Comprehensive reconstruction of the musculoskeletal anatomy in the shoulder using a hybrid 3D ultrasound mosaicking workflow: A pilot study

Abstract

Background Three-dimensional (3D) ultrasound (US) imaging of complex joints such as the shoulder is limited by a small field of view and shallow imaging depth. As a result, capturing the entire shoulder anatomy in a single scan is not feasible. This study evaluates a hybrid 3D US mosaicking workflow that combines probe pose tracking, image-based alignment, and magnetic resonance imaging (MRI) guidance to reconstruct the full shoulder musculoskeletal anatomy. Methods Five healthy volunteers underwent shoulder imaging using a robot-assisted 3D US probe. Each shoulder was divided into five scan regions, yielding approximately 45 partially overlapping US volumes per trial. Volumes were first positioned using probe pose information and then refined using a hybrid alignment strategy that combined semi-automatic US-US registration with expert manual adjustments guided by subject-specific MRI bone models. Reconstruction performance was assessed by measuring alignment consistency of the humerus across overlapping volumes using Dice similarity coefficients and multimodal overlap metrics. Results The hybrid workflow improved reconstruction accuracy by approximately 17% compared to pose estimation alone. The refined reconstructions showed clearer anatomical continuity and fewer alignment artefacts. Reconstruction accuracy was influenced by the amount of overlap between scans and by differences in shoulder position between US and MRI acquisitions. Weighted mean compounding performed best for regional reconstructions with sufficient overlap, while maximum compounding was more effective for whole-shoulder mosaics. Conclusion This study demonstrates the feasibility of reconstructing the full shoulder anatomy using a hybrid 3D US mosaicking approach. The proposed workflow improves alignment accuracy and anatomical visualisation compared with pose estimation alone and provides a practical reference framework for future large-scale and automated, real-time musculoskeletal US reconstruction.