Open-Set Active Learning for Nucleus Detection From the Histopathological Images.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Deep learning for medical imaging

Background:

• Deep learning models for nucleus detection require extensive labeled data.
• Active learning aims to reduce annotation efforts but struggles in open-set environments with unknown classes.
• Nucleus detection in open-set environments using active learning is an underexplored area.

Purpose of the Study:

• To propose an effective active learning framework for nucleus detection in open-set environments.
• To address the challenge of non-target samples from unknown classes in active learning.
• To reduce the annotation burden for expert histopathologists.

Main Methods:

• A two-stage active learning framework, OpAL4ND, is proposed.
• Stage 1: A prototype-based query strategy using an auxiliary detector selects pure candidate samples from known classes.
• Stage 2: The target detector queries uncertain and representative samples from the candidate set.

Main Results:

• OpAL4ND improves the purity of selected samples from known classes.
• The framework achieves higher nucleus detection accuracy compared to existing methods.
• OpAL4ND significantly lowers the annotation burden required for training.

Conclusions:

• The proposed OpAL4ND framework effectively handles nucleus detection in open-set environments.
• This method enhances the efficiency and accuracy of histopathological examination through reduced annotation.
• OpAL4ND offers a promising solution for applying active learning in challenging medical imaging scenarios.