A new generation of a multimodal large model for lung cancer pathological genomics is released in Guangzhou, the capital of Guangdong province, on Saturday. [Photo provided to chinadaily.com.cn]

A new generation of a multimodal large model for lung cancer pathological genomics was released in Guangzhou, the capital of Guangdong province, on Saturday, helping reduce the cost of lung cancer genetic testing.

The DeepGEM 2.0, jointly developed by the First Affiliated Hospital of Guangzhou Medical University and Guangzhou KingMed Diagnostics Group, a testing company, achieves image-based gene recognition in as little as one minute.

It can simultaneously complete pre-screening for gene targets corresponding to approved lung cancer targeted therapies in China, according to the company.

Along with its release, a multicenter research project involving 50 medical institutions nationwide has been launched, marking the model's entry into the large-scale clinical application validation phase.

The model is planned to be deployed in over 500 medical institutions across the country.

By relying on AI pre-screening and specific gene-targeted verification, the technology is expected to reduce the cost of lung cancer genetic testing to a few hundred yuan, Liang Wenhua, vice-president of the First Affiliated Hospital of Guangzhou Medical University, said.

"As the large model technology is further translated and implemented, it will effectively bridge the last mile of targeted diagnosis and treatment for lung cancer," Liang said.

The technology will help enhance the diagnostic capabilities for major diseases at the grassroots level, reducing medical costs and the financial burden on patients, according to Liang.

The model's first generation was released in 2025 and uses pathological slides as input to predict the probability of lung cancer-related gene mutations and their spatial distribution within tissues.

The prediction accuracy of the model's new version can reach up to 90 percent, Liu Si, head of the AI special project at KingMed Diagnostics, said.

"The model predicts the risk levels of target genes, enabling rapid pre-screening of individuals who may have opportunities for targeted therapies," Liu said.