Beijing, April 22 (SANA) A team of researchers led by the Hong Kong University of Science and Technology has unveiled an artificial intelligence-based pathology system that can diagnose multiple types of cancer using minimal data, marking a potential shift in how such conditions are detected.
The system, known as PRET, relies on a technique called “in-context learning,” originally used in natural language processing, and applies it to the analysis of tissue images, according to a report published Tuesday by the scientific website MedicalXpress.
Researchers say the approach allows the model to quickly adapt to new tumor types and perform a range of diagnostic tasks, including cancer detection and classification, without requiring extensive retraining.
In tests using datasets from several countries, the system achieved accuracy rates exceeding 97 percent in most cases, and reached 100 percent in certain colorectal cancer screenings. The results surpassed the average performance of several pathologists in complex diagnostic tasks.
Developers say the system’s ability to operate with limited data could reduce reliance on large datasets and continuous training, potentially easing pressure on healthcare systems and improving access to accurate diagnoses, particularly in regions with a shortage of specialists.
The project was developed in collaboration with international medical and research institutions, including Harvard Medical School. Findings were published in the journal Nature Cancer, where the system outperformed several existing models currently used in the field.
R A H
