How precisely doesa 3D scanner measurethe eye socket?

Living people move. They breathe, blink and pull faces, and every one of those movements lands in the result as measurement error. To find out how precisely a device actually measures, those confounders have to be removed first.

So the measurements were taken on two mannequin heads and five body donors, not on patients. Two observers captured each dummy six times independently, a third observer captured each body donor four times. Whatever deviation remains comes from the device, the software and the operator, and from nothing else.

Volume is not measured across a whole face but within a defined outline. Five such outlines were specified, from the lower eyelid alone to a large area spanning both eyes and the bridge of the nose. The larger the area, the more surface there is for error to accumulate.

Before volume comes distance. Twenty distances were defined between 18 anatomical landmarks and each was measured five times, once digitally on the 3D model and once by hand with a calliper.

The mean deviation between the two methods is 0.019 millimetres per millimetre. Over a 50 millimetre distance that is less than a single millimetre. Repeatability reached an ICC of 0.997 for the digital and 0.996 for the manual measurement.

Across all modes the mean volume deviation in body donors is 0.032 ± 0.113 millilitres. In 14.1 per cent of measurements there was no deviation at all. For the dummies it was 8.1 per cent at a mean of -0.005 ± 0.083 millilitres.

The pattern is clear: the larger the region measured, the larger the deviation. Mode A stays at 0.037 millilitres, mode D reaches 0.099.

Up to this point everything looks excellent. The decisive difference lies elsewhere: evaluate the same scan twice and the results agree almost perfectly. Photograph twice from scratch and agreement drops sharply.

The most striking part is the left to right difference in body donors. The left eye reaches an ICC of 0.83, the right only 0.15. A plausible but unverified explanation is the camera handling of a right handed operator. It is this asymmetry, not the average, that currently limits use for longitudinal monitoring.

The most precise outline is not automatically the most clinically useful one. Mode B measures most accurately but leaves out the eyebrow, and for eyelid surgery that is exactly the structure that matters.

The VECTRA H2 measures distances and volumes in the eye region reproducibly as long as the work stays within a single capture. For documentation and surgical planning that makes it usable.

For monitoring over weeks and months it is not yet sufficient on this evidence, because too much scatter arises between two separate captures. The next step is therefore a study on living patients, under exactly the conditions deliberately excluded here.

The study was carried out at Göttingen University Medical Center together with the University of Applied Sciences and Arts, the Institute of Anatomy and Embryology and the Department of Medical Statistics. Contributors are listed by initials only.

S. S.

Conceptualization, dummy measurements, analysis, manuscript

C. C.

Methodology, second observer, supervision

M. S.

Conceptualization, clinical infrastructure, supervision

C. V.

Conceptualization, access to body donors

A.-K. S.

Methodology, body donor measurements

A. L.

Statistical analysis and consultation

Status

Manuscript submitted and under review. Ethics approval from Göttingen University Medical Center is in place.