The aim of the study was to create an accurate method of automated subcutaneous (SAT) and visceral (VAT) adipose tissue detection basing on three-dimensional (3D) computed tomography (CT) scans.

One hundred and forty abdominal CT examinations were analysed. An algorithm for automated detection of SAT and VAT consisted of following steps: thresholding of an analysed image, detection of a patient’s body region, separation of SAT and VAT. The algorithm was sequentially applied to each 2D axial slice of a 3D examination. To assess the accuracy of the proposed method, automated and manual segmentations (performed by two readers) of SAT and VAT were compared using Dice similarity coefficient (DSC) and average Hausdorff distance (AHD).

Mean DSC was equal to 99.6% ± 0.4% for SAT and 99.6% ± 0.5% for VAT, which was equal to DSC obtained for comparison between both readers. In 90% of cases DSC was equal or above 99.0% and the minimal DSC was 97.6%. AHD equalled to 0.04 ± 0.06 for SAT and 0.13 ± 0.23 for VAT (automated vs. manual segmentations), while AHD for comparison of two manual segmentations was 0.03 ± 0.07 for SAT and 0.09 ± 0.20 for VAT. The processing time for a single slice was 0.16 s for an automated segmentation and 5−10 min for a manual segmentation. The processing time of an entire 3D stack (around 40 2D slices) was on average 6.5 s.

Our algorithm for the automated detection of SAT and VAT on 3D CT scans has the same accuracy as manual segmentation and performs equally well for both adipose tissue compartments.

该研究的目的是基于三维（3D）计算机断层扫描（CT）扫描创建一种准确的自动皮下（SAT）和内脏（VAT）脂肪组织检测方法。

分析了140例腹部CT检查。用于自动检测SAT和VAT的算法包括以下步骤：阈值化分析图像，检测患者的身体区域，SAT和VAT分离。该算法被依次应用于3D检查的每个2D轴向切片。为了评估该方法的准确性，使用Dice相似系数（DSC）和平均Hausdorff距离（AHD）比较了SAT和VAT的自动和手动分割（由两个读者执行）。

SAT的平均DSC等于99.6％±0.4％，VAT的平均DSC等于99.6％±0.5％，这等于两个阅读器之间进行比较所获得的DSC。在90％的情况下，DSC等于或高于99.0％，最低DSC为97.6％。SAT的AHD等于0.04±0.06，增值税的AHD等于0.13±0.23（自动与手动细分），而SAT的两个手动细分的比较的AHD分别为0.03±0.07和VAT的0.09±0.20。自动切片的单个切片处理时间为0.16 s，手动切片的处理时间为5-10分钟。整个3D堆栈（大约40个2D切片）的处理时间平均为6.5 s。

我们在3D CT扫描中自动检测SAT和VAT的算法与手动分割具有相同的准确性，并且在两个脂肪组织隔室中均表现出色。