Original Research Automatic Correction of Intensity Inhomogeneities Improves Unsupervised Assessment of Abdominal Fat by MRI Vincenzo Positano, MSc, 1,2 * Kenneth Cusi, MD, 3 Maria Filomena Santarelli, PhD, 1,2 Annamaria Sironi, MD, 1,2 Roberta Petz, MSc, 1,2 Ralph DeFronzo, MD, 3 Luigi Landini, PhD, 2,4 and Amalia Gastaldelli, PhD 1–3 Purpose: To demonstrate that unsupervised assessment of abdominal adipose tissue distribution by magnetic reso- nance imaging (MRI) can be improved by integrating auto- matic correction of signal inhomogeneities. Materials and Methods: Twenty subjects (body mass index [BMI] 23.7– 44.0 kg/m 2 ) underwent abdominal (32 slices) MR imaging with a 1.9T Elscint Prestige scanner. Many images were affected by relevant intensity distortions. Unsupervised segmentation of subcutaneous adipose tissue (SAT) and vis- ceral adipose tissue (VAT) was performed by a previously validated algorithm exploiting standard fuzzy clustering seg- mentation. Images were also processed by an improved ver- sion of the software, including automatic correction of inten- sity inhomogeneities. To assess the effectiveness of the two methods SAT and VAT volumes were compared with manual analysis performed by a trained operator. Results: Coefficient of variation between manual and unsu- pervised analysis was significantly improved by inhomogene- ities correction in SAT evaluation. Systematic underestima- tion of SAT was also corrected. A less important performance improvement was found in VAT measurement. Conclusion: The results of this study suggest that the com- pensation of signal inhomogeneities greatly improves the ef- fectiveness of the unsupervised assessment of abdominal fat. Correction of intensity distortions is important in SAT evalu- ation and less significant in VAT measurement. Key Words: abdominal fat; image processing; fuzzy clus- tering; MRI J. Magn. Reson. Imaging 2008;28:403– 410. © 2008 Wiley-Liss, Inc. OBESITY IS AN IMPORTANT risk factor for the devel- opment of cardiovascular and metabolic diseases (1). Many studies have recognized the importance of differ- ent locations of adipose tissue depots, in particular, visceral adiposity (i.e., the amount of fat deposited around the internal organs, VAT). However, VAT can be measured accurately only by imaging techniques, since waist circumference, often used as index of abdominal adiposity in the clinical setting, is associated, but not as well as VAT, with the risk of cardiovascular disease (CVD) (2). Another important index is the ratio between visceral and subcutaneous adipose tissue (SAT) that is associated with the development of all features of the metabolic syndrome, accompanying insulin resistance and CVD (2,3). Therefore, detection and quantification of VAT and SAT is a crucial issue for identifying sub- jects with abdominal obesity-related risks. Imaging techniques are certainly the most precise and reliable methods for a qualitative and quantitative SAT and VAT analysis. Although several imaging meth- ods have been proposed for the assessment of abdom- inal fat distribution (4), magnetic resonance imaging (MRI) represents one of the safer and more accurate noninvasive techniques. The most common approach is the analysis of the area of a single abdominal slice that gives the advantage of simple acquisition and analysis (5). However, slice location affects the amounts of SAT and VAT measured, and the correlations to the total volumes of SAT and VAT (6). Thus, accurate determi- nation of SAT and VAT requires multislice imaging and analysis (7). Fully manual analysis of 3D datasets is time-consuming and operator-dependent, so that it is important to develop computer assisted methods. Starting from the seminal work by Lancaster et al (8), several image analysis techniques were proposed to ad- dress this issue. Semiautomated methods speed up the analysis and reduce operator dependence (7,9). Auto- matic methods remove the operator bias and, if accu- rately performed, increase the reproducibility and re- duce the time required for the analysis (10 –13). In particular, an unsupervised methodology based on a fuzzy clustering approach was demonstrated to allow processing of 3D MRI datasets in a short time without 1 Institute of Clinical Physiology, CNR, Pisa, Italy. 2 Fondazione Toscana Gabriele Monasterio, Pisa, Italy. 3 Division of Diabetes, Department of Medicine, University of Texas Health Science Center at San Antonio, San Antonio, Texas. 4 Department of Information Engineering, University of Pisa, Pisa, Italy. *Address reprint requests to: V.P., Institute of Clinical Physiology, CNR, Via Moruzzi, 1, 56124, Pisa, Italy. E-mail: positano@ifc.cnr.it Received September 14, 2007; Accepted April 7, 2008. DOI 10.1002/jmri.21448 Published in Wiley InterScience (www.interscience.wiley.com). JOURNAL OF MAGNETIC RESONANCE IMAGING 28:403– 410 (2008) © 2008 Wiley-Liss, Inc. 403