Original Research Automated Segmentation of Visceral and Subcutaneous (Deep and Superficial) Adipose Tissues in Normal and Overweight Men Suresh Anand Sadananthan, PhD, 1,2 Bhanu Prakash, KN, PhD, 3 Melvin Khee-Shing Leow, MMed, PhD, 1,4 Chin Meng Khoo, MBBS, 5 Hong Chou, MMed, 6 Kavita Venkataraman, MBBS, PhD, 2,7 Eric Y.H. Khoo, MBChB, 5 Yung Seng Lee, MMed, PhD, 1,8 Peter D. Gluckman, DSc, 1 E. Shyong Tai, MBChB, PhD, 1,5 and S. Sendhil Velan, PhD 1,3,9 * Purpose: To develop an automatic segmentation algo- rithm to classify abdominal adipose tissues into visceral fat (VAT), deep (DSAT), and superficial (SSAT) subcutane- ous fat compartments and evaluate its performance against manual segmentation. Materials and Methods: Data were acquired from 44 normal (BMI 18.0–22.9 kg/m 2 ) and 38 overweight (BMI 23.0–29.9 kg/m 2 ) subjects at 3T using a two-point Dixon sequence. A fully automatic segmentation algorithm was developed to segment the fat depots. The first part of the segmentation used graph cuts to separate the subcutane- ous and visceral adipose tissues and the second step employed a modified level sets approach to classify deep and superficial subcutaneous tissues. The algorithmic results of segmentation were validated against the ground truth generated by manual segmentation. Results: The proposed algorithm showed good perform- ance with Dice similarity indices of VAT/DSAT/SSAT: 0.92/0.82/0.88 against the ground truth. The study of the fat distribution showed that there is a steady increase in the proportion of DSAT and a decrease in the propor- tion of SSAT with increasing obesity. Conclusion: The presented technique provides an accu- rate approach for the segmentation and quantification of abdominal fat depots. Key Words: visceral fat; deep subcutaneous fat; superfi- cial subcutaneous fat; graph cut; level sets J. Magn. Reson. Imaging 2015;41:924–934. V C 2014 Wiley Periodicals, Inc. INCREASED LEVELS OF ADIPOSITY are associated with elevated risk of metabolic diseases including type 2 diabetes and cardiovascular disease (1,2). Adipose tissues exert different physiologic effects based on their anatomical location (2–4). In the abdomen, the adipose tissue compartment is divided into visceral (VAT) and subcutaneous adipose tissues (SAT). The SAT is separated into deep (DSAT) and superficial subcutaneous adipose tissue (SSAT) compartments by a fascial plane known as fascia superficialis (Fig. 1). These adipose depots exhibit differences in terms of their 1) anatomical location and venous drainage; for example, VAT drains directly into the hepatic portal vein; 2) adipocyte size and number; and 3) metabolic activity (4). With the advancement of magnetic reso- nance (MR) technology in imaging the adipose tissues, there is an increasing interest in estimating and quantifying the volume of these tissues to provide a better indicator of metabolic and cardiovascular risk factors than total fat volume. The ability to image the volume of fat in various depots has led to the observation that VAT is more strongly associated with metabolic and cardiovascular risks than SAT (1,2). A recent study that examined fat 1 Singapore Institute for Clinical Sciences, Agency for Science, Tech- nology & Research (A*STAR), Singapore. 2 Department of Obstetrics & Gynaecology, Yong Loo Lin School of Medicine, National University of Singapore and National University Health System, Singapore. 3 Singapore Bioimaging Consortium, Agency for Science, Technology & Research (A*STAR), Singapore. 4 Department of Endocrinology, Tan Tock Seng Hospital, Singapore. 5 Department of Medicine, Yong Loo Lin School of Medicine, National University of Singapore and National University Health System, Singapore. 6 Department of Diagnostic Radiology, Khoo Teck Puat Hospital, Singapore. 7 Saw Swee Hock School of Public Health, National University of Sin- gapore and National University Health System, Singapore. 8 Department of Pediatrics, Yong Loo Lin School of Medicine, National University of Singapore and National University Health System, Singapore. 9 Clinical Imaging Research Centre, Agency for Science, Technology & Research (A*STAR), Singapore. Contract grant sponsor: National Medical Research Council Transla- tional and Clinical Research Flagship Programme; Contract grant number: NMRC/TCR/004. *Address reprint requests to: S.S.V., Singapore Bioimaging Consor- tium, 11 Biopolis Way, #02-02, Singapore 138667. E-mail: sendhil_velan@sbic.a-star.edu.sg Received September 13, 2013; Accepted April 17, 2014. DOI 10.1002/jmri.24655 View this article online at wileyonlinelibrary.com. JOURNAL OF MAGNETIC RESONANCE IMAGING 41:924–934 (2015) V C 2014 Wiley Periodicals, Inc. 924