Automated Classification of Usual Interstitial Pneumonia Using Regional Volumetric Texture Analysis in High-Resolution Computed Tomography

Abstract

Objectives We propose a novel computational approach for the automated classification of classic versus atypical usual interstitial pneumonia (UIP). Materials and Methods 33 patients with UIP were enrolled in this study. They were classified as classic versus atypical UIP by a consensus of two thoracic radiologists with more than 15 years of experience using the American Thoracic Society evidence–based guidelines for CT diagnosis of UIP. Two cardiothoracic fellows with one year of subspecialty training provided independent readings. The system is based on regional characterization of the morphological tissue properties of lung using volumetric texture analysis of multiple detector CT images. A simple digital atlas with 36 lung subregions is used to locate texture properties, from which the responses of multi-directional Riesz wavelets are obtained. Machine learning is used to aggregate and to map the regional texture attributes to a simple score that can be used to stratify patients with UIP into classic and atypical subtypes. Results We compared the predictions based on regional volumetric texture analysis with the ground truth established by expert consensus. The area under the receiver operating characteristic curve of the proposed score was estimated to be 0.81 using a leave-one-patient-out cross-validation, with high specificity for classic UIP. The performance of our automated method was found to be similar to that of the two fellows and to the agreement between experienced chest radiologists reported in the literature. However, the errors of our method and the fellows occurred on different cases, which suggests that combining human and computerized evaluations may be synergistic. Conclusions Our results are encouraging and suggest that an automated system may be useful in routine clinical practice as a diagnostic aid for identifying patients with complex lung disease such as classic UIP, obviating the need for invasive surgical lung biopsy and its associated risks.

Subacute and Chronic Left Ventricular Myocardial Scar: Accuracy of Texture Analysis on Nonenhanced Cine MR Images

“…Because traditional visual inspection of images might not allow for recognizing subtle differences in textural information, the application of TA has the potential to further enhance the diagnostic and prognostic value of imaging. The increase of interest in evaluating TA in radiology is reflected by recent literature describing potential applications of the technique in oncologic (11)(12)(13), neuroradiologic (14), musculoskeletal (15,16), and lung imaging (17). To our knowledge, only one preliminary study, published as a conference abstract, has evaluated the potential of TA for cardiac MR imaging so far (18).…”

Optimized steerable wavelets for texture analysis of lung tissue in 3-D CT: Classification of usual interstitial pneumonia

“…The performance improvement is large when compared to our previous work: AUC = 0.811 in [9] using the same cohort of patients. It can provide valuable computer-aided assistance in less specialized practice centers without access to experts experienced in interstitial lung disease, as well as avoiding unnecessary surgical biopsies.…”

“…The pulmonary volumes were separated into 36 subregions obtained from all intersections between apical, vertical center, basal, peripheral, left, right, anterior and posterior lung regions [9]. The average energies of the coefficients from each subband E(R RC i {ψ m * f }(x k )) were computed for each subregion k and concatenated to form the feature vector v, where i = 1, .…”

“…2. Construction of the feature vector v from regional averaging of the energies E(R RC i {ψ m * f }(x)) [9]. that the Papadakis wavelet [13] with a radial bandwidth of [3π/10, π] outperformed other proposed wavelets like Simoncelli [6], Meyer [16] that have radial bandwidth of [π/4, π], we designed a family of maximally localized wavelets over the radial bandwidth [Ω, π] for Ω ∈ [π/4, π/2] to simultaneously find the optimal bandwidth.…”

“…The latter are characterized by local morphological distortions of the lung parenchyma. Second, the tissue types must be located in the lung anatomy (e.g., lung bases and periphery), since their presence only is not specific to the classic UIP pattern [9]. The image operators must therefore have a spatial support that is large enough to differentiate the tissue types and yet small enough to reduce the influence of the lung boundaries in both basal and peripheral regions (see Fig.…”