AFNet Algorithm for Automatic Amniotic Fluid Segmentation from Fetal MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Fetal Medicine

Background:

• Amniotic Fluid Volume (AFV) is a critical indicator for diagnosing fetal abnormalities.
• Accurate AFV quantification is essential for prenatal care and monitoring.
• Current methods for AFV assessment can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate a novel Convolutional Neural Network (CNN) model, AFNet, for automated segmentation of amniotic fluid (AF).
• To improve the accuracy and efficiency of AFV evaluation in clinical settings.
• To enable better diagnosis of gestational disorders through precise AF quantification.

Main Methods:

• Development of AFNet, a CNN architecture featuring efficient feature mapping and transposed convolutions.
• Training and testing AFNet on a manually segmented and radiologist-validated dataset of fetal ultrasound images.
• Comparative analysis of AFNet against state-of-the-art segmentation models like ResUNet++ and UNet++.

Main Results:

• AFNet achieved a mean Intersection over Union (mIoU) of 93.38% on the AF dataset.
• AFNet demonstrated superior performance compared to ResUNet++.
• AFNet achieved comparable performance to UNet++ with significantly fewer parameters (less than half).

Conclusions:

• AFNet provides a highly accurate and efficient method for amniotic fluid segmentation.
• The developed model facilitates objective and reliable AFV quantification in clinical practice.
• This AI-driven approach has the potential to enhance the diagnosis and management of fetal conditions related to amniotic fluid abnormalities.