AI bone measurements: Cobb, EOS, lower limbs, bone age

TechCare Metrics computes up to 60 musculoskeletal measurements in a single pass: Cobb angle, spine telemetry, lower limbs in 2D and 3D (EOS-type) and bone age. One thing sets the tool apart: the measurement is produced from the image, not copied off a worksheet. Values are written straight into the report and are reproducible from one operator to the next. The radiologist validates and keeps the decision.

The gain comes from removing the manual tracing that drives inter-operator variability. The figure is measured: on weight-bearing feet, manual measurement averages 203 seconds per patient (AI/MDPI, 2025), while the AI computes them near-instantly, giving back more than three minutes per case. The pre-report is ready in under 30 seconds, with the full analysis returned to the PACS in under 2 minutes. For pre-operative planning, axes, angles and lengths that stay reproducible case after case give a stable baseline.

Yes, on the parameters studied, and the evidence is published. On weight-bearing feet in adults, a study (AI/MDPI, 2025, 105 patients) reports AI accuracy of 94% for hallux valgus (sensitivity 91.1%, specificity 97.2%), on par with an expert radiologist. On the spine, a study (Diagnostic & Interventional Imaging, 2023, mostly pediatric) validates automated alignment with a mean Cobb-angle error of 2.9° (ICC 0.95-0.97 in the coronal plane) and 99% classification of severe scoliosis. The radiologist checks every value before validating.

Honestly, not everywhere in the same way. Automatic reproducibility reduces the variability of manual tracing, but a measurement still depends on image quality: positioning, projection, or surgical hardware can degrade a landmark. The spine study shows it plainly: full alignment was reliable from the outset in 72 of 100 patients, the rest needing a correction. Hence the principle: the value comes back pre-filled for review, never locked, and the radiologist corrects an outlier in one click. The right question to ask any vendor is not whether it is automatic, but whether the measurement is computed from the image and verifiable.

Measurements integrate with the RIS/PACS over DICOM and HL7, with no change to how you work: the pre-filled report comes back in under 2 minutes. Bone age, lower limbs (EOS) and the Cobb angle are computed in a single pass, then inserted into the report ready to validate. Combined with fracture detection, together they cover routine MSK reading, from adults to pediatrics.

TechCare Metrics is part of the Milvue Suite, CE class IIa (EU MDR 2017/745, BSI CE 2797), trained on over one million images. The measurements are published: weight-bearing feet in an AI/MDPI 2025 study and spinal alignment in a Diagnostic & Interventional Imaging 2023 study. The solution is deployed across 600+ sites in 25 countries. Evidence before the promise.