Towards Automated Prenatal Care: Attention-Based Deep Learning for Fetal Head Circumference Measurement

Introduction: Accurate fetal head circumference (HC) estimation from ultrasound images is critical for prenatal assessment, yet current deep learning approaches are limited by scarce annotated training data and inherently low image contrast. These limitations compromise the model's capacity to reliably delineate fetal head boundaries from surrounding uterine structures, directly impacting clinical utility. Material and Methods: This study introduces an attention-based deep learning framework designed to optimize feature extraction by selectively emphasizing diagnostically relevant regions within ultrasound images. The attention mechanism guides the network to prioritize fetal head boundaries while suppressing irrelevant background information, thereby enhancing segmentation precision and feature discrimination during training. Results: Comprehensive evaluation on benchmark ultrasound datasets validates the clinical effectiveness of our approach. The proposed attention-based model achieves a ۲% improvement in fetal head detection accuracy compared to current state-of-the-art methods, while simultaneously reducing overfitting probability by ۵۰%. These gains translate to more robust and reliable HC measurements across diverse imaging conditions. Conclusion: Integration of attention mechanisms into deep neural networks substantially advances automated fetal biometry by addressing two critical challenges: measurement accuracy and model generalization. The demonstrated improvements in both detection performance and overfitting mitigation establish attention-guided learning as a viable pathway toward clinically deployable ultrasound analysis systems, with potential to enhance prenatal care quality and consistency.

کلیدواژه ها:

Fetal Head Circumference ، Ultrasonic Image ، Deep learning ، Attention Based Semantic Segmentation