A tiny slice of tissue can hold the key to detecting disease – but reading it accurately often requires highly trained specialists.

As China works to expand precision healthcare, a shortage of pathology experts has become a challenge. The Jinfeng Laboratory in southwest China's Chongqing is turning to AI to develop next-generation diagnostic tools that could make medical expertise more accessible.

In a quiet pathology lab, a doctor places a fingernail-sized glass slide under a microscope, searching for microscopic signs that can reveal what is happening inside a patient's body.

"This is a biopsy slide taken after brain tumor surgery," said Yao Rui, an attending pathologist at Hygeia Hospital. "Under an optical microscope, the differences between normal and tumor cells are measured in micrometers – just a fraction of the diameter of a human hair."

Pathologists must identify these subtle differences under a microscope to support accurate diagnoses, a process that requires extensive expertise and careful analysis. At Hygeia Hospital, the pathology department receives slides covering more than 100 types of diseases every day.

For complex cases, patients may also need additional molecular testing, with results often taking more than 10 days, further increasing the time and workload involved in diagnosis.

With an industry benchmark of one to two pathologists per 100 hospital beds, China faces a shortage of more than 100,000 pathology specialists, especially in community hospitals. AI-assisted pathology offers a new way to boost diagnostic efficiency and help bridge regional gaps in healthcare services.

Jinfeng Laboratory is a key strategic research platform in Chongqing's drive to build a nationally influential innovation center.

Compared with conventional pathology, AI-assisted diagnosis can rapidly analyze digital slides and related clinical data, supporting doctors in diagnosis and treatment.

"We receive pathology data from community hospitals in real time," said Yan Xiaochu, chief scientist at the Intelligent Pathology Center of Jinfeng Laboratory. Yan said AI models can provide diagnostic results, predict disease outcomes, and offer treatment recommendations, allowing community hospitals to access more consistent, high-quality pathology services.

Accurate AI diagnosis depends on large volumes of high-quality pathology data. Drawing on its growing data resources, Jinfeng Laboratory is building a pathology data network covering medical institutions across different regions and levels.

Li Sen, project director of the Intelligent Pathology Center of Jinfeng Laboratory, said Jinfeng Laboratory has partnered with 49 medical institutions across 21 provinces. It has also been approved to establish a national pathology phenotype data sub-center.

According to Li, it is the only institution in China capable of collecting and standardizing pathology data across provinces, medical centers and institutional levels.

Supported by this data network, Jinfeng Laboratory has connected 43 medical institutions across 35 districts in Chongqing. Its system integrates six intelligent diagnostic modules, reducing the time needed to annotate a pathology slide from five minutes to 50 seconds, with an accuracy rate of 87%. It has also completed remote diagnoses for more than 400 complex cases.

Building on these advances, Jinfeng Laboratory released China's first standardized brain tumor pathology dataset certified by the National Institutes for Food and Drug Control in April.

Yang Yi, a researcher at the Intelligent Pathology Center, said the dataset covers 474 clinical cases and centers on 3,420 whole-slide digital images, forming a multimodal data matrix.

Building a 'common language' for AI pathology

Companies and research teams developing AI-assisted brain tumor diagnostic tools in China must submit their products to the National Institutes for Food and Drug Control for review before receiving approval.

Their models are then evaluated through standardized blind testing using this dataset, allowing diagnostic accuracy and stability to be compared under the same benchmark.

The key value of a standardized dataset lies in setting common rules for the industry. Yan said the dataset provides a unified performance benchmark for AI research and development. Once established, the standard is expected to attract medical institutions, research teams and companies from across the country, helping Chongqing become a national hub for pathology data resources.

Globally, rising demand for cancer diagnosis is accelerating the pathology sector's shift toward digital slides, remote consultations and AI-assisted diagnosis. For example, the US Food and Drug Administration has authorized AI software to help pathologists detect prostate cancer, though it remains an assistive rather than independent diagnostic tool.

Meanwhile, institutions including the US National Cancer Institute are building large databases integrating pathology images, genomic data and clinical information, shifting industry competition from algorithm accuracy alone toward high-quality data, multicenter validation, unified standards and regulatory compliance.

(Ran Zheng, Huxin Luo, Xudong Yang, Vivian Yan, Kenny Dong and Dan Liu contributed to the story.)