WISE: Efficient WSI selection for active learning in histopathology

Area of Science:

• Medical Image Analysis
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Deep neural networks (DNNs) achieve high performance in medical image analysis but require continuous improvement.
• Active learning (AL) is a promising strategy for enhancing DNNs in the medical domain.
• Current AL methods in histopathology focus on image patches, neglecting whole-slide image (WSI) selection, which limits performance gains.

Purpose of the Study:

• To introduce a novel WSI-level active learning method, WSI-informative selection (WISE).
• To address the limitations of patch-based AL by focusing on informative WSI selection.
• To enhance the performance and efficiency of DNN models in histopathology using WSI-level AL.

Main Methods:

• Developed WISE, a WSI-level AL method utilizing a new class distance metric.
• WISE identifies diverse and uncertain WSIs for model training.
• Evaluated WISE on real-world Colon and Stomach datasets and the public DigestPath dataset.

Main Results:

• WISE achieved state-of-the-art performance across all tested datasets.
• The method significantly reduced the number of required WSIs by over threefold compared to traditional one-pool settings.
• Demonstrated the effectiveness of WSI-level AL in improving DNN performance.

Conclusions:

• WISE offers an effective and sustainable strategy for enhancing DNNs in histopathology.
• WSI-level AL is crucial for optimizing deep learning model development in cancer diagnostics.
• WISE improves model performance while substantially decreasing data requirements.