Automated Segmentation of Abdominal Skeletal Muscle on Pediatric CT Scans Using Deep Learning

Abstract

S arcopenia, characterized by the loss of skeletal mass, has been linked to outcomes for various disease states and is a marker of overall health (1,2). A commonly used method of skeletal muscle mass quantification is to calculate the skeletal muscle mass index (SMI) in which the cross-sectional area of the skeletal muscle at the L3 vertebral level is normalized to patient height squared (SMI = skeletal muscle area/height 2) (3). The muscles included in this measurement are the psoas, paraspinal (erector spinae and quadratus lumborum), and abdominal wall muscles (external and internal abdominal obliques, transversus abdominis, and rectus abdominis) (example shown in Figure E1 [supplement]). Consensus normal ranges and cutoff values of SMI for the diagnosis of sarcopenia have been reported primarily for adult populations (1,2,4-7). For children, only a few studies have reported the diagnosis of sarcopenia using measurements of skeletal muscle mass, and the reference ranges used were derived from adult measurements without age-specific cutoff values (8,9). Availability of reference ranges from only adults is problematic, as skeletal muscle mass increases throughout childhood and adolescence and

End-to-end automated body composition analyses with integrated quality control for opportunistic assessment of sarcopenia in CT

“…Compared to previous research in the field of automated body composition analyses, we observed similar or superior performance values for slice extraction task and tissue segmentation in our study [12][13][14][15][16][17]. In previous work, the slice extraction task was formulated either as a regression problem, a classification task, or, similar to our approach, a segmentation problem [15][16][17]. While the methods proposed so far for slice extraction are based on 2D images or require the generation of a maximum intensity projection in a preprocessing step, the use of the nnU-Net framework allows the direct input of 3D CT datasets of different sizes.…”

“…Compared to previous research in the field of automated body composition analyses, we observed similar or superior performance values for slice extraction task and tissue segmentation in our study [12][13][14][15][16][17]. In previous work, the slice extraction task was formulated either as a regression problem, a classification task, or, similar to our approach, a segmentation problem [15][16][17].…”

“…While the methods proposed so far for slice extraction are based on 2D images or require the generation of a maximum intensity projection in a preprocessing step, the use of the nnU-Net framework allows the direct input of 3D CT datasets of different sizes. For tissue segmentation, different variants of a 2D U-Net architecture have been used [12,[15][16][17]. The CDFNet architecture applied in the current study is an extension of a Dense-UNet architecture with max-out activation units, which has recently also been successfully used for body composition analyses in magnetic resonance imaging [20].…”

“…However, manual interaction was still required for extraction of the single slice on which the automatic segmentation is performed. Only very recent work also includes deeplearning-based automated slice extraction as the next step for truly automated body composition analyses [15][16][17].…”

Deep generative models for automated muscle segmentation in computed tomography scanning

“…Lee et al [10] proposed a method for automatically segmenting muscles from CT cross-sectional images at the third lumbar level using deep learning; however, the muscles were treated as a lump and not individual segments. Castiglione et al [11] proposed a U-Netbased convolutional neural network model that could accurately identify the L3 levels and segment the skeletal muscle in pediatric CT scans. Kamiya et al [12] proposed an automatic segmentation method for the psoas major muscle using a shape model; however, it is difficult to apply this approach to the GMd because the periphery of the psoas major muscle is covered with adipose tissue, and the boundary is clear.…”