Standard methods for creating digital skeletal models using structure-from-motion photogrammetry.

Abstract

This article assesses best practices for producing 3D digital cranial models through structure-from-motion (SfM) photogrammetry, and whether the metric accuracy and overall presentation of photogrammetric models are comparable to physical crania. It is intended to present a user-friendly standard method of creating accurate digital skeletal models using Agisoft PhotoScan. Approximately 200 photographs were taken of three different crania, and were separated into series consisting of 50, 75, 100, 150, and approximately 200 photos. Forty-five cranial models were created using different photo series and a variety of PhotoScan settings. These models were assessed based on defined qualitative criteria, and model measurement estimates were compared with physical skeletal measurements using Bland-Altman plots. The majority of all models (37/45) produced measurement estimates with mean differences of 2 mm or less regardless of PhotoScan settings, and therefore demonstrated high levels of agreement with the physical measurements. Models created with 150 photographs and on "high" PhotoScan settings scored the highest in terms of qualitative appearance in the shortest amount of time. In PhotoScan, it is recommended to create cranial models using 150 photographs and "high" settings; this produces digital cranial models that are comparable to physical crania in both appearance and proportion. SfM photogrammetry is a convenient, noninvasive, and rapid 3D modeling tool that can be used in almost any setting to produce digital models, and following the guidelines established here will ensure that these models are metrically accurate.