Accurate 3D reconstruction of eardrum perforations using Structure from Motion photogrammetry

Abstract

Objective

3D reconstruction of the ear canal and eardrum perforations of known dimensions using routine endoscopy and the computer vision algorithm Structure from Motion (SfM) photogrammetry.

Methods

Thirteen 3D-printed ear models were created featuring anterior-inferior perforations (ranging 0.7–4.0 mm). One human patient was also included in data collection. A 3.0 mm 0° rigid endoscope connected to a high-definition camera captured endoscopic videos of eardrum perforations. Optical calibration included a chessboard target in coordination with a reference cylinder placed on the concha cavum. Endoscopy was performed to 1 mm from the eardrum, angling the endoscope 10–15° from the external canal axis. SfM photogrammetry was utilized to generate 3D point clouds for perforation measurements. High-resolution microCT scans (12-μm slice thickness) and 3D printed models served as ground-truths to compare against corresponding SfM eardrum reconstructions.

Results

The average absolute difference between microCT and SfM measurements were 0.09 mm with a percentage error value < 11 % amongst the thirteen 3D printed specimens. Bland-Altman plots demonstrated no bias between large and small perforations. In the live patient, 3D reconstruction measurements (1.87 mm length, 1.41 mm width) deviated approximately 6 % from manual ruler measurements of 2.0 mm and 1.5 mm.

Conclusion

This pilot study demonstrates that SfM can generate highly accurate 3D reconstructions of eardrum perforations of varying sizes in 3D-printed models and one human subject. The promising ability to reconstruct live intraoperative patient data highlights its clinical viability, particularly for adding objective measurements to clinical exam, surgical planning, and potentially patient-specific graft design.

Level of evidence

4.