A novel method of 3D image analysis of high-resolution cone beam CT and multi slice CT for the detection of semicircular canal dehiscence.

Hypothesis: We investigated if current-generation computed tomographic (CT) scanners have the resolution required to objectively detect bone structure defects as small as 0.1 mm. In addition, we propose that our method is able to predict a possible dehiscence in a semicircular canal.

Background: In semicircular canal dehiscence (SCD), the bone overlying the superior canal (SC) is partially absent, causing vertigo, autophony, hyperacusis or hearing loss. Diagnosis of SCD is typically based on multi-slice computed tomography (MSCT) images combined with the consideration of clinical signs and symptoms. Recent studies have shown that MSCT tends to overestimate the size of dehiscences and may skew the diagnosis towards dehiscence when a thin bone layer remains. Evaluations of CT scans for clinical application are typically observer based.

Methods: We developed a method of objectively evaluating the resolution of CT scanners. We did this for 2 types of computed tomography: MSCT, and cone beam computed tomography (CBCT), which have been reported to have a higher resolution for temporal bone scans. For the evaluation and comparison of image accuracy between different CT scanners and protocols, we built a bone cement phantom containing small, well-defined structural defects (diameter, 0.1-0.4 mm). These small inhomogeneities could reliably be detected by comparing the variances of radiodensities of a region of interest (i.e., a region containing a hole) with a homogenous region. The Fligner-Killeen test was used to predict the presence or absence of a hole (p ≥ 0.05). For our second goal, that is, to see how this technique could be applied to the detection of a possible dehiscence in a SC, a cadaveric head specimen was used to create an anatomic model for a borderline SCD; the SC was drilled to the point of translucency. After semi-automatically fitting the location of the canal, our variance-based approach allowed a clear, significant detection of the thin remaining bone layer.

Results: Our approach of statistical noise analysis on bone cement phantoms allowed us to distinguish real irregularities from measured image noise or reconstruction errors. We have shown that with computed tomography, an approach comparing radiodensity variance in regions of interest is capable of detecting inhomogeneities down to 0.1 mm (p ≤ 0.0001).

Conclusion: Our analysis of data from the cadaveric head specimen demonstrates that this approach can be used to objectively detect thin layers of bone overlying an SC. This should provide the basis for using this approach for a semi-automated, objective detection of SCD.