Aiforia and the University of Bern have worked together for many years, utilizing each other's strengths to accelerate innovation. One of the most recent outcomes of this collaboration is an AI solution developed for lymph node metastasis detection. Using the models and training data from the collaboration as the core and starting point, the new AI model has been taken through Aiforia’s development processes and CE-IVD marking. This is an excellent proof of how collaboration between academia and industry can result in solutions that will improve healthcare.
We interviewed key stakeholders from the University of Bern’s and Aiforia’s teams to hear their thoughts on the collaboration so far. Keep reading to learn more or watch the video interview.
| Inti Zlobec Director of the Department of Digital Medicine and Professor of Digital Pathology at the University of Bern |
|
|
|
Bastian Dislich Surgical Pathologist, Institute of Tissue Medicine and Pathology at the University of Bern |
| Panu Kauppila Chief Product Officer at Aiforia Technologies |
|
| Richard Doughty Medical Director at Aiforia Technologies |
“Industry-academia partnerships are absolutely essential for innovation. Complimentarity of skills and knowledge is strength and helps to bridge the gap between research and clinical products. Where one side knows the ins and outs of their unmet clinical need and has the domain knowledge, the other side has the technical expertise and can help to create and market a product. And not only that, but it can be clinically approved by regulatory bodies, which is practically unachievable by us in academia,” Inti Zlobec, Director of the Department of Digital Medicine and Professor of Digital Pathology at the University of Bern, describes.
Aiforia has a long history of working with research partners and customers using Aiforia® Create as the common platform for developing AI models for a multitude of purposes. “These collaborations with academia are of utmost importance for us. By having researchers, pathologists, students, everyone using the same tool for building AI models opens a wealth of opportunities,” Panu Kauppila, Aiforia’s Chief Product Officer, explains.
Aiforia’s collaboration with the University of Bern is an excellent example. Since couple of years, the team at the University of Bern has built AI models, for instance, for H. pylori and lymph node metastasis detection for their own research purposes and for Aiforia’s product development. “Having the many insights from their research work helps us shape the products we bring to the market for clinical practice. Also, as part of our collaboration, we get to leverage their images and annotations in the actual product development. And of course, access to high-quality data is imperative for any AI algorithm training. With the University of Bern, we obtain data that has been prepared and documented according to the highest standards with their Swiss accuracy. This makes the overall process of getting to high algorithm quality very efficient,” Kauppila continues.
“One of the advantages of having such an industry-academic partnership, such as the one we have with Aiforia, is also to reduce the turnaround time. So, basically, when you have an idea and you want to bring it to market, and you want to bridge this, Aiforia was really there to help us to do this in a quick way.” – Inti Zlobec, Director of the Department of Digital Medicine and Professor of Digital Pathology at the University of Bern
“The main thing that I really valued about working together with Aiforia is that you can always reach out to them and you will have an answer via email if things can be clarified in a few sentences, or you can request a meeting to discuss more complicated things or to get their advice on how to continue with a project or how to modify an algorithm. They have been very helpful with us. So, the overall experience that we have was a very positive one. – Bastian Dislich, Surgical Pathologist, Institute of Tissue Medicine and Pathology at the University of Bern
When asked why the University of Bern wanted to start investing in AI, Bastian Dislich, Surgical Pathologist at the Institute of Tissue Medicine and Pathology at the University of Bern, lists three main reasons:
1. It helps the pathologists with the increased demand.
“There's more and more specimens getting sent in, the diagnoses are more complex, and on the other hand, we have a shortage of pathologists, and to help counterbalance that imbalance, AI can really help, sort of as a second digital virtual helper that helps you along to get your daily routine tasks done efficiently.”
2. For certain tasks, AI is more attentive or precise than a pathologist.
“For example, if you have to count things and search for certain things, those are very tedious tasks. They don't need any intellectual input from our side. We just need to focus and pay attention. And we know that we get more tired towards the end of the day, and having an AI by our side really helps us to be more precise and also not to miss anything, especially when we're kind of a bit worked over or at the end of a long day.”
3. AI sees things in tissue specimens that the human eye cannot see.
“If you train enough images, for example, of a certain tumor type and that tumor type is associated with a specific molecular phenotype, for example, then AI can draw the conclusion that whatever morphological feature is there is very often seen in combination with that molecular phenotype and very often we as the pathologists, we cannot even say why the AI is making that connections because there's certain features that are just not visible for us humans.”
The Aiforia® Lymph Node Metastasis model is an AI-powered image analysis solution that scans H&E whole slide images of lymph nodes and highlights even the smallest metastatic foci from breast, colorectal, and melanoma cases. It is trained on thousands of expertly annotated images in collaboration with the University of Bern and fine-tuned in the Aiforia® Clinical Suite. It is CE-IVD marked for diagnostic use in EU and EEA countries (IVDR) and for Research Use Only (RUO) and Performance Studies Only (PSO) in all other market areas.
“We developed the AI lymph node metastasis model because lymph node screening is one of the most labor-intensive steps in cancer staging. A single case can require the pathologist to go through tens or even hundreds of slides under time pressure. By automation, the read and flagging of suspicious regions frees the pathologist to focus on interpretation and reporting, and not exhaustive searching for the metastasis.”, Richard Doughty, Aiforia’s Medical Director, explains.
The model solves three challenges:
Aiforia® Lymph Node Metastasis AI model in Aiforia® Clinical Suite Viewer
“We found that using this model, we really have a great AI companion. The current model really does not miss anything. It is very sensitive. So, you basically do not miss any lymph node metastasis. If I screen lymph nodes in a rather quick fashion with a conventional or digital microscope and I don't see any metastasis, then I will also ask the model what its outcome is. And if the model is also negative, I'm reassured that I did not miss anything. The overall outcome is that it saves you a lot of time in the end and increases the accuracy of your diagnosis,” Dislich describes.
Dislich estimates that more general AI models will become available in the upcoming years: “At the moment, AI is really good at answering a very specific task, like counting a certain number of cells. What is coming, I think, in the next years is more generalistic models where you basically just feed the model a slide, and it can by itself decide what is going on: is this a neoplastic process, or is this an inflammatory process? If it is a neoplastic process, it can already come up with the differential diagnosis, it can suggest certain stains to be done, and order them automatically from the lab as a more agentic way. So, having your own digital companion helps you triage cases and assists you in your workload.”
He also sees that in the mid-term future, the AI model could handle the straightforward cases independently, and the pathologist could focus on the more tricky cases.
“I see AI going a lot more integrative, a lot more multimodal. And it will definitely be exciting times in pathology,” Zlobec concludes.
Learn about other clinical collaborations: