Image-Guided Tethering Spine Surgery With Outcome Prediction Using Spatio-Temporal Dynamic Networks

Area of Science:

• Orthopaedic Surgery
• Medical Imaging
• Artificial Intelligence

Background:

• Fusionless surgical techniques like Anterior Vertebral Body Growth Modulation (AVBGM) treat spinal deformities while preserving flexibility.
• Predicting AVBGM outcomes in immature patients is challenging but crucial for surgical planning.

Purpose of the Study:

• To introduce an intra-operative framework for forecasting AVBGM surgery outcomes in scoliosis patients.
• To aid orthopaedic surgeons in planning and tailoring AVBGM procedures.

Main Methods:

• A spatial-temporal corrective network was developed, learning segmental correction similarities and incorporating a long-term shifting mechanism.
• The model integrates dynamic geometric dependencies, temporal dynamics of curve evolution, and features from T2-weighted MRI inter-vertebral disks.
• A novel loss function with regularization ensures coherence of corrective transformations with observed spine evolution.

Main Results:

• The network achieved high accuracy, with prediction errors of 1.8 ± 0.8mm in 3D anatomical landmarks.
• Generated spine geometries closely matched ground-truth reconstructions at one and two-year follow-ups.
• Demonstrated significant improvements over comparative deep learning and biomechanical models.

Conclusions:

• The developed intra-operative framework accurately forecasts AVBGM outcomes for scoliosis patients.
• This tool can enhance surgical planning and personalization of AVBGM procedures.
• The spatial-temporal network shows promise in improving predictive accuracy for spinal deformity correction.