Segmentation and Estimation of Fetal Biometric Parameters using an Attention Gate Double U-Net with Guided Decoder...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Fetal Medicine

Background:

• Fetal biometric parameters are crucial for assessing fetal health.
• Current methods using conventional image processing are prone to errors.
• Accurate fetal biometry is essential for timely diagnosis and intervention.

Purpose of the Study:

• To develop an advanced deep learning model for precise fetal biometric parameter prediction.
• To enhance the accuracy of measurements from low-resolution ultrasound images.
• To improve upon existing fetal ultrasound analysis techniques.

Main Methods:

• Introduction of the Attention Gate Double U-Net with Guided Decoder (ADU-GD) model.
• Utilizing attention mechanisms and guided decoders for feature merging.
• Benchmarking ADU-GD against established deep learning models.

Main Results:

• ADU-GD achieved a Mean Absolute Error (MAE) of 0.99 mm and 99.1% segmentation accuracy.
• High Dice index (99.1 ± 0.8), low Hausdorff distance (1.01 ± 1.07), and low Average Symmetric Surface Distance (0.25 ± 0.21).
• Outperformed models like Double U-Net, DeepLabv3, and Trans U-Net in predicting Head Circumference, Abdomen Circumference, Femur Length, and BiParietal Diameter.

Conclusions:

• The ADU-GD model demonstrates superior performance in fetal biometric parameter prediction.
• The model offers enhanced precision and accuracy for fetal ultrasound analysis.
• ADU-GD represents a significant advancement in automated fetal health assessment.