Data from: Field-based adipose tissue quantification using bioelectrical impedance spectroscopy validated with CT scans and deep learning

Full description

Dataset used in the manuscript "Field-based adipose tissue quantification using bioelectrical impedance spectroscopy validated with CT scans and deep learning"

Software/equipment used to create/collect the data: (a) Bioelectrical impedance measurements: Bioelectrical impedance measurements were performed using a BIS device (SFB7, Impedimed, Brisbane, QLD, Australia), that measures resistance and reactance to an applied harmless, alternating electric current at 256 logarithmically-spaced frequencies in the range of 3-1000 kHz. Device calibration was verified daily. The extracted data of interest were resistance at infinite frequency (Rinf, predictor of total body water and FFM); resistance at zero frequency (R0, predictor of extracellular water); intracellular resistance (Ri, an index of intracellular water); and, for comparison with studies using single-frequency (50 kHz) impedance devices, resistance at 50 kHz (R50); reactance at 50 kHz (Xc50) and phase angle at 50 kHz (PhA50).

(b) Computed tomography scans: Volumetric data of total body scans were acquired in helical scan mode, with 1.25 mm slice thickness and spacing between slices set at 0.625 mm (Optima CT660 16 slice scanner, GE Medical Systems, Milwaukee, WI USA; and Aquilion Lightning 160, Canon, Tokyo, Japan). The typical number of CT slices per animal was around 1,600. In this study, the total number of slices was approximately 80,000, and approximately 50,000 CT slices were used in the final calculations.

Software/equipment used to manipulate/analyse the data: (a) Adipose tissue identification: Adipose tissue Hounsfield units (HU) (i.e., attenuation ranges) were identified by three-dimensional rendering using a commercially available, validated software for DICOM (Digital Imaging and Communications in Medicine) visualisation and body composition analysis (NovaPACS, NovaradTM, American Fork, USA), following the methods described in Gibby et al. (2017) and DePersio et al. (2019).

(b) Adipose tissue quantification: An open-sourced machine learning framework (PyTorch) was used for implementing the U-Net (Paszke et al., 2019), and was adapted from Yakubovskiy (2020).

Notes