CCMT: Cross Collaboration Mult-Task Network for Neonatal Hip Bone Intelligent Diagnosis

Abstract

Ultrasonography plays a vital role in the early detection of Developmental Dysplasia of the Hip (DDH). The key to DDH diagnosis lies in calculating α and β angles, which are determined by specific bones, landmarks, and lines. Sonographers often face challenges of low accuracy and inefficiency in DDH diagnosis. To address this, we propose the Cross Collaboration Multi-Task (CCMT) Network, utilizing a high-resolution network as the representation module. Subsequently, through multi-task learning, CCMT implicitly leverages the interconnections between bone segmentation, landmark localization, and line detection. Furthermore, cross-enhancement of different tasks is achieved by exploiting the spatial dependence between structure-landmark and landmark-line. Extensive experiments on our ultrasound dataset of 1211 images reveal that the average errors in α and β angles are 2.19° and 2.73°, confirming the effectiveness of our approach.

Ähnliche Arbeiten