NordForsk Machine learning applied to predicting and preventing production loss in aquaculture

Hvert år dør opptil 20 % av laksefisk I oppdrett før slakt. Episoder med dødelighet skjer under stressende hendelser og intervensjoner som avlusning, forringet vannkvalitet og transport til slakt. Stor fisk som dør under stressende prosedyrer før slakt er av spesielt bekymringsfullt, med tanke på investeringene som kreves for å få fisken opp til denne størrelsen. Slik dødelighet er et alvorlig velferdsspørsmål og utgjør en betydelig hindring for bærekraft, etikk og økonomi i den nordiske havbruksnæringen. Foreløpig er årsakene til stressrelatert dødelighet usikre, men stadig flere bevis tyder på at flertallet dør av hjertesvikt på grunn av unormale hjerteformer eller andre typer hjertesykdommer. Vi har i dette prosjektet utviklet kunnskap om hvordan hjertemorfologi kan knyttes til hjertesykdom og stressindusert dødelighet. Dermed kan unormal hjerteform representere en ny og lovende indikator på hjertefunksjon, ytelse og risiko for dødelighet. Vi har også utviklet nye verktøy for vurdering av hjertemorfologi manuelt og ved maskinsyn og er nå i stand til å utføre automatisk vurdering av hjertemorfologi. Dette verktøyet har vi blant annet brukt til å sammenlikne hjertemorfologi på tvers av produsenter og produksjonsland (Norge og Færøyene) og vil kunne videreutvikles sånn at det kan brukes som beslutningsstøtteverktøy i oppdrettsnæringen.

We have created a computer vision model for automating the measurement and detection of deviations in fish heart morphology. This can replace manual measurements to extract features such as angles and lengths from heart images, which is a time-consuming process that require training and alignment of personnel to ensure consistent measurements. Besides enabling faster and more consistent measurement of heart morphology, the measurements from the computer vision model will also be used as input to probabilistic causal models that connect the heart morphology to heart function, the influence of environmental factors, and the effect on fish mortality and biological performance. We have also developed new standardized methods for assessing heart morphology in salmon. First, we have developed new nomenclature for important reference points on the pyramidal fish heart. We use this nomenclature 1) to standardize and more easily describe existing and new quantitative measures of heart morphology and 2) in the development of a qualitative scoring system that includes the most common phenotypic traits observed on the hearts of Norwegian farmed salmon. Many of these traits represent deviations from the wild salmon heart. These methods can be used in research and by the aquaculture industry and will facilitate further work to understand heart disease in farmed salmon. Lastly, we have developed tools to assess cardiac function to be able to establish standardized indicators of cardiac function. In short, we show that deviating heart morphology affects cardiac function in Atlantic salmon. More precicely, heart rate, hemodynamics and contractility are altered in deviating hearts. In addition, specific deviating traits have distinct impacts on cardiac function. In addition, we have tested if ECG can also provide a simple and non-lethal test for evaluating fish heart health status, and connected that to cardiac morphology. ECG data from the present work offers a promising screening tool that can be used to diagnose heart disease of farmed salmonids. It has, therefore, the potential to be used by farmers to take preventive measures, improve fish welfare, and minimize costly production losses.

Every year, up to 20% of farmed salmonids die before slaughter. Episodes of mortality happen during stressful incidents and interventions such as parasite treatment, deteriorating water quality, and transport to slaughter. Large fish that die during stressful procedures before slaughter is of particular concern, considering the investment required for growing fish to this size. Such mortality is a serious welfare issue and poses a considerable obstacle to sustainability, ethics and economy of the Nordic aquaculture industry. Currently, the causes of stress-related mortality are uncertain, but an increasing body of evidence suggests that the majority die from heart failure due to abnormal heart shapes or other types of heart disease. We have access to preliminary data that link cardiac morphology to cardiac morbidity and stress-induced mortality. Thus, abnormal heart shape may represent a novel and promising indicator of heart function, performance and risk of mortality. There are likely several causes underlying development of heart disease in farmed salmonids. Similar to humans, development of heart disease in fish may be linked to factors such as diet, physical activity, physical condition and stress from the environment. The DigiHeart project, a consortium of researchers and industry partners form Norway, Sweden and the Faroe Island will develop technology and control systems that use machine learning to continuously survey data about all these factors, in addition to operational and environmental conditions to identify causes behind heart disease and mortality in farmed salmonids. In addition, the project will develop on-site tools to evaluate mortality risk prior to stressful operational practices on fish farms. Thus, by identifying factors leading to heart disease and mortality and developing tools that can predict performance and mortality, the consortium aims at improving sustainability and animal welfare and reducing mortality in salmonid aquaculture.