Weakly supervised classification of aortic valve malformations using unlabeled cardiac MRI sequences

Area of Science:

• Medical Imaging
• Machine Learning
• Cardiology

Background:

• Biomedical repositories offer vast unlabeled cardiac imaging data, posing challenges for supervised machine learning.
• Aortic valve malformations require accurate classification for timely intervention.

Purpose of the Study:

• To develop a weakly supervised deep learning model for classifying aortic valve malformations from unlabeled cardiac MRI data.
• To establish a scalable strategy for training machine learning models on large, unlabeled medical image datasets.

Main Methods:

• Utilized weak supervision with expert-defined heuristics to generate imperfect labels for training.
• Developed a deep learning model for aortic valve classification using up to 4,000 unlabeled cardiac MRI sequences.
• Validated the model's clinical utility by assessing its association with major adverse cardiac events.

Main Results:

• The weakly supervised model significantly outperformed a traditional supervised model trained on hand-labeled data.
• The model identified individuals with a 1.8-fold increased risk of major adverse cardiac events.
• Demonstrated the feasibility of using weak supervision for large-scale medical image analysis.

Conclusions:

• Weak supervision offers a powerful alternative to fully supervised learning for analyzing unlabeled medical imaging data.
• The developed deep learning model provides a robust baseline for aortic valve malformation classification.
• This strategy enables the effective use of big data in biomedical research and clinical applications.