Fetal-Net: enhancing Maternal-Fetal ultrasound interpretation through Multi-Scale convolutional neural networks and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Obstetrics

Background:

• Ultrasound imaging is crucial for fetal growth and maternal-fetal health assessment.
• Interpreting fetal ultrasound images is challenging due to complex anatomy and variable image quality.
• Existing deep learning models often focus on single tasks like segmentation or detection, lacking an integrated approach.

Purpose of the Study:

• To develop an integrated deep learning solution for comprehensive fetal ultrasound analysis.
• To improve the accuracy and robustness of fetal structure identification and anomaly detection.
• To create a tool that accounts for the interplay between fetal anatomical structures.

Main Methods:

• Development of Fetal-Net, a novel deep learning architecture.
• Integration of Multi-Scale Convolutional Neural Networks (CNNs) and transformer layers.
• Training on a large dataset of over 12,000 expertly annotated fetal ultrasound images across diverse anatomical planes.

Main Results:

• Fetal-Net demonstrated excellent performance in anomaly detection.
• Achieved high precision (96.5%), accuracy (97.5%), and recall (97.8%).
• Showed robustness against various imaging settings, indicating reliability in real-world scenarios.

Conclusions:

• Fetal-Net offers a potent, integrated solution for fetal ultrasound interpretation.
• The model significantly enhances the identification of fetal structures and anomaly detection.
• Fetal-Net has the potential to augment prenatal care by refining ultrasound image analysis.