QualityDDH: visualized standardization of neonatal hip ultrasound via a structural prior regression framework

Ruhan Liu, Yuan Zhang, Xiaoxiao Luo, Yiwen Zheng, Qirong Liu, Mengyao Liu, Lixin Jiang

Abstract: Early ultrasound screening of developmental dysplasia of the hip (DDH) is crucial for timely intervention and preventing hip replacement. However, the lack of standardization in image acquisition during DDH ultrasound screening often hinders the accuracy and consistency of the screening process, posing a significant challenge. Therefore, there is an urgent need to develop automated, visual, and high-precision methods to assist in image standardization. Conventional quality classification methods, which rely on the recognition and judgment of anatomical structures, struggle to achieve accurate results. To address this, we propose the QualityDDH framework, a visual quality assessment tool. It uses structural priors to obtain key structural segmentation maps and assesses anatomical availability based on standardized guidelines. We applied the QualityDDH framework to clinical prospective validation. It assisted ultrasound physicians of different levels in making standardized judgments using an independent external validation dataset of 600 infants. The QualityDDH framework improved performance for ultrasound physicians at all levels: Expert (Area Under the Receiver Operating Characteristic Curve, AUC, increased by 4.70%), Attending (AUC increased by 12.95%), and Resident (AUC increased by 20.85%). This lays the foundation for the clinical application of intelligent auxiliary screening for DDH. Code available at: https://github.com/Liuruhan/QualityDDH.

External IDs:dblp:journals/vc/LiuZLZLLJ25