MsGA: Gestational Age Estimation with Multi-plane Unified Measurements Driven by Anatomic Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Obstetrics

Background:

• Accurate gestational age estimation is crucial for optimal prenatal care and clinical decisions.
• Current ultrasound methods face limitations due to segmentation model capacity, image noise, and measurement variability.
• Existing approaches struggle with insufficient information representation and inter-observer variability.

Purpose of the Study:

• To develop an advanced model for accurate gestational age estimation using multi-plane unified measurements driven by anatomic segmentation.
• To overcome the limitations of conventional methods in ultrasound-based gestational age estimation.
• To introduce a novel deep learning framework for improved prenatal assessment.

Main Methods:

• Proposed the MsGA model incorporating a lightweight LGF-UNet for anatomic segmentation and a Point Regression module for landmark localization.
• Developed a Deep Patch Embedding module, Local-Global Fusion Transformer block, and Focusing Attention Bottleneck module within LGF-UNet.
• Created a fully annotated ultrasound plane dataset spanning various gestational stages for model training and validation.

Main Results:

• The MsGA model demonstrated superior performance compared to existing methods on the Gestational Age Estimation task.
• Extensive experiments validated the effectiveness of the overall MsGA model and its individual components.
• The proposed model achieved state-of-the-art results with a reduced number of parameters.

Conclusions:

• The MsGA model offers a significant advancement in automated gestational age estimation from ultrasound images.
• The proposed LGF-UNet and Point Regression modules effectively address challenges like image noise and landmark localization.
• This novel approach provides a more accurate and efficient tool for prenatal care, enhancing clinical decision-making.