Automated Annotator: Capturing Expert Knowledge for Free.

Area of Science:

• Medical image analysis
• Artificial intelligence in medicine
• Computational pathology

Background:

• Deep learning for medical image analysis requires extensive expert annotations, which are time-consuming and expensive.
• Automated annotation methods are needed to reduce the burden on medical experts.

Purpose of the Study:

• To develop and evaluate a cost-effective automated annotation pipeline for medical images.
• To leverage autoencoder-based heatmaps for extracting high-level information from histology images.
• To demonstrate the pipeline's effectiveness in the context of coeliac disease histology.

Main Methods:

• An automated annotation pipeline utilizing autoencoder-based heatmaps was developed.
• The model predicts heatmaps on unseen images, functioning as a 'robot annotator'.
• The method was applied to coeliac disease histology images and evaluated by a pathologist and a deep network classifier.

Main Results:

• The automated annotation pipeline demonstrated encouraging initial results.
• The approach effectively extracts high-level information from histology images.
• Empirical evaluation using a coeliac disease classification network showed positive outcomes.

Conclusions:

• The proposed automated annotation pipeline is a simple yet effective method for medical image analysis.
• This task-agnostic approach has the potential for broad application in various medical image annotation tasks.
• Further investigation is warranted, especially concerning scalability for widespread user adoption.