Deep neural network for automatic volumetric segmentation of whole-body CT images for body composition assessment

Abstract

Background & aims: Body composition analysis on CT images is a valuable tool for sarcopenia assessment. We aimed to develop and validate a deep neural network applicable to whole-body CT images of PET-CT scan for the automatic volumetric segmentation of body composition. Methods: For model development, one hundred whole-body or torso 18 F-fluorodeoxyglucose PETeCT scans of 100 patients were retrospectively included. Two radiologists semi-automatically labeled the following seven body components in every CT image slice, providing a total of 46,967 image slices from the 100 scans for training the 3D U-Net (training, 39,268 slices; tuning, 3116 slices; internal validation, 4583 slices): skin, bone, muscle, abdominal visceral fat, subcutaneous fat, internal organs with vessels, and central nervous system. The segmentation accuracy was assessed using reference masks from three external datasets: two Korean centers (4668 and 4796 image slices from 20 CT scans, each) and a French public dataset (3763 image slices from 24 CT scans). The 3D U-Net-driven values were clinically validated using bioelectrical impedance analysis (BIA) and by assessing the model's diagnostic performance for sarcopenia in a community-based elderly cohort (n ¼ 522). Results: The 3D U-Net achieved accurate body composition segmentation with an average dice similarity coefficient of 96.5%e98.9% for all masks and 92.3%e99.3% for muscle, abdominal visceral fat, and subcutaneous fat in the validation datasets. The 3D U-Net-derived torso volume of skeletal muscle and fat tissue and the average area of those tissues in the waist were correlated with BIA-derived appendicular lean mass (correlation coefficients: 0.71 and 0.72, each) and fat mass (correlation coefficients: 0.95 and 0.93, each). The 3D U-Net-derived average areas of skeletal muscle and fat tissue in the waist were independently associated with sarcopenia (P < .001, each) with adjustment for age and sex, providing an area under the curve of 0.858 (95% CI, 0.815 to 0.901).

Impacts of body composition parameters and liver cirrhosis on the severity of alcoholic acute pancreatitis

“…Abdominal CT images were uploaded to commercially available deep learning-based software for body composition analysis (DeepCatch R v1.0.0.0; Medicalip Co. Ltd., Seoul). This software package provides automatic volumetric segmentation of the following seven body components with an accuracy of 97% [ 30 ]: skin, muscle, abdominal visceral fat, subcutaneous fat, bone, internal organs and vessels, and the central nervous system. In addition, the software provides automatic localization of the third lumbar vertebral body (L3), and automatically quantifies L3 sectional area (cm 2 ) and mean CT attenuations (Hounsfield Units, HU) of visceral abdominal fat, subcutaneous abdominal fat, and abdominal muscle components ( Fig 1 ).…”

Impact of Computed Tomography-Based, Artificial Intelligence-Driven Volumetric Sarcopenia on Survival Outcomes in Early Cervical Cancer

“…In brief, we used this deep neural network-based software for automatic volumetric segmentation of body composition (skeletal muscle, abdominal visceral fat, and subcutaneous fat) from anonymized, precontrast CT images in DICOM format. According to the previous validation study, the software's average segmentation accuracy was reported as 97% compared with manual segmentation (21). After segmentation, the abdominal waist was automatically labeled based on WHO's waist definition (29): between the lower end of the thoracic ribs and the upper end of the iliac crest.…”

“…Researchers have measured individuals' area of skeletal muscle and adipose tissue at the third lumbar vertebral body (L3)-level cross-sectional image of CT scans, which is known to reflect amounts of total body muscle and adipose tissue well (19,20). In addition, the latest high-throughput technology allows automated and fast volumetric measurements of each component from CT scans (21,22). With the use of such an advanced tool, tracking the volumetric change of specific body composition components is feasible (23), which has not yet been investigated in early cervical cancer.…”

Prognostic role of computed tomography-based, artificial intelligence-driven waist skeletal muscle volume in uterine endometrial carcinoma

“…To determine sarcopenia, these studies commonly measured skeletal muscle area from a single computed tomography (CT) scan image, based on previous findings that the third lumbar vertebra (L3)-level cross-sectional image reflects total body muscle mass and adipose tissues well [ 15 , 16 ]. Beyond the areal measurement, recent technological advances enable the volumetric measurement of a specific body composition component, such as skeletal muscle, visceral fat, and subcutaneous fat, from the CT scans that were not feasible due to the requirement of substantial time and human effort [ 17 ]. The volumetric measurement of body composition may contain more abundant and precise information than areal measurements in a single cross-sectional image [ 18 ].…”