Maximising animal health and welfare is an important breeding objective for aquaculture industries. Cardiac morphology (ventricular size and shape) and epicardial fat (EF) coverage are known to be associated with the health and performance of Atlantic salmon. The degree of EF coverage in salmon is currently measured using a visual scoring system (Grade 1 to 3), which is a labour-intensive, and subjective categorical trait which lacks the precision and resolution of a truly quantitative trait. This can be limiting for statistical analyses and genetic evaluations. The aims of this study were therefore to 1) evaluate the use of image analysis to estimate EF percentage (EF%) and other cardiac morphometric traits in Atlantic salmon, and 2) estimate the genetic parameters of these cardiac traits.
Phenotypic development:
Ground truth measurements of EF% were obtained from 71 salmon hearts with a wide distribution of EF coverage using computer tomography (CT). The same 71 hearts were imaged in four planes (right and left lateral, caudal, and dorsal) using a digital SLR camera. Image processing involved background removal, contrast enhancement, and Gaussian filtering. EF% was then estimated using 1) grayscale conversion and binary thresholding, and 2) calculation of the average value of each of the red, green, and blue colour channels. Cardiac morphometric traits (heart size, ventricular height to width ratio, and internal heart angles) were estimated from a binary silhouette of the dorsal view.
We found that the best prediction of EF% (r = 0.80) was achieved using data from 3 images (left and right lateral and caudal views) and binary thresholding. Best results from a single image were achieved using binary thresholding applied to the dorsal view (r = 0.75).
Genetic analysis:
A categorical EF score and an image of the dorsal aspect of the heart were available for each of 1071 animals with known pedigree. Genetic parameters were estimated for heart size, EF%, EF score, ventricular height to width ratio, and the three internal heart angles (α1, α2 and α3), using univariate (variance components and heritability) and bivariate (phenotypic and genetic correlations) animal models.
We found that all traits except the angle α1 (h2 = 0.04 ± 0.04) were moderately heritable (0.16 ± 0.05 ≤ h2 ≤ 0.37 ± 0.07). Encouragingly, the genetic correlation between image-predicted EF% and categorical EF score was 0.98 ± 0.03, indicating that from a breeding perspective they are the same trait. Furthermore, the heritability of the image-predicted EF% was greater than that of the manual heart fat score (h2 = 0.28 ± 0.04 and 0.32 ± 0.07, respectively), suggesting that the more quantitative phenotyping approach captures more of the genetic variation of the trait. This result coupled with the possibility of evaluating other heritable cardiac traits leads us to conclude that image analysis is a valuable high-throughput phenotyping method for measuring cardiac traits in Atlantic salmon.