Overcoming an Annotation Hurdle: Digitizing Pen Annotations from Whole Slide Images

Area of Science:

• Digital Pathology
• Artificial Intelligence
• Computational Pathology

Background:

• Artificial intelligence (AI) in pathology relies on digitally annotated whole slide images (WSI).
• Manual digital annotation is time-consuming and costly.
• Pathologists use pen annotations on glass slides, which are currently excluded from computational analysis.

Purpose of the Study:

• To develop a novel method for segmenting and converting hand-drawn pen annotations into a digital format.
• To make existing pen annotations accessible for AI training in computational pathology.

Main Methods:

• A Python-based, open-source software tool was developed.
• The method segments and fills hand-drawn pen annotations from WSI.
• The approach is robust against text annotations.

Main Results:

• The method successfully extracts pen annotations and saves them as masks.
• Validation on 319 WSI achieved a Dice metric of 0.942, Precision of 0.955, and Recall of 0.943.
• Processing time is 15 minutes, compared to 5 hours for manual annotation.

Conclusions:

• The method can utilize existing pen-annotated slides for training AI models.
• This approach facilitates the collection of large training datasets from digitized pathology archives.
• It enhances the utility of historical pathology data for AI development.