FetalMLOps: operationalizing machine learning models for standard fetal ultrasound plane classification

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Machine Learning Operations (MLOps)

Background:

• Fetal standard plane detection is crucial for prenatal care, aiding in assessing fetal development and identifying anomalies.
• Despite ML advancements, clinical integration is hindered by a lack of standardized operational frameworks.

Purpose of the Study:

• To introduce FetalMLOps, the first comprehensive MLOps framework tailored for fetal ultrasound (US) imaging.
• To bridge the gap between ML innovation and clinical application in prenatal diagnostics.

Main Methods:

• A ten-step MLOps methodology adapted for clinical needs, covering the entire ML lifecycle.
• Standardized ETL processes for data curation, anonymization, and harmonization.
• Deployment via RESTful API with continuous monitoring, emphasizing explainability and sustainability.

Main Results:

• FetalMLOps provides an end-to-end operational framework for ML in fetal US.
• The framework ensures alignment with clinical objectives and real-world medical practice.
• It facilitates the deployment of accurate, efficient, and clinically relevant ML models.

Conclusions:

• FetalMLOps establishes a precedent for trustworthy and scalable AI adoption in prenatal care.
• Operationalizing ML models through this framework enhances clinical workflows and patient outcomes.
• It promotes ethical, transparent, and responsible AI in medical diagnostics.