FUSC: Fetal Ultrasound Semantic Clustering of Second-Trimester Scans Using Deep Self-Supervised Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Manual labeling of fetal ultrasound images is labor-intensive and time-consuming.
• Accurate classification of fetal ultrasound views is crucial for diagnosis and analysis.
• Existing methods often require extensive manual annotation, limiting scalability.

Purpose of the Study:

• To introduce an unsupervised approach, fetal ultrasound semantic clustering (FUSC), for automatic clustering of fetal ultrasound images.
• To reduce or eliminate the need for manual labeling in large-scale ultrasound datasets.
• To improve the efficiency and accuracy of fetal ultrasound image analysis.

Main Methods:

• Developed the FUSC method using an unsupervised clustering approach.
• Trained and validated the method on a large dataset of 88,063 fetal ultrasound images.
• Evaluated performance on an independent dataset of 8187 images, assessing clustering purity.

Main Results:

• The FUSC method achieved over 92% clustering purity on the evaluation dataset.
• Demonstrated the feasibility of automatically clustering fetal ultrasound images without manual labels.
• Showcased the potential for handling large volumes of clinical ultrasound scans efficiently.

Conclusions:

• The FUSC method shows significant promise for automating fetal ultrasound image analysis.
• Automated clustering can reduce the manual labeling burden, enhancing clinical efficiency.
• This approach paves the way for advanced automated labeling solutions in medical imaging.