Enhancing Annotation Efficiency with Machine Learning: Automated Partitioning of a Lung Ultrasound Dataset by View

Area of Science:

• Medical imaging analysis
• Machine learning in healthcare

Background:

• Annotating large medical imaging datasets is time-consuming and costly.
• Existing datasets are often not optimized for deep learning requirements.
• Hierarchical task organization can improve annotation efficiency.

Purpose of the Study:

• To develop and evaluate a method for optimizing medical image annotation efficiency.
• To leverage hierarchical task structures for improved data labeling.
• To reduce the cost and time associated with creating large, annotated medical datasets.

Main Methods:

• Trained a machine learning model to classify lung ultrasound (LUS) views.
• Utilized 2908 LUS clips for model training.
• Developed a strategy for partitioning datasets based on view classification.

Main Results:

• Automatic partitioning of a 780-clip dataset by view saved 42 minutes of manual annotation time.
• Achieved an increase of 55±6 relevant labels per hour in a sample task.
• Demonstrated significant efficiency gains in view-specific annotation.

Conclusions:

• Automatic partitioning of LUS datasets by view enhances annotator efficiency and throughput.
• The proposed strategy increases the relevance of annotated labels.
• This hierarchical annotation approach is applicable to other medical imaging datasets.