Deep-Prenatal: A Deep Learning Approach for Detection and Analysis of Nuchal Translucency for Enhanced Prenatal Screening

Nuchal Translucency (NT) observed in ultrasound scans is widely utilized to identify genetic abnormalities in foetuses. However, NT measurements may be challenging to perform as a mainstay of prenatal screening. These challenges include operator dependency, which primarily relies on the technician’s expertise, and variations in imaging settings that can introduce data anomalies. Thus, the precision of NT measurements is not always definitive, raising questions about their dependability in clinical practice. A Convolutional Neural Network (CNN) framework is proposed to address these constraints. This architecture comprises of Conv2D, Max Pooling, Dense, and Flatten layers and advanced components. Together, these elements enable the CNN to effectively identify and analyse NT measurements, thereby automating the process and reducing reliance on human involvement. The proposed CNN architecture performs well across critical criteria such as accuracy, precision, recall, and Intersection over Union (IoU). Furthermore, the framework has demonstrated stability when applied to diverse datasets, owing to its robustness and flexibility across various clinical databases. This proposed efficient and reliable preliminary screening method, based on the enhanced CNN architecture, addresses the inherent limitations of NT measurements. It establishes a new benchmark for accuracy in the field and introduces a transformative approach that enhances clinical outcomes.