European Journal of Radiology 72 (2009) 278–283 Contents lists available at ScienceDirect European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad Computer-assisted quantification of interstitial lung disease associated with rheumatoid arthritis: Preliminary technical validation K. Marten a , V. Dicken d , C. Kneitz c , M. Hoehmann b , W. Kenn b , D. Hahn b , C. Engelke a,∗ a Department of Radiology, Georg August University of Göttingen, Robert-Koch-Strasse 40, 37075 Göttingen, Germany b Department of Radiology, University Hospital of Würzburg, Josef-Schneider-Strasse 2, 97080 Würzburg, Germany c Department of Rheumatology and Clinical Immunology, Medizinische Klinik and Poliklinik, University Hospital of Würzburg, Klinikstrasse 6, 97070 Würzburg, Germany d MeVis Research GmbH, Universitätsallee 29, 28359 Bremen, Germany article info Article history: Received 1 April 2008 Received in revised form 6 July 2008 Accepted 7 July 2008 Keywords: Computer-aided diagnosis Interstitial lung disease Collagen vascular disorders abstract Purpose: To validate a threshold-based prototype software application (MeVis PULMO 3D) for quantifi- cation of chronic interstitial lung disease (ILD) in patients with rheumatoid arthritis (RA) using variable threshold settings for segmentation of diseased lung areas. Methods: Twenty-two patients with rheumatoid arthritis were included and underwent thin-section CT (4 × 1.25mm collimation). CT scans were assessed by two observers for extent of ILD (EoILD), and twice by MeVis PULMO 3D for each protocol. MeVis PULMO 3D used four segmentation threshold (ST) settings (ST = -740, -780, -800 and -840 HU). Pulmonary function tests were obtained in all patients. Statistical evaluation used 95% limits of agreement (LoA) and linear regression analysis. Results: There was total concordance between the software measurements. Interobserver agreement was good (LoA = -28.36 to 17.58%). EoILD by readers correlated strongly with DL CO (r = -0.702, p < 0.0001) and moderately with FVC (r = -0.523, p =0.018). There was close correlation between readers and MeVis PULMO 3D with best results for ST <780 HU (EoILD vs. MeVis PULMO 3D: r = 0.650 for ST = -800 and -840 HU, respectively; p = 0.002). MeVis PULMO 3D correlated best with DL CO at ST of -800 HU (r = -0.44, -0.49, -0.58 and -0.57 for ST = -740, -780, -800 and -840, respectively; p = 0.007–0.05) and moder- ately with FVC (r = -0.44, -0.51, -0.59 and -0.45 for ST = -740, -780, -800 and -840), respectively; p = 0.007–0.05). Conclusion: The MeVis PULMO 3D system used holds promise to become a valuable instrument for quan- tification of chronic ILD in patients with RA when using the threshold value of -800 HU, with evidence of the closest correlations, both with human observers and physiologic impairment. © 2008 Elsevier Ireland Ltd. All rights reserved. 1. Introduction Rheumatoid arthritis (RA) is a chronic inflammatory autoim- mune disorder that not only causes joint destruction, but also affects the lungs, heart and vascular endothelium. Pulmonary dis- ease occurs in about 50% of cases and may account for mortality in 18% of patients, coming second only to infections as a source of morbidity [1,2]. Thin-section chest CT constitutes a sensitive and reproducible method for detection and monitoring of interstitial lung disease (ILD) associated with RA [3,4]. The clinical course of these patients may vary: some individuals display a slowly pro- gressive course that may last for a decade, whereas others have a fulminant course leading to death within less than 6 months after the onset of respiratory symptoms [5–7]. In all patients with pul- ∗ Corresponding author. E-mail address: c.engelke@med.uni-goettingen.de (C. Engelke). monary involvement it is of paramount importance to accurately stage the extent of ILD and to monitor its progression, making non-invasive and reproducible scoring systems employed in high- resolution CT highly desirable. Computer-aided diagnosis (CAD) has long been recognized as a valuable means for improved performance and decision-making due to enhanced detection and evaluation of complex imaging features [8]. In thoracic imaging, the most powerful impetus for CAD has been automated detection and volumetry of pulmonary nodules [9]. Although less intensely evaluated, CAD of ILD offers inherent advantages, as widely accepted standardized systems for quantification of the extent of disease currently do not exist [8]. However, treatment selection and monitoring as well as the evaluation of novel treatment regimens require disease quantifi- cation based on non-invasive and reproducible scoring systems. Preliminary studies employing various computer-derived indices obtained from frequency histograms of thin-section CT scans of the lung as well as more sophisticated image analysis techniques 0720-048X/$ – see front matter © 2008 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2008.07.008