Automatic Video Segmentation-Based 3D Reconstruction for Comprehensive Median Nerve Assessment in Healthy, T1D, and SMA Pediatric Populations

Background/Purpose: The assessment of peripheral nerves in children remains challenging. Cross-sectional area (CSA) of nerves changes significantly during development and in conditions like Type 1 diabetes (T1D) and spinal muscular atrophy (SMA). Current ultrasound techniques are limited to single-point measurements at one to three locations, limiting comprehensive evaluation of nerve morphology.[1] We aim to develop a physician-guided method for comprehensive median nerve assessment that enables tracking of structural changes along the nerve in healthy children and those with neuromuscular conditions.

Methods: We capture ultrasound videos of the median nerve and implement segmentation using the segment anything model 2 (SAM2),[2] an open-source computer vision model, leveraging the spatial relationships between frames when scanning distal to proximal. Our approach balances computational efficiency with physician control through selective refinement. From segmented frames, we reconstruct 3D models to measure novel metrics including CSA changes, shape variations, and volume. We compare algorithmic segmentations to physician-annotated ground truth in healthy children and patients with T1D and SMA.

Results: Preliminary results show successful nerve segmentation with high concordance between algorithm and physician annotations, as judged by a physician, with promising reductions in annotation time. Three-dimensional reconstruction enables visualization and quantification of metrics unavailable with conventional point measurements. Initial 3D nerve reconstructions reveal distinct pattern variations between healthy subjects and those with neuromuscular conditions. ([Figs. 1] and [2])

Conclusion: SAM2-based segmenting with 3D reconstruction enables high-precision analysis of complete nerve sections rather than single-point measurements, enhancing detection of pathological changes and making comprehensive nerve assessment feasible in pediatric clinical practice.

Fig. 1 Process of automatic video segmentation-based 3D reconstruction for comprehensive median nerve assessment.

Fig. 2 Comparison of physician- and SAM2-based nerve segmentation.

Publikationsverlauf

Artikel online veröffentlicht:
26. September 2025

• References

• 1 Broser P, Lütschg J. Funktion und Struktur peripherer Nerven aus neuropädiatrischer Sicht: Wie klinische, neurophysiologische und Ultraschalluntersuchungen ergänzende diagnostische Informationen liefern. Klin Neurophysiol 2021; 52
• 2 Ravi N, Gabeur V, Hu Y-T. et al. SAM 2: Segment Anything in Images and Videos. 2024