Obesity, metabolic syndrome and diabetes are associated with an increased morbidity and mortality. The accumulation and regional distribution of abdominal adipose tissue and organ fat plays an important role in these diseases. Lifestyle changes such as exercise programs or dietary interventions leading to weight loss have a positive effect on preventing disease progression [1]. Therefore, there is a growing interest on investigating the association between weight loss and regional changes of different adipose tissue compartments, which may help to find optimal lifestyle intervention strategies for disease prevention in these patient groups. Chemical shift encoding-based water-fat MRI has emerged as a reliable method to measure fat content throughout the body [2]. However, the rapid and accurate extraction of abdominal adipose tissue volumes and organ fat content based on water-fat imaging requires a fully-automated and reliable image analysis methodology. The present work proposes a fully automatic multi-atlas-based approach for abdominal organs segmentation and adipose tissue measurement based on longitudinal water-fat MRI data.

Subjects:

Twenty obese women (age range: 24-65 years, BMI: 34.9 ± 3.8 kg/m²) were recruited for this study and underwent a defined dietary intervention with a total daily energy intake of 800 kcal and additional 200 g of vegetables for 28 days. All subjects underwent MR imaging one day before the start and one day after the end of the dietary intervention.

MR Imaging:

The abdominal region of the subjects was scanned on a 3.0 T MR scanner (Ingenia, Philips Healthcare) using anterior and posterior coil arrays. Axial two-point Dixon images based on a 3D spoiled gradient echo sequence were acquired using two stacks with identical imaging parameters: TR = 4.0 s, TE1/TE2 = 1.32/2.6 ms, flip angle = 10°, bandwidth = 1004 Hz/pixel, 332x220 acquisition matrix size, FOV = 500x446 mm², acquisition voxel = 1.5x2.0x5.0 mm³, 44 slices, parallel imaging using SENSE with a reduction factor R = 2.5. The two stacks were aligned to cover the entire abdominal region starting at the top of the liver. The acquisition time for each stack was 10.6 s and the scanning of each stack was performed in a single breath-hold. Water and fat images were separated online on the scanner using the mDixon algorithm [3].

MR Image Post-Processing:

Firstly, manual segmentation was performed to generate atlases as reference standard for the fully automatic segmentation algorithm.

Figure 1 shows the steps of the proposed segmentation algorithm. Step (I) included the manual generation of atlases. Step (II) included the procedure of registration-based segmentation. Each of preprocessed atlas dataset was co-registered with preprocessed patient dataset. The output deformation field transformed segmentations of interested organs to the patient spatial coordinate system. The registration consisted of affine and diffeomorphic deformation provided by Advanced Normalization Tools (ANTs) framework [4]. In this way, a group of normalized segmentation labels was automatically yielded. Step (III) included the label fusion method Selective And Iterative Method For Performance Level Estimation (SIMPLE) [5] that fused the normalized and resampled segmentation labels to compute the final segmentation.

Figure 2 shows the flow chart of the employed fat classification algorithm. K-means clustering was used to classify the adipose tissue within the abdomen, and morphological operations further separated abdominal adipose tissue into visceral and subcutaneous adipose tissue (VAT and SAT). With the use of vertebral bodies, SAT could be classified as anterior, posterior and total SAT. SAT and VAT compartments were computed across the feet-head direction using blocks of L1-L2, L2-L3, L3-L4, L4-L5 and L5-caudal. Organ-specific VAT was determined within 5 cm from the periphery of the kidneys, spleen, liver and pancreas.

Figure 3, 4 and 5 display representative organ, SAT, and VAT segmentations in patients before and after intervention.

The accuracy of organ segmentation represented by Dice coefficients ranged from 67.21% ± 15.51% for the pancreas to 94.25% ± 2.25% for the liver. SAT changes were significantly greater in the posterior than the anterior compartment (-11.4±5.1% versus -9.5±6.3%, p<0.001). SAT changes were greater in the L2-3 (total: -12.5±12.5%) and L3-4 (total: -11.8±12.4) regions and less pronounced in the L1-2 region (total: -6.6±13.5). However, these differences were not statistically significant (p>0.05). VAT loss located not around any organ (-16.1±8.9%) was statistically significant greater than VAT loss around liver, left and right kidney, spleen, and pancreas (p<0.05). A significant reduction of organ fat content (-2.7±3.3%, p=0.002) was only observed in the liver.A fully automatic multi-atlas-based approach was developed for abdominal organs segmentation and adipose tissue volume quantification based on water-fat MRI data. The developed fully automatic algorithm showed good performance in abdominal fat and organ segmentation, and allowed the detection of regional changes of adipose tissue depots in obese women undergoing a calorie restriction intervention.