Open World Active Learning for Echocardiography View Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Automated echocardiography view classification typically assumes a closed world, limiting its real-world applicability.
• Classical approaches struggle with unseen views, reducing robustness in open environments.

Purpose of the Study:

• To develop an open world active learning framework for echocardiography view classification.
• To enhance the robustness and data labeling efficiency of automated view classification systems.

Main Methods:

• Implemented an open world active learning approach for classifying known and identifying unknown echocardiography views.
• Utilized a clustering method to group unknown views for expert labeling.
• Iteratively updated the classification network with newly labeled data.

Main Results:

• The proposed open world approach demonstrated superior performance compared to closed world methods.
• Significantly improved the efficiency of data labeling for echocardiography datasets.
• Enhanced the overall robustness of the view classification model.

Conclusions:

• Open world active learning is a more robust and efficient strategy for echocardiography view classification in real-world scenarios.
• The developed approach effectively handles unseen views, increasing classifier reliability.