Vertebral segmentation & alignment

Spinal alignment, measured automatically.

SpineContour is an open-source research application that measures cervical, lumbar and standing full-spine radiographs, detecting which it has been given. It reports the alignment parameters used in spine research: C2–C7 Cobb angle and cervical SVA, lumbar lordosis, pelvic incidence, pelvic tilt, sacral slope and PI–LL mismatch, and the global C7–S1 SVA on standing films. It runs in seconds, entirely on your workstation.

Download Measurement accuracy

01 · What it does

Segmentation and landmarks

Contours for L1–S1 with a per-study confidence score, and 23-point landmarks on cervical films. Review and correct either directly on the image.

Alignment parameters

C2–C7 Cobb and cervical SVA, lumbar lordosis, pelvic incidence, pelvic tilt, sacral slope and PI–LL mismatch, plus segmental lordosis and disc angulation at every level. Exportable as CSV or a PDF report.

Local by design

Nothing is uploaded. Files are read from disk on this workstation, and de-identified research images never leave your machine.

Your data never leaves your workstation

SpineContour runs entirely on the machine you install it on. Radiographs and clinical data are read from disk and stay there. There is no mechanism in the application to send them anywhere: no upload, no cloud processing, no telemetry, no account.

No upload path for images, measurements, or reports
No account, login, or license server
No analytics, telemetry, or crash reporting
Clinical data is imported from a CSV file you choose
Exports are written to a location you pick on disk
Works with networking disabled

02 · Inside the app

SpineContour study analysis of a standing lateral lumbar film: coloured outlines on L1 to L5, femoral head circles, and the measurements panel with sagittal parameters, segmental angles and disc heights — Study analysis: per-vertebra outlines, sagittal parameters, segmental angles and disc heights, shown here on real model output.

Studies screen: searchable table of studies with view, status and lordosis columns — Organise studies: search, batch status, key measurements at a glance.

Studies screen in dark mode: searchable table of studies with view, status and lordosis columns — Organise studies: search, batch status, key measurements at a glance.

Workspace screen: image folder and clinical data CSV import steps — Import a folder of films and their clinical data from CSV.

Workspace screen in dark mode: image folder and clinical data CSV import steps — Import a folder of films and their clinical data from CSV.

03 · Measurement accuracy

Externally validated against expert measurement.

Four networks were trained on 10,387 images from four public datasets: three HRNet models for the vertebrae (C2–C7 and L1–S1 endplates) and a binary U-Net for the femoral heads. Testing used 100 lateral full-spine radiographs from the Lateral Full-spine X-ray Image Dataset, a separate collection excluded entirely from model development. Uncorrected automated measurements were compared with a neurosurgery spine fellow's reference measurements by mean absolute error, root mean square error, prediction R², and absolute-agreement ICC(A,1).

0.998 ICC(A,1) · Global SVA external test set, n = 100

0.834 Lowest ICC(A,1) of seven every parameter good to excellent

1.2 s Inference per radiograph Apple M2 Max laptop (SD 0.02 s)

External test set, n = 100 · automated vs. reference measurement
Parameter	ICC(A,1)	Agreement
Global SVA	0.998	Excellent
Pelvic tilt	0.983	Excellent
C2–C7 SVA	0.960	Excellent
Sacral slope	0.933	Excellent
Pelvic incidence	0.916	Excellent
C2–C7 Cobb angle	0.835	Good
L1–L5 lordosis	0.834	Good

Disc heights: MAE 0.68 mm (C2–C3, middle) to 2.83 mm (L4–L5, anterior). Segmental angulation: MAE 2.70° (C4–C5) to 4.36° (L2–L3). Images with a missing landmark, such as an obscured C7, were not evaluated on the parameters that depend on it. Full MAE, RMSE and R² results are in the forthcoming methods paper.

04 · Research use

Research use only

Not for diagnostic or clinical use.

SpineContour is an investigational research tool. Its measurements, segmentations, and derived metrics have not been cleared or approved by any regulatory body. All output requires independent verification by a qualified reviewer.

The model weights, training and evaluation code, and annotation tool are released openly. We invite other centers to test the platform, use it in their own research, and contribute annotations so that accuracy can be established across sites.

We're committed to keeping SpineContour open source and free to use. Please contact spine-contour@gmail.com with questions or if you'd like to collaborate.

spine-contour: An Open Source Platform for Automated Spine Segmentation and Spinopelvic Parameter Measurement (methods paper · forthcoming)

Datasets used in model development · 10,387 images

BUU-LSPINE

Lateral lumbar radiographs with published vertebral endplate coordinates, from Burapha University Hospital.

Paper · Dataset

VinDr-SpineXR

Full-spine radiographs from two hospitals in Hanoi. Together with Merlin, contributes manually annotated images adding vertebroplasty cement and posterior instrumentation examples.

Paper · Dataset

Merlin

Abdominal CT volumes from a United States academic center, released with the Merlin vision–language foundation model.

Paper

CSXA

Cervical Spine X-ray Atlas: 4,963 lateral cervical radiographs with 23 annotated landmarks each, used to train the cervical model.

Paper · Dataset

05 · Get SpineContour

Download SpineContour.

Version 1.1.1. Builds are published on GitHub, each with its checksum, so the link here is always the current release.

Download for Windows · 1023 MB Download for macOS · Apple Silicon · 1001 MB

Neither build is code-signed yet, so your machine will warn you the first time.

Windows shows a blue "Windows protected your PC" screen. Choose More info, then Run anyway.
macOS refuses to open an unidentified developer. Control-click the app and choose Open, or allow it under System Settings, Privacy & Security.

The macOS build is for Apple Silicon only. Intel Macs are not supported.

SHA-256 · v1.1.1

Spine-Contour-Windows.exe

4c0f16a9af09cc12ddd6748536875ac60846bbd59be89970be8d1c896a221b16

Spine-Contour-macOS-arm64.dmg

0a678c5698c16fe614dd6f892aad81b0534d0e0fe1171e5b0b2a5591336245c4

Check yours with Get-FileHash .\Spine-Contour-Windows.exe in PowerShell, or shasum -a 256 Spine-Contour-macOS-arm64.dmg in Terminal.