CONNECT seminar on Digital Pathology and AI
Experts from regional, national and international groups shared their perspectives on digital pathology and artificial intelligence (AI) at an educational seminar hosted by CONNECT Working Group 1 on 23 May 2022 at Oslo Cancer Cluster Innovation Park.
The first speaker was Professor Emiel Janssen from the Pathology Unit at Stavanger University Hospital. Janssen shared the experiences from implementing digital pathology in the Norwegian Western Regional Health Authority (Helse Vest). With new and efficient equipment, all histological slides are now stored as digital images and can easily be shared between hospitals and health professionals, without the need for physical slides to be sent around. Janssen said:
“Digital pathology opens up a lot of possibilities for research, development, training and consultations.”
Janssen presented the regional Pathology in West project (link in Norwegian), which is supported by strategic funds from Helse Vest with the goal to develop, test, validate and implement computational pathology in the region.
The next focus of the seminar was how to use AI to benefit from the full potential of digital pathology. Hrafn Weishaupt, postdoctoral researcher at Haukeland University Hospital and part of the Pathology in West project, described several deep learning methods. The methods ranged from supervised to fully unsupervised efforts to train AI models in disease prediction, diagnostics and treatment decision.
Weishaupt concluded with some remarks on the challenges with deep learning:
“A lot of choices have to be taken about image processing before AI training and we do not know what effects these choices have.”
The next speaker was Associate Professor Marit Valla, leader of the research group Artificial intelligence and digital pathology in cancer (AICAN) at the Department of Clinical and Molecular Medicine at NTNU. Valla shared valuable insights from the research on AI and interpretation of histopathological slides from breast cancer. One of Valla's main concerns is having enough data to validate the AI models:
“AI models have to be used across borders. Validation of the AI models is critical."
Torbjørn Furuseth, CEO and co-founder of DoMore Diagnostics, stressed the need for better predictors in present tumor diagnostics. Furuseth said the big benefit from AI in digital pathology is the possibility of developing better prognostic and predictive tumor markers:
“We believe AI is on the verge of breakthrough in diagnostics in digital pathology.”
Jeroen van der Laak, professor at Radboud Medical Center, and Dr Pierre Moulin, pathologist at Roche and architect of the project, presented the BIGPICTURE-project. This is an EU-funded public-private partnership with the purpose of creating a European repository for high-quality pathology data and AI algorithms. A main concern in AI development is to have enough data for training and validation.
“AI offers big promises, but the challenge is to have enough data. Here, BIGPICTURE comes into play.”
Van der Laak and Moulin further stressed the importance of making an infrastructure of trust, where data and algorithms can be stored and shared in a safe and compliant way. They stated that BIGPICTURE should be an inclusive and safe arena where specialists from different fields can collaborate in solving present and future challenges in computational pathology. This will include many types of cancers, sometimes rare, which would be otherwise at risk to be neglected, or as expressed by Moulin: “beyond the obvious”.
You can access the video of the educational seminar below.