Quantification of brown and white adipose tissue based on Gaussian mixture model using water-fat and T2* MRI in adolescents.

ABSTRACT

PURPOSE: To develop a technique for the separation and quantification of brown adipose tissue (BAT) and white adipose tissue (WAT) using fat fraction and T2* intensity based on the Gaussian mixture model (GMM). MATERIALS AND METHODS: Chemical-shift water-fat and T2* images were acquired at the neck, supraclavicular, interscapular, and paravertebral regions in 24 volunteers (Obese n = 12, female/male = 6/6, body mass index [BMI] = 31.3 ± 2.3 kg/m2 , age = 16.1 ± 0.6; Normal weight n = 12, female/male = 6/6, BMI = 21.2 ± 2.4 kg/m2 , age = 12.9 ± 2.4) using a 3T scanner with the chemical-shift water-fat mDixon sequence. BAT and WAT were clustered based on the Gaussian mixture model using the expectation-maximization algorithm. Results and reproducibility were compared and assessed using independent t-tests and intraclass correlation coefficient. RESULTS: BAT in obese participants was predominately found at the supraclavicular region and in normal-weight participants it was more scattered and distributed in interscapular-supraclavicular, axillary, and spine regions. Absolute volume of BAT was higher in the obese group (Obese 315.2 mL [±89.1], Normal weight 248.5 mL [±86.4]), but BAT/WAT ratios were significantly higher (P = 0.029) in the normal group. T2* of BAT (P = 0.04) and volume of WAT (P < 0.001) were significantly lower in the normals. Within-group comparison between male and female indicated no significant differences were found in volume (P = 0.776 (normal), 0.501 [obese]), T2* (P = 0.908 [normal], 0.249 [obese]) and fat-fraction of BAT (P = 0.985 [normal], 0.108 [obese]). The intraclass correlation coefficient showed a good reproducibility in volume (BAT 0.997, WAT 0.948), T2* (BAT 0.969, WAT 0.983), and fat-fraction (BAT 0.952, WAT 0.517). CONCLUSION: BAT identified by this method was in agreement with other studies in terms of location, fat-fraction value, and T2* intensity. The proposed GMM-based segmentation could be a useful nonradiation imaging method for assessment of adipose tissue, in particular for serial follow-up of volume changes after drug or lifestyle interventions for obesity. LEVEL OF EVIDENCE 2 Technical Efficacy Stage 1 J. MAGN. RESON. IMAGING 2017;46758-768.