Determination of astaxanthin concentration in Rainbow trout (Oncorhynchus mykiss) by multispectral image analysis

Astaxanthin is the single most expensive constituent in salmonide fish feed. Therefore control and optimization of the astaxanthin concentration from feed to fish is of paramount importance for a cost effective salmonide production. Traditionally, methods for astaxanthin determination include extraction of astaxanthin from the minced sample into a suitable solvent such as acetone or hexane before further analysis.

The existing methods have several drawbacks including being destructive and labour consuming. Current state-of-the art vision systems for quality and process control in the fish processing industries are typically based on traditional trichromatic (Red Green Blue) imaging. The relative presence of some wavelengths and absence of others is a specific characteristic of many material properties.

Consequently, the adaption of multispectral imaging technology can reveal relevant information and measurement of more biological quality parameters such as fat, astaxanthin and cartilage content, simultaneously. A multispectral image may also be referred to as a surface chemistry map where a set of neighbouring spectra are recorded, revealing information about the surface chemistry to a larger degree than in a trichromatic image.

In this study multispectral imaging has been evaluated for characterization of the concentration of astaxanthin in rainbow trout fillets. Rainbow trout’s (Oncorhynchus mykiss), were filleted and imaged using a rapid multispectral imaging device. The multispectral imaging device captures reflection properties in 19 distinct wavelength bands.

Subsequently, the astaxanthin concentration was determined by a traditional chemical method. The astaxanthin concentration of the analysed samples ranged from 0.20 to 4.34 ppm. In total 7 samples were detected as outliers and removed from the data set before further analysis. A partial least squares regression (PLSR) model was build to predict the astaxanthin concentration from novel images.

The obtained model was evaluated with a test set. The root mean square error of prediction obtained from the test set was 0.27 ppm and a goodness of fit of 0.86. The PLSR model made it possible to predict the astaxanthin concentration in each pixel of the image – surface chemistry map - and thereby show the astaxanthin distribution in the fillet.

The projected images clearly show a difference in astaxanthin distribution, showing that the upper part of the fillet contains the highest concentration of astaxanthin. This study has shown that multispectral imaging is a promising method for rapid and non-destructive analysis of astaxanthin concentration of rainbow trout, and thereby a qualified candidate for replacement of traditional laborious and destructive analysis of the astaxanthin concentration.