Reliability of automated topographic measurements for spine deformity

Benjamin Groisser, Howard J. Hillstrom, Ankush Thakur, Kyle W. Morse, Matthew E. Cunningham, M. Timothy Hresko, Ron Kimmel, Alon Wolf, Roger F. WidmannSpine 2022 · DOI

TL;DR

Researchers have developed a fully automated, markerless 3D surface-scanning system that generates highly reliable torso measurements for adolescent idiopathic scoliosis (AIS) patients. By eliminating the need for manual landmarking and ionizing radiation, this system offers a "near-perfect" reliable method for long-term monitoring of spinal deformities.

What Problem Does This Paper Address?

Adolescent Idiopathic Scoliosis (AIS) is a complex 3D spinal deformity. Currently, the gold standard for diagnosis and monitoring is radiography (X-rays). However, the typical patient may receive up to 16 radiographs over their treatment course, significantly increasing the risk of radiation-induced cancer.

While surface topography (3D scanning of the skin) has been proposed as a safer alternative, existing systems have struggled with two main issues:

  1. Manual Labor: Most systems require a trained technician to manually place physical markers on the patient or click on digital landmarks, which is slow and introduces human error.
  2. Reliability: Because patients move and breathe, getting consistent measurements across different sessions or different clinicians has been difficult.

This paper introduces a fully automated pipeline that aims to make surface topography fast, objective, and reliable enough for clinical use.

Methods

The researchers utilized the 3dMDbody system, a photogrammetric rig consisting of 30 cameras (20 black-and-white stereo pairs and 10 RGB cameras) capable of capturing a 360° torso reconstruction in milliseconds.

The "secret sauce" of this study is the automated processing pipeline:

  • Atlas Registration: Instead of a human finding landmarks, the software takes a generic human torso "atlas" and deforms it to fit the raw scan data. This creates a "topologically watertight manifold" where every point on the scan corresponds to a specific anatomical location on the atlas.
  • Measurement Suite: The system automatically calculates two types of measures:
  • Intrinsic Measures: Geometrically invariant properties like spine length, back surface area, and section volumes.
  • Pose-Dependent Measures: Parameters sensitive to how the patient stands, such as the Angle of Trunk Rotation (ATR) and "Qangle" (a topographic analogue to the radiographic Cobb angle).

Datasets

The study evaluated the system using two primary datasets:

  1. Human Cohort: 46 adolescents (26 with AIS and 20 healthy controls). Each subject was scanned in three poses (EOS-pose, A-pose, and Adam’s Bend) across multiple trials to test intra-rater and inter-rater reliability.
  2. Rigid Phantom: A lightweight torso mannequin was used as a "ground truth" control. This phantom was scanned both by the 3dMD system and by a high-resolution CT scanner to validate absolute measurement accuracy.

Key Insights and Ideas

  • Automation Boosts Reliability: By removing the "human in the loop" for landmarking, the system achieved remarkably high consistency. When a human doesn't have to decide exactly where "C7" or the "PSIS" is on a digital mesh, the variance drops significantly.
  • Intrinsic vs. Pose-Dependent: The researchers found a clear divide in reliability. Intrinsic measures (like total back area) had near-perfect reliability (ICC > 0.98) because they are mathematically robust to minor shifts in posture. Pose-dependent measures (like rotation) are more volatile because they reflect actual changes in how the patient holds themselves.
  • Speed and Workflow: The total average time for optical scanning was 2.7 minutes—comparable to a standard EOS X-ray—but the automated analysis takes only about 10 minutes, making it viable for a busy clinic.

Key Results

  • Accuracy: On the rigid phantom, the system showed a mean RMS error of only 1.7% compared to the gold-standard CT scan.
  • Human Reliability: 80% of all measured parameters achieved an Intraclass Correlation Coefficient (ICC) of ≥ 0.90 (where 1.0 is perfect).
  • Clinical Correlation: In a preliminary investigation of 105 patients, the topographic "Back Surface Rotation" (BSR) showed a strong correlation ($R=0.72$) with the radiographic Cobb angle, suggesting that surface shape is a meaningful proxy for internal skeletal alignment.

Limitations and Future Work

While the results are promising, the authors note a few hurdles:

  • Postural Sway: Even with a fast scanner, pose-dependent measures like the "Qangle" showed lower reliability (ICC as low as 0.49 in some poses) because patients naturally sway or shift between scans.
  • The "Black Box" of Bone vs. Skin: Surface topography measures the "envelope" of the body. While it correlates with the spine, it isn't a 1:1 replacement for seeing the vertebrae. Future work is needed to create more sophisticated models that can better predict internal skeletal alignment from surface data.
  • Clinical Integration: Further research is required to determine if these objective surface measurements can effectively track the progression of scoliosis over years as accurately as X-rays.

Reference: Groisser, B.N., Hillstrom, H.J., Thakur, A. et al. Reliability of automated topographic measurements for spine deformity. Spine Deformity (2022). https://doi.org/10.1007/s43390-022-00505-9

Summary updated

Abstract

clinical parameters can be reliably measured in a fully automated system, paving the way for objective analysis of symmetry, body shape pre/post-surgery, and tracking of pathology without ionizing radiation.
BibTeX
@inproceedings{groisser2022reliability,
  author = {Benjamin Groisser and Howard J. Hillstrom and Ankush Thakur and Kyle W. Morse and Matthew E. Cunningham and M. Timothy Hresko and Ron Kimmel and Alon Wolf and Roger F. Widmann},
  title = {{Reliability of automated topographic measurements for spine deformity}},
  booktitle = {Spine},
  year = {2022},
  doi = {10.1007/s43390-022-00505-9},
}
Figure