomography
DOI: 10.1038/s41586-025-10094-y
This article has been translated by JST with permission from The Science News Ltd. (https://sci-news.co.jp/). Unauthorized reproduction of the article and photographs is prohibited.