Contrast and Gain-Aware Attention: A Plug-and-Play Feature Fusion Attention Module for Torso Region Fetal Plane...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Fetal Medicine

Background:

• Accurate fetal torso ultrasound plane identification is crucial for detecting fetal malformations.
• Clinical expertise is vital but limited, necessitating advanced diagnostic support tools.
• Image artifacts and complex fetal anatomy present challenges in ultrasound interpretation.

Purpose of the Study:

• To develop an efficient diagnostic support tool for fetal torso ultrasound plane identification.
• To address challenges posed by image artifacts and varying gain/contrast settings.
• To improve the accuracy and efficiency of identifying key fetal planes.

Main Methods:

• Proposed a contrast and gain-aware attention mechanism integrated into ResNet18 and ResNet34 models.
• The mechanism generates images under varying gain/contrast conditions and uses attention to mimic clinical decision-making.
• A lightweight attention module performs feature fusion directly on images with different gain/contrast settings.

Main Results:

• The proposed method significantly enhanced performance in identifying key fetal torso planes (abdomen, spine, kidney).
• Performance improvement was observed compared to traditional models.
• The approach demonstrated efficiency with minimal addition to model parameters.

Conclusions:

• The contrast and gain-aware attention mechanism offers an effective solution for fetal torso ultrasound plane identification.
• This AI-driven approach can support clinicians in prenatal examinations.
• The method shows promise for improving early detection of fetal malformations.