PRET is a few-shot system for pan-cancer recognition without example training.

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Medical diagnostics

Background:

• Pathological examination is crucial for cancer diagnosis, but faces global pathologist shortages.
• Current AI models require extensive labeled data, limiting scalability and practicality.
• Few-shot learning offers a potential solution for data-efficient AI in pathology.

Purpose of the Study:

• To introduce PRET (pan-cancer recognition without examples training), a few-shot AI system for cancer recognition.
• To enable flexible, scalable, and effective cancer recognition across diverse organs, hospitals, and tasks without prior training.
• To address the limitations of conventional AI models in pathological diagnostics.

Main Methods:

• Developed PRET, a few-shot learning system for pan-cancer recognition.
• Evaluated PRET on 23 international benchmarks with 4,484 whole-slide images.
• Assessed performance across 20 diverse pathological tasks.

Main Results:

• PRET outperforms existing approaches across 20 tasks, achieving >97% AUC on 15 benchmarks.
• Demonstrated a maximum performance improvement of 36.76% compared to other methods.
• Achieved clinical-grade diagnostic performance in lymph node metastasis detection using only eight examples, surpassing 11 pathologists.

Conclusions:

• PRET offers a flexible, scalable, and cost-effective solution for AI-based pan-cancer recognition.
• The system paves the way for accessible and equitable AI pathology, benefiting underserved regions and minority populations.
• Few-shot learning significantly enhances AI's practical application in pathology, overcoming data limitations.