Five-landmark photogrammetric quantification of frontal-plane head and neck alignment in cervical dystonia: A six-patient proof-of-concept study

Quantitative monitoring of cervical dystonia (CD) posture usually requires clinical ratings or specialized equipment. A transparent method based on routine frontal photographs may support low-complexity longitudinal documentation, but its measurement properties must be distinguished from treatment effects. We aimed to assess the computational feasibility and behavior of a five-landmark photogrammetry pipeline and to identify requirements for prospective measurement validation. We used six complete photographic pairs identified within a retrospective source cohort of 73 unique patients with CD and longitudinal visual analog scale data. Twenty-four patients had an initial post-correction photograph (T0), and six of these also had an end-of-course photograph (T1); all six complete pairs (12 photographs) were analyzed. Both time points recorded an instructed upright posture. Bilateral shoulder, nose-tip, chin, and neck-center landmarks defined face-centered and shoulder-referenced measures. We normalized horizontal deviations to shoulder width. Although bilateral shoulder landmarks were retained separately, all reported metrics were absolute magnitudes; thus, the direction of tilt or deviation was not represented in the outcomes. Paired changes were calculated as T1 – T0. Individual paired values and effect estimates were primary; two-sided exact sign tests with false-discovery-rate correction were exploratory sensitivity analyses. The pipeline returned values for all 12 images. Face-axis angle decreased from 12.97 ± 9.97° at T0 to 4.57 ± 2.90° at T1 (mean change, −8.39°; 95% CI, −20.75 to 3.96), chin-neck deviation from 8.30 ± 7.55% to 2.23 ± 0.58% of shoulder width (−6.06; 95% CI, −13.75 to 1.62), and nose-neck deviation from 7.81 ± 10.95% to 2.53 ± 0.90% (−5.29; 95% CI, −16.89 to 6.31). Individual trajectories were heterogeneous; one operational collis-axis value was geometrically unstable, and no exploratory sign-test result survived multiplicity correction (all q ≥ 0.094). Signed image-plane sensitivity values crossed the reference axis in one or two of six participants, depending on the metric. The pipeline generated candidate continuous image-plane measures from all 12 images, demonstrating technical computability only. The uncontrolled dataset lacks an untreated baseline, and instructed posture acquisitions thus do not establish validity, responsiveness, or treatment efficacy. Its contribution is methodological: it identifies requirements for prospective validation.