• There are no suggestions because the search field is empty.

Food safety case study: salmon skin analysis by AI

Artificial intelligence can be used for image analysis tasks in quality control (QC) and food safety assessments, like studying the health of salmon skin.
Written by Aiforia

Interview with Nofima scientist Dr Lene Sveen

What is your job and what do you do?

Lene: I am a researcher in the fish health group at Nofima (an institute for applied research within the fields of fisheries, aquaculture and food research).

A summary of your project with Aiforia:

In Aiforia we have developed a model that recognizes the successive tissues in Atlantic salmon skin.

salmon skin analysis by deep learning AIThe skin of fish are  complex organs, with multiple cell types and tissues that responds to the environment. Healthy skin is important for the robustness of the fish, and skin disease is a primary constraint to the commercial aquaculture industry.

We have analyzed skin from Atlantic salmon with bacterial diseases, parasites and how the skin responds to handling operations such as mechanical de-lousing and transportation.  

What does Aiforia enable you to do that you otherwise could not (or would find difficult) to with traditional or other methods?

There are many advantages with Aiforia. We can measure and count tissue, which is impossible to do manually. We can have large sample sets, which makes the analysis more accurate, and continuous data is also a big advantage.

What were your expectations of using AI?

My expectations were not too high. Last year we decided to give Aiforia a try, however I am very impressed with the platform and what it can learn!

Did you know anything about AI before starting your work with Aiforia?


Some details around your project with Aiforia (slide numbers, staining types, etc.):

We stained our skin sections with AB/PAS. Now we have 200 slides in our project. The AI-model worked very well with fewer slides, but whenever we have experiments with different fish sizes, infection, wound healing trial, or batches that stain in different shades of the AB/PAS color we add some more training in order to make the AI-model even more efficient.

What stage are you at with your work with Aiforia?

The AI-model is in continuous work. We are about to publish results from one trial, however, the AI-model will have to evolve with the input samples and staining. 

What is the most unique aspect of your work with Aiforia?

During my work developing the AI-model I have spent many hours drawing annotation regions, this has given me a deeper understanding of the tissue that I work with, and maybe we even have discovered some new functions of the skin which we plan to look at closer in future projects. 

Read the publication of Dr Sveen's study here.