Markus B. Huber PhD, Presenter: Nothing to Disclose

Mahesh Nagarajan, Abstract Co-Author: Nothing to Disclose

Gerda Luise Leinsinger, Abstract Co-Author: Nothing to Disclose

Lawrence Ray PhD, Abstract Co-Author: Employee, Carestream Health, Inc

Sven E. Ekholm MD, Abstract Co-Author: Nothing to Disclose

Axel Wismueller MD, PhD, Abstract Co-Author: Research grant, Carestream Health, Inc

To classify morphological patterns (honey-combing) indicative of fibrotic interstitial lung disease in multi-detector computed tomography (MDCT) images with recently developed topological and geometrical texture features and to compare the results against standard texture features.

For five patients with known occurrence of honeycombing, a stack of 70 axial, lung kernel reconstructed images were acquired from MDCT chest exams. 964 regions of interest (23x23 pixels) of both health and pathological (356) lung tissue were identified by an experienced radiologist. Texture features were extracted using six properties calculated from gray-level co-occurrence matrices (GLCM), Minkowski Dimensions (MDs), three Minkowski Functionals (MFs, e.g. MF.euler) and Scaling Index Method (SIM). A fuzzy k-nearest-neighbor (k-NN) classifier and a Multilayer Radial Basis Functions Network (RBFN) were optimized in a 10-fold cross-validation for each texture feature, and the area under the ROC curve (AUC) was calculated on independent test sets as a measure of automated tissue characterization. A Wilcoxon signed-rank test was used to compare two AUC distributions including the Bonferroni correction.

The best classification results were obtained by the MF and SIM features, which performed significantly better than all the standard GLCM and MD features (p<0.005) for both classifiers. The highest AUC was found for the MF.euler (0.987±0.002, 0.980±0.005; for the k-NN and RBFN classifier, respectively) and slightly lower values for SIM (0.972±0.007, 0.974±0.007). The best standard texture feature, GLCM.homogeneity, had a significantly higher AUC (0.961±0.007, 0.963±0.012) than the other GLCM and MD features for both classifers (p<0.001).

The results indicate that advanced topological and geometrical texture features can provide superior classification performance in computer-assisted diagnosis of interstitial lung diseases when compared to standard texture analysis methods.

Pathological lung pattern classification using novel texture analysis approaches could improve the radiologist’s ability to accurately classify interstitial lung diseases.

Huber, M, Nagarajan, M, Leinsinger, G, Ray, L, Ekholm, S, Wismueller, A, Classification of Interstitial Lung Disease Patterns with Topological Texture Features.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8009183.html