(Yicai) July 1 -- Ruijin Hospital, an affiliate of Shanghai Jiao Tong University School of Medicine, and tech giant Huawei Technologies have open-sourced their artificial intelligence pathology model to give smaller hospitals access and thereby improve diagnostic speed and accuracy across China.

RuiPath has been trained on over one million of Ruijin Hospital's high-quality digital pathology slide images covering seven common cancers and was fine-tuned using Huawei's full-stack AI toolchain Model Engine, the hospital said yesterday. In seven out of 14 common diagnostic tests, it has achieved state‑of‑the‑art performance and already has clinical validation capability.

The decision to make RuiPath freely available makes Ruijin Hospital the first Shanghai-based medical institution to release an open‑source pathology diagnosis AI.

“By open-sourcing the diagnostic model -- trained and fine-tuned on Ruijin Hospital’s high‑quality data accumulated over many years -- we will help grassroots hospitals save a lot of effort on model deployment work and promote the widespread use of AI-assisted pathology diagnosis,” Peter Zhou, president of Huawei Information Technology Product Line, told Yicai.

Pathology diagnosis involves putting human tissue or cell samples under a microscope to identify diseases, with tests usually taking three to five days. More complex illnesses can take as long as 10 days. A qualified diagnostician needs five to eight years of training, posing a challenge for smaller hospitals and possibly affecting the quality of diagnosis and thus affect the treatment of patients.

AI pathology models can only generate value and achieve inclusiveness when they are truly applied in clinical practice, said Ning Guang, president of Ruijin Hospital. Making RuiPath open source can also help it improve, delivering a win-win for the developer and grassroots hospitals, he added.

RuiPath covers the diagnosis of 19 common cancer types, accounting for 90 percent of cases in China, Ning said, adding that it will likely be able to diagnose more tumor types after becoming open source.

Editors: Tang Shihua, Martin Kadiev