Ruediger Egbert Schernthaner MD, Presenter: Nothing to Disclose

Christoph Arnoldner, Abstract Co-Author: Nothing to Disclose

Marius Erik Mayerhoefer, Abstract Co-Author: Nothing to Disclose

Christian Czerny MD, Abstract Co-Author: Nothing to Disclose

To determine the value of texture analysis for the differentiation of cholesteatoma and middle ear inflammation in non-enhanced high-resolution (HR-) CT images.

25 patients with histologically proven cholesteatoma and 25 consecutive patients with severe middle ear inflammation were included in the study. Fisher coefficients were used to select an optimized subset of 10 texture features extracted from a manually segmented region of interest within the lesions. For lesion classification, k-nearest neighbor classification using the leave-one-out cross-validation technique were performed for “raw” texture features, for texture features after principal component analysis (PCA) and for texture features after linear discriminant analysis (LDA). Postoperative histology served as standard of reference; routine CT reports were used for comparison with the results of texture-based lesion classification.  

Seven out of 25 cholesteatomas (28%) were misinterpreted as inflammation on the routine CT report, and in another three patients (12%), the report was inconclusive. In 9 of the 25 patients with middle ear inflammation (36%), cholesteatoma could not be excluded as differential diagnosis on the routine CT report. In summary, 38% of the routine CT reports (n=19) were inconclusive or wrong. Conversely, based on the raw texture features, only five cholesteatomas (20%) were misclassified as inflammation, and only four inflammations (16%) were categorized as cholesteatoma. Thus, the overall misclassification rate dropped to 18% (n=9) using raw texture analysis. PCA and LDA were less successful than raw data analysis for the differentiation of cholesteatoma and inflammation, but were still superior to the routine CT reports, with misclassification rates of 20% (n=10) and 32% (n=16), respectively.  

Texture-analysis improves the differentiation of cholesteatoma and middle ear inflammation in non-enhanced HR-CT images.

HR-CT as the standard imaging technique for the temporal bone cannot adequately differentiate between cholesteatoma and inflammation; texture analysis improves lesion classification in HR-CT images.

Schernthaner, R, Arnoldner, C, Mayerhoefer, M, Czerny, C, Texture Analysis for the Differentiation of Cholesteatoma and Middle Ear Inflammation at Non-enhanced High-Resolution CT.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12036041.html