Related Experiment Videos
Intra- and Interobserver Variability of Cochlear Length Measurements in Clinical CT
John E Iyaniwura1, Mai Elfarnawany, Sadegh Riyahi-Alam
1*Biomedical Engineering Graduate Program †Department of Otolaryngology-Head and Neck Surgery ‡Department of Medical Imaging §Department of Medical Biophysics ||Department of Electrical and Computer Engineering, Western University, London, Ontario, Canada.
Hypothesis:
The cochlear A-value measurement exhibits significant inter- and intraobserver variability, and its accuracy is dependent on the visualization method in clinical computed tomography (CT) images of the cochlea.
Background:
An accurate estimate of the cochlear duct length (CDL) can be used to determine electrode choice, and frequency map the cochlea based on the Greenwood equation. Studies have described estimating the CDL using a single A-value measurement, however the observer variability has not been assessed.
Methods:
Clinical and micro-CT images of 20 cadaveric cochleae were acquired. Four specialists measured A-values on clinical CT images using both standard views and multiplanar reconstructed (MPR) views. Measurements were repeated to assess for intraobserver variability. Observer variabilities were evaluated using intra-class correlation and absolute differences. Accuracy was evaluated by comparison to the gold standard micro-CT images of the same specimens.
Results:
Interobserver variability was good (average absolute difference: 0.77 ± 0.42 mm) using standard views and fair (average absolute difference: 0.90 ± 0.31 mm) using MPR views. Intraobserver variability had an average absolute difference of 0.31 ± 0.09 mm for the standard views and 0.38 ± 0.17 mm for the MPR views. MPR view measurements were more accurate than standard views, with average relative errors of 9.5 and 14.5%, respectively.
Conclusion:
There was significant observer variability in A-value measurements using both the standard and MPR views. Creating the MPR views increased variability between experts, however MPR views yielded more accurate results. Automated A-value measurement algorithms may help to reduce variability and increase accuracy in the future.