AI Model Detects Rare Malignant Cells in Medical Images With Minimal Labeling

Researchers developed one-class representation learning techniques that detect malignant cells in whole-slide cytology images without instance-level supervision, achieving state-of-the-art performance even when malignant cells comprise less than 1% of tissue. The approach outperforms conventional methods in ultra-rare scenarios, addressing a critical challenge in computational pathology where manual annotation is impractical.