Chinese Team Develops Versatile AI Model Capable of Analyzing Pathological Images
Beijing, The Gulf Observer: A team of researchers in China has unveiled the country’s first versatile artificial intelligence (AI) model, capable of analyzing a wide array of pathological images spanning more than 20 human organs, including lungs, breast, and liver.
Named PathOrchestra, this large language model (LLM) marks a significant advancement in AI-assisted disease diagnosis, moving beyond single-task models focused on specific cancers to a versatile platform addressing multiple types.
Developed by researchers from Air Force Medical University (AFMU), Tsinghua University, and SenseTime, PathOrchestra utilized China’s largest domestic dataset of nearly 300,000 whole-slide digital pathology images, totaling 300 terabytes of data. Employing self-supervised learning techniques, the model “cross-learned” to analyze diverse organs, performing tasks such as pan-cancer classification, lesion identification, multi-cancer subtype differentiation, and biomarker assessment.
“The diversity in pathological images poses a formidable challenge for AI applications, and PathOrchestra represents a jewel in the crown of image processing,” commented Wang Zhe, a professor from AFMU’s Basic Medical Science Academy.
PathOrchestra has achieved impressive results, boasting over 95 percent accuracy in nearly 50 clinical tasks, including diagnoses of lymphoma subtypes and screening for bladder cancer, according to an AFMU news release.
The model is expected to significantly alleviate pathologists’ workload and enhance efficiency in medical image review, noted the researchers.
PathOrchestra exemplifies China’s dynamic growth in the AI landscape, with the country accounting for 36 percent of more than 1,300 AI LLMs globally, making it the second-largest producer after the United States, as highlighted in a recent white paper on the global digital economy by the China Academy of Information and Communications Technology at the Global Digital Economy Conference 2024.
