Shapeshift 3D Repair - A Fully Automated and Unsupervised Cloud API for the Reliable Reconstruction of Raw 3D Surface Scans Data

Abstract

This paper presents a study to quantify the reliability of the automatic reconstruction tool Shapeshift 3D Repair TM to create watertight, genus 0, precise and accurate 3D surface scan of the human body. Our methodology uses a precise baseline 3D scan acquired from a full body 3D scanner as an input of a scanning process simulator that emulates the properties of a common 3D scanner, the Structure TM by Occipital TM , and the behavior of a typical untrained handheld 3D scanner operator. The output of the simulator is a raw scan (noisy and incomplete). Afterward, the raw scan is fed to Shapeshift 3D Repair TM which outputs a reconstructed scan. We express the reliability of the process in terms of Standard Error of Measurement (SEM). Using the girth difference between the baseline scan and the reconstructed scan, we express the compatibility in terms of Signed Mean Difference (Bias) and Mean Absolute Error (MAE). We compare our results with common reconstruction methods found in the literature and with other studies about the reliability of 3D Scanning, Plaster Casting and Traditional Anthropometry. Context: To create custom medical devices and wearables, the patient's 3D geometry can be acquired using a 3D scanner. The raw 3D scans require post-processing as they are often noisy and incomplete. While organizations using 3D scanners put in place training programs, scans are often of poor quality. To this day, this issue is a hindrance to business models centered on 3D scanning such as the novel 3Dscan-to-3Dprint business model; inadequate scans must be manually corrected by the operators, which is a time-consuming offline process. The study is focused on the simulation of the scanning process and the scan reconstruction of the knee. Results: A fully automated and unsupervised cloud processing service for the reconstruction of the knee has been implemented and is ready to be tested by users and vendors of 3D scanners. Reconstructed scans exhibit leg, knee and max thigh girth error under 0.1 cm, 0.3 cm and 0.4 cm respectively with 95% confidence level while producing properly defined surface that are manifold, genus 0, have good triangle aspect ratio, and have a single surface. With the recent boom of devices featuring an embedded 3D scanner, we believe that in due time, our technology can be accessible to millions of users without the needs of industry-specific hardware or skills.

Reliability and Accuracy of Mobile 3D Scanning Technologies for the Customization of Respiratory Face Masks

“…These measurements are taken using anatomical landmarks on the face and can be classified in two groups: Euclidean distance and geodesic distance. We obtained these landmarks using the Deep-MVLM AI algorithm [17]. This AI takes the 3D scan or model as an input and outputs 73 landmarks on the face.…”

Reliability of Mobile 3D Scanning Technologies for the Customization of Respiratory Face Masks

“…ScandyPro and 3DSizeMe provide a full model of the head, but each output meshes are unregistered and require a reconstruction process. We are using Shapeshift 3D Repair algorithms [27] to reconstruct the scan, we then use the Deep-MVLM AI model to locate 73 landmarks [28]. We use the 73 landmarks to find the symmetry plane and mirror the mesh, this helps improve the mesh accuracy by averaging the left and right sides.…”