Lepeophtheirus salmonis has always plagued and parasitized wild salmonids, but the aquaculture boom has given this tiny louse exceptionally good living conditions. Despite years of intense work, research, preventative measures and regulations, this highly resilient salmon louse remains a major threat to wild fish stocks and one of the salmon industry’s biggest problems.

Researchers from the Norwegian University of Science and Technology (NTNU) and Wageningen University in the Netherlands have now developed a new method that could provide better control over these parasites. Using real-life images of louse larvae in the sea, the production of synthetic data (real data edited and pasted together in new ways to increase the amount of data) and artificial intelligence, they have developed large datasets that AI models can be trained on to detect salmon lice. In a new study, the researchers demonstrate the method, highlighting how it can make lice detection much more efficient and effective.

The latest study shows that trained biologists needed more than 30 hours over several days to identify 82 per cent of the salmon lice larvae in one large and complex seawater sample. The AI model developed by the researchers took just 30 minutes to identify 97.5 per cent of the larvae in the same sample.

“A number of different measures are being used and tested to combat salmon lice. It is often the combined effect of several measures that will best improve the health of both farmed salmon and wild salmonids,” said researcher Lars Christian Gansel. He is head of NTNU’s Department of Biological Sciences in Ålesund and has helped develop the method.

“More information is needed about the spread of salmon louse larvae to document the effectiveness of current methods, as well as develop and customize new measures. Our model makes it possible to obtain this information,” he added.

Every year, between 400 and 450 million salmon and rainbow trout fry are released into net-pens in Norway. A single fish farm can contain millions of salmon and spread millions of salmon louse larvae into the fjords every single day.

The industry is regulated through a traffic light system, which is designed to protect wild fish (The aim is that salmon lice cause less than 10 per cent mortality in wild populations. Fish farmers must regularly count and report the number of parasites on the fish in their net-pens. Facilities are given a traffic light rating every other year, and the number of lice determines whether they are permitted to increase production (green light), continue as before (yellow), or must reduce production (red)).

“If we are to succeed in eradicating salmon lice, the best approach is to prevent contact between the parasite and the fish. To develop, evaluate and document the effectiveness of preventive methods, it is important to detect the larvae while they are still drifting around in the sea,” said the researcher.

Salmon louse larvae are not the only organisms floating around in Norwegian fjords. There can be as many as hundreds of thousands or even millions of other organisms per salmon louse larva in the sea.

Relative to the total amount of plankton and other particles, the salmon louse can actually be considered a rare organism, according to Gansel.

“We must therefore analyze large volumes of water in order to monitor salmon lice in the sea. If we don’t use enough water, we can easily overestimate or underestimate the number,” he added.

Many methods have been used to continuously monitor and count salmon louse larvae, but the vast majority of them have been cumbersome, imprecise, time-consuming and expensive.

“Currently available camera systems for plankton analysis often lack the resolution needed to distinguish between species and developmental stages. There is still no fully documented method for continuous monitoring of salmon lice in the ocean,” emphasized Gansel.

Artificial intelligence and machine learning have emerged as a promising possibility. The challenge has been the lack of high-resolution, clear, detailed images of larvae in a real seawater environment on which to train the AI models. However, the researchers at NTNU and Wageningen University may have found a solution. They have built their own video microscope and taken more than 120,000 images featuring close-ups of louse larvae and other organisms. As a result, they have been able to train AI models using synthetic data.

“The models performed just as well as the experts using microscopes. Even though some other marine species may look quite similar, the model was able to identify salmon lice in large seawater samples,” explained Gansel.

The researchers have concentrated particles the size of salmon lice from hundreds of seawater samples. In total, they have collected and filtered several thousand cubic metres of seawater from fish farms and marine areas near Ålesund. Sea conditions vary depending on the season and location. When there are few salmon lice in circulation, it takes a long time to compile reliable datasets from the samples.

So, in order to create more training material, the researchers hatched salmon louse larvae themselves and released them into the water samples. They then let the water flow slowly through a glass tube while filming the particles in the water stream with a video microscope (pictured).

Using software capable of tracking and selecting individual video elements, they isolated images of larvae at two different stages: newly hatched lice, or nauplii, and the slightly larger copepodites that are ready to attach themselves to fish.

“The individual frames from the videos will not show all sides of the larvae. They might be moving, or perhaps they are just floating about in certain parts of the tube. Because we only see a few of all the possible scenarios, we can improve the models by generating synthetic data to use alongside a large number of real-life videos. Salmon lice can vary slightly in size. To account for the differences, we can scale the lice, rotate and flip them, and include multiple lice in the same image. The same can be done with plankton and organisms that resemble salmon lice to improve the model even further,” explained Gansel.

The models can be used to monitor salmon lice infestation in areas where wild fish are expected. They can also be used to calculate the release of larvae and to study how they spread, grow and develop. Monitoring can help in assessing potential measures and is important when estimating the risk of transmission between farmed and wild salmonids.

“Measuring larvae directly in the sea will eliminate some of the uncertainty in the current system, where the number of larvae is estimated based on the number of lice on farmed fish. This will make the salmon lice map much more accurate. Production can be planned more effectively, and we can make better decisions about where to operate fish farms and what measures to take against salmon lice,” concluded Gansel.