Field‐based adipose tissue quantification in sea turtles using bioelectrical impedance spectroscopy validated with CT scans and deep learning

• Published in: Ecology and evolution 2022-12, Vol.12 (12), p.e9610-n/a
• Main Authors: Kophamel, Sara, Ward, Leigh C., Konovalov, Dmitry A., Mendez, Diana, Ariel, Ellen, Cassidy, Nathan, Bell, Ian, Balastegui Martínez, María T., Munns, Suzanne L.
• Format: Article
• Language: English
• Subjects: Adipose tissue; Aquatic reptiles; Artificial intelligence; Automation; Bias; Bioelectricity; Bland–Altman; Body composition; body condition; Body fat; Body mass; Body size; Calibration; Computed tomography; Deep learning; Ecoinformatics; Ecological studies; Ecophysiology; Electric currents; Electrodes; Environmental protection; Environmental stress; Impedance; Impedance spectroscopy; Laboratory animals; Medical imaging; Nutrition assessment; Nutritional status; Performance prediction; Population; sea turtle; Sea turtles; Spectroscopy; Spectrum analysis; Vertebrates; Wildlife; Wildlife conservation; Wildlife management
• ISSN: 2045-7758; 2045-7758
• DOI: 10.1002/ece3.9610

Loss of adipose tissue in vertebrate wildlife species is indicative of decreased nutritional and health status and is linked to environmental stress and diseases. Body condition indices (BCI) are commonly used in ecological studies to estimate adipose tissue mass across wildlife populations. However, these indices have poor predictive power, which poses the need for quantitative methods for improved population assessments. Here, we calibrate bioelectrical impedance spectroscopy (BIS) as an alternative approach for assessing the nutritional status of vertebrate wildlife in ecological studies. BIS is a portable technology that can estimate body composition from measurements of body impedance and is widely used in humans. BIS is a predictive technique that requires calibration using a reference body composition method. Using sea turtles as model organisms, we propose a calibration protocol using computed tomography (CT) scans, with the prediction equation being: adipose tissue mass (kg) = body mass − (−0.03 [intercept] − 0.29 * length2/resistance at 50 kHz + 1.07 * body mass − 0.11 * time after capture). CT imaging allows for the quantification of body fat. However, processing the images manually is prohibitive due to the extensive time requirement. Using a form of artificial intelligence (AI), we trained a computer model to identify and quantify nonadipose tissue from the CT images, and adipose tissue was determined by the difference in body mass. This process enabled estimating adipose tissue mass from bioelectrical impedance measurements. The predictive performance of the model was built on 2/3 samples and tested against 1/3 samples. Prediction of adipose tissue percentage had greater accuracy when including impedance parameters (mean bias = 0.11%–0.61%) as predictor variables, compared with using body mass alone (mean bias = 6.35%). Our standardized BIS protocol improves on conventional body composition assessment methods (e.g., BCI) by quantifying adipose tissue mass. The protocol can be applied to other species for the validation of BIS and to provide robust information on the nutritional and health status of wildlife, which, in turn, can be used to inform conservation decisions at the management level. Loss of body fat (i.e., adipose tissue) in vertebrate wildlife species is indicative of decreased nutritional and health status and is linked to environmental stress and diseases. Here, we calibrate bioelectrical impedance spectroscopy (BIS) for assessin