Automatic Orientation Estimation of Inertial Sensors in C-Arm CT Projections

Mareike Thies,Björn Eskofier,Andreas Maier,Marc Levenston,Jennifer Maier,Garry Gold,Rebecca Fahrig

doi:10.1515/cdbme-2019-0050

Abstract

Abstract To obtain CT images of the knee joint in a more lifelike position, data acquisition can be performed with patients in standing rather than in lying position. However, in that situation, people tend to show involuntary motion. One possibility to compensate for that motion is the use of Inertial Measurement Units, that capture the accelerations during the scan. For this purpose, their local coordinate system needs to be known. An estimation based on the SIFT algorithm was implemented and compared to an existing approach that uses the Fast Radial Symmetry transform and to expert labels for evaluation. The SIFT method showed to be superior to the existing approach as it could extract stable feature points from the projections that were used to estimate the three-dimensional coordinate system in a reliable manner. The final algorithm achieved a mean euclidean distance of 2.61 mm between the calculated position of the origin and the assumed ground truth by the expert labels.

Highlights

Cone-beam C-arm CT systems enable the scanning of patients in standing weight-bearing position, thereby giving useful insights into the knee joint for monitoring the progression of Osteoarthritis
Existing approaches for patient motion estimation propose to perform e.g. an image-based registration to a motionfree data set acquired in supine position [1], or to use external cameras that capture the surface of the knee during the scan
The origins of the proposed algorithm and the expert labels differ by a mean of 2.61 mm with a maximum of 4.62 mm, whereas the mean distance of reference approach to expert labels is larger with 11.79 mm and little variation over the different data sets

Summary

Introduction

Cone-beam C-arm CT systems enable the scanning of patients in standing weight-bearing position, thereby giving useful insights into the knee joint for monitoring the progression of Osteoarthritis. In order to obtain data usable for diagnosis in clinical applications, the patient’s motion has to be estimated for the reconstruction. Existing approaches for patient motion estimation propose to perform e.g. an image-based registration to a motionfree data set acquired in supine position [1], or to use external cameras that capture the surface of the knee during the scan [2]. Maier et al proposed to use Inertial Measurement Units (IMUs) for the purpose of motion estimation, where one single IMU attached to the knee is able to record the knee’s motion [8]

Methods

Results

Discussion

Conclusion