Automatic Spine Ultrasound Segmentation for Scoliosis Visualization and Measurement.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• X-rays are standard for scoliosis measurement but involve radiation exposure.
• Ultrasound imaging offers a radiation-free alternative but faces challenges in spine visualization and accurate measurement.
• Developing non-invasive, accurate scoliosis assessment tools is crucial for patient safety and accessibility.

Purpose of the Study:

• To integrate tracked ultrasound and AI for a safer, more accessible scoliosis measurement alternative to X-ray.
• To develop automatic ultrasound segmentation for 3D spine visualization and scoliosis measurement.
• To address the inherent difficulties in using ultrasound for spine imaging.

Main Methods:

• A convolutional neural network (CNN) was trained for spine segmentation using ultrasound scans from healthy adults.
• The trained CNN was tested on pediatric patients to evaluate segmentation and 3D volume reconstruction for scoliosis measurement.
• Performance metrics (recall, precision) were analyzed for fuzzy and binary segmentation, and 3D reconstructions were used for scoliosis measurement.

Main Results:

• Fuzzy segmentation metrics showed expected decreases when translating from healthy volunteers to patients (recall: 0.72 to 0.64; precision: 0.31 to 0.27).
• Binary segmentation metrics performed better on patient data after threshold optimization (recall: 0.98 to 0.97; precision: 0.10 to 0.06).
• 3D reconstructions enabled scoliosis measurement in under 1 minute with a maximum error of 2.2° compared to X-ray.

Conclusions:

• Automatic spine segmentation using tracked ultrasound and AI provides an efficient and accurate method for scoliosis measurement.
• This AI-driven approach overcomes previous limitations of ultrasound, positioning it as a viable alternative to X-ray for scoliosis assessment.