Efficient fetal ultrasound image segmentation for automatic head circumference measurement using a lightweight deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Obstetrics

Background:

• Fetal head circumference (HC) is crucial for assessing fetal development.
• Current deep learning methods for HC measurement are accurate but often lack efficiency.
• There is a need for faster, resource-efficient HC measurement tools in clinical practice.

Purpose of the Study:

• To develop a highly efficient deep learning model for automatic fetal HC measurement.
• To improve the speed of HC measurement for clinical applications.
• To create a model suitable for deployment on resource-constrained devices.

Main Methods:

• A lightweight deep convolutional neural network (CNN) was designed for fetal head segmentation.
• Sequential prediction network architecture and depthwise separable convolutions were employed to enhance efficiency.
• Post-processing techniques, including morphological operations and ellipse fitting, were used to determine HC.
• Experiments were conducted on the public HC18 dataset.

Main Results:

• The model boasts only 0.13 million parameters, achieving rapid inference speeds (28 ms/CPU, 0.194 ms/GPU).
• It demonstrated comparable accuracy to state-of-the-art methods, with a mean absolute difference of 1.97 mm and a Dice score of 97.61%.
• The model significantly outperforms existing deep learning models in terms of efficiency.

Conclusions:

• A very lightweight deep learning model enables fast and accurate fetal head segmentation for HC calculation.
• The proposed method enhances efficiency and accuracy in fetal HC measurement for obstetricians.
• This approach is suitable for clinical settings with limited computational resources.