Nuchal Translucency Assessment in Prenatal Screening: An Attention-Based U-Net Approach

Accurate identification and measurement of nuchal translucency (NT) region from fetal ultrasound images is critical in prenatal screening for chromosomal and congenital disorders. However, manual demarcation of the NT region can be difficult and subject to inter-operator heterogeneity. This work describes an automated solution for nuchal translucency segmentation based on an Attention U-Net architecture. The suggested method builds on the U -Net model's strengths for biomedical image segmentation by incorporating an attention mechanism that improves the model's capacity to focus on relevant characteristics and capture long-range dependencies. The Attention U-Net model is trained on a large dataset of fetal ultrasound images, and its performance can be evaluated using a different metrics, including Dice Similarity Coefficient, Jaccard Index, Specificity, Sensitivity and Area Under the Receiver Operating Characteristic Curve (AUC-ROC) as 0.91, 0.8, 96.17, 96.17 and 99.8 respectively. The findings indicate that the suggested method outperforms standard U-Net and traditional image processing techniques, with high segmentation accuracy as 99.7 and robust performance across a variety of imaging conditions.