Adrien Raphael Depeursinge PhD, Presenter: Nothing to Disclose

Anne Shu-Lei Chin MD, Abstract Co-Author: Nothing to Disclose

Ann N. C. Leung MD, Abstract Co-Author: Nothing to Disclose

Glenn Rosen PhD, Abstract Co-Author: Nothing to Disclose

Daniel L. Rubin MD, MS, Abstract Co-Author: Nothing to Disclose

A confident CT diagnosis of classic usual interstitial pneumonia (UIP) can eliminate the need for an invasive surgical biopsy to confirm this diagnosis. This task is often challenging, particularly in less specialized practice centers without access to experts experienced in interstitial lung disease. We propose a novel computational approach for the automated classification of classic versus atypical UIP. A score is derived from regional volumetric texture analysis of CT images.

CT examinations of 33 patients with biopsy proven UIP from -anonymous- were retrospectively reviewed in this study. Two thoracic radiologists with more than 15 years of experience worked in consensus to classify each patient as classic (15 patients) versus atypical (18 patients) UIP based on the American Thoracic Society guidelines. The responses of 3-D wavelets are localized using a simple digital atlas of 36 subregions of the lungs. The decision function of support vector machines (SVM) trained in a feature space spanned by the regional texture features is used as a score to stratify patients with UIP into classic and atypical subtypes. Receiver operating characteristics (ROC) analysis was used to evaluate the ability of the score to discriminate between classic versus atypical UIP.

An area under the ROC curve (AUC) of 0.81 was obtained using a leave-one-patient-out cross-validation, with high specificity for classic UIP. We compared this performance with a global characterization of the volumetric texture properties of the lungs, which led to an AUC of 0.72. This highlighted the importance of localizing tissue texture properties, which is consistent with the medical knowledge.

We propose a novel computational method for the automated classification of classic versus atypical UIP based on regional volumetric texture analysis. Overall, the proposed approach successfully predicts UIP subtypes for more than 4 out of 5 patients (AUC=0.81) with high specificity for classic UIPs.

With further validation, our system may be useful in the clinical setting for identifying patients with classic UIP for which an unnecessary surgical biopsy can be avoided.

Depeursinge, A, Chin, A, Leung, A, Rosen, G, Rubin, D, A Computerized Score for the Automated Differentiation of Usual Interstitial Pneumonia from Regional Volumetric Texture Analysis.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14015198.html