Swin-BCPN: A Fetal Ultrasound Four-Chamber Quality Assessment Model Based on Structure Detection

Prenatal ultrasonography is an important method for screening fetal congenital heart diseases. However, the acquisition of standard ultrasound planes heavily depends on the experience of sonographers, which introduces a high degree of subjectivity. To alleviate the workload of sonographers and improve the quality of standard plane acquisition, we propose a fetal four-chamber heart ultrasound image quality assessment model based on anatomical structure detection (Swin-BCPN). The backbone network of the model adopts the Swin Transformer architecture, and we independently designed a Bidirectional Cascade Pyramid Network (BCPN) in the neck network to more effectively integrate contextual features. Additionally, an Efficient Multi-scale Attention (EMA) mechanism was introduced into the BCPN, forming the BCPN_EMA module, to further extract critical features. Experimental results demonstrate that the proposed model outperforms other competing baseline models in terms of detection accuracy, validating its effectiveness and offering a practical solution for prenatal ultrasound examinations.