Validation of the FXA Software

To ensure the highest possible precision and accuracy, several studies were performed exploring the efficacy and limits of the software.

Method
The accuracy in determining the centers of rotation (CoR) requires a sufficient resolution of theradiographs and depends on the matching accuracy for the CoR evaluation. This process is very sensitive to proper image correlation as small inaccuracies of 0.1mm lead to significant errors in the location of the CoR. The aim of this work was to assess the core accuracy of the software using synthetic images from a lumbar spine (L1-5). These images were generated using a 3-D CAD-model (SolidWorks 2010), thereby excluding artifacts from the image acquisition process such as noise or out-of-plane effects. Each vertebra in the CAD-model was rotated about pre-defined CoRs. Various rotation angles were produced (0.1°,0.5°, 1°, 2°, 3° and 5°). The FXA software determined RoM and CoR of each segment and the measurements were compared to their nominal values.

Results
The smallest input rotation angle (0.1°) resulted in a detected RoM of 0.10±0.03°. Accuracy increased for RoM≥1° with a SD of ±0.02°. Significant CoR deviations ranged from 2.7mm to 79.9mm to the ideal CoR location (0.0mm) for RoM=0.1°. Images with RoM=0.5° were found to yield a CoR location accuracy of 1.6±0.7mm. A mean CoR location offset of 0.9±0.7mm was found for RoM=1.0°. Stable CoR output was found for RoM≥2° that resulted in 0.6±0.3mm.

Method
The in vitro study was performed with n=6 spinal specimens (L3-4) taken from calves. The specimens were attached in a robot unit (KR125, Kuka, Augsburg). Subsequently, CoR were predefined at specimens using a cross laser level. The CoR was pinpointed using an x-ray marker. A baseline x-ray was taken at the unloaded position and the marker was removed. Specimens were moved to flexion/extension movements about the previously defined CoR at magnitudes of 0.1°, 0.5°, 1°, 2°, 3° and 5°. Lateral x-rays were taken at the end positions using a mobile x-ray machine (PX-15HF, Raytech Diagnostics, Canada) and digital memory foils (ADCC-MD-plate, Agfa). The x-ray source was placed at a distance of approx. 75 cm away from the specimens. The x-ray beam was aligned to target the center of the intervertebral disc. Movements were additionally recorded using an optical motion tracking system (Optotrak Certus, NDI, Radolfzell). X-ray images and motion tracking data were evaluated for RoM/CoR and subsequently compared.

Results
Moving the specimens to a RoM=0.1° resulted in recordings of 0.12°±0.04° with the motion tracking system and 0.14°±0.07° using the FXA method. With increasing RoM the measurement errors increased slightly for both methods. Maximum deviation was found for the largest RoM of 5°. The motion tracking system yielded 4.84°±0.16° and the x-ray analysis 4.85°±0.16°. There were no statistical differences detected between both methods. The maximum accuracy for the CoR location was found for a RoM≥2°.

Discussion
The aim of the study was to produce intersegmental positions and movements on x-ray films that were uniquely defined in RoM and CoR. For this reason, we employed a robot unit due to its capability of moving objects about a predefined CoR with a high reproducibility. The study confirmed the software accuracy established in the previous validation. The validation was maintained by two distinct methods (robot and motion tracking system) against which the FXA results were compared. Utilizing the automated functional x-ray analysis method, poly-segmental RoM, CoR evaluation and implant migration assessments can be conducted in daily practice. This could enhance the quality of clinical investigations and increase the scientific value of clinical studies.

Literature Review

The accuracy of orthopedic parameters obtained from radiological images has been evaluated and published in the past. In order to compare the different methods, the average error and the standard deviation of angular measurements was analyzed. A systematic measurement error is expressed by the average error, the quality and reproducibility of a measurement method is characterized by the standard deviation.
With the exception of the QMA software, all measurement methods showed no systematic measurement error (average error near zero).
The manual methods (angles are determined by drawing lines on the x-ray film or computer monitor) showed the largest scatter range with a standard deviation of 2°, which can be significantly improved with semi-automatic methods.
The FXA software showed a further reduced standard deviation of 0.13°, which is essential for the reliable determination of the location of the center of rotation. The theoretical accuracy of the FXA software can be expressed with a standard deviation of 0.02°.