General Methodology for Accurate MRI Abdominal Adipose Tissue Quantification

INTRODUCTION Fast and accurate quantification of abdominal adipose tissues has been a challenge. Traditionally, non-water-saturated (NWS), T1-weighted (T1W) techniques have been used for abdominal fat quantification on humans. Although fat signal is usually significantly higher than non-fat tissues, image quality is complicated by the relatively low signal-to-noise ratio and/or contrast-to-noise ratio, motion blurring, and other artifacts inherent to the abdomen and viscera such as blood flow and peristalsis. In addition, the signal intensity of different non-fatty tissues might be different due their own T1 and T2 differences. Despite the difficulties, a few fast fat quantification methods based on image gray-level histogram analysis have been proposed. These methods include manual, semi-automated, and fully-automated thresholding to identify fat voxels (1-3). Other approaches include curve fitting on image histogram to estimate total number of fat voxels (4). All these methods are based on histogram analysis, assuming that fat and non-fat tissue voxels can be successfully labeled solely based on their signal intensity. This approach, however, is inherently flawed because severe partial volume effects in human abdomen MR images are not taken into account leading to large systematic errors and variability, particularly during visceral adipose tissue (VAT) amount quantification. Here, we propose a general framework for VAT quantification based on a combination of fuzzy c-means (FCM) clustering and signal intensity thresholding, so that both full-volume (FV) and partial-volume (PV) fat amounts can be accurately quantified.