Automatic Detection of Fetal Nuchal Translucency Based on Ultrasound Images Using Deep Learning Techniques

Among the most major indicators in prenatal ultrasound screening is fetal nuchal translucency (NT). An increased in the thickness of NT is regarded as an effective marker for detecting chromosomal and structural abnormalities in the fetus. This study introduces a state-of-the-art deep learning approach to automatic detection of the NT using YOLOv8 technique based on a dataset of ultrasound images for fetal NT at the end of the first trimester of pregnancy. The YOLOv8-seg model was implemented for NT detection with different variants: nano (n), small (s), and medium (m), A comparative analysis was performed for these versions. The model performance was evaluated by calculating P, R, and mAP stand for precision, recall, and mean average precision. The results showed that the YOLOv8m-seg (medium) emerged as the superior model. The performance metrics of this model of (P, R, mAP50, mAP50-95) for the bounding box were (0.998, 1, 0.995, 0.991), respectively. The outcomes reflect the excellent performance of the proposed approach for automatic NT detection.