The UMCG website shows YouTube videos. YouTube places cookies if you watch them. Choose ‘Rather not’ if you do not want these cookies. We also place cookies of our own. These help us improve the usability of our website.
In pathology, tissue samples from patients are routinely taken and analyzed for diagnosis. The analysis is often performed with the naked eye or with microscopes to determine, for example, whether cells are healthy or cancerous. These tasks can be time consuming and the assessments may vary per pathologist, to which digital pathology presents a promising solution. Here, glass tissue slides are digitalized and analyzed with the help of computer algorithms. “I spend an increasing amount of my time investigating possibilities for these algorithms, how we can validate them and implement them into our clinical workflow”, Bert van der Vegt explains. “The use of digital image analysis algorithms can save the pathologist a lot of time and leads to better cell and tissue based diagnostic and predictive tests for tasks with a poor intra-observer reproducibility.”
With the recent advancements in and growing popularity of deep-learning platforms based on convolutional neural networks, it is now possible to train complex digital image algorithms. These algorithms do not depend on pre-defined parameters, giving researchers the possibility to study more intricate relationships between cells, such as the relation between a tumor’s microenvironment and its response to treatment.
Challenges in setting up a tissue and cell-based deep learning facilities are the high amount of data that is needed to train robust algorithms and ensuring the patient data is well protected. “I am a pathologist, not an AI or data scientist, so I needed help on the technical part of the project”, Bert van der Vegt says.
“DASH helped me to get in contact with other people in the UMCG who are already working on similar problems to learn how we can approach this in our department.” A connection was made between Bert van der Vegt and the Machine Learning Lab of the UMCG, so they could team up and exchange knowledge. Earlier this year, Bert van der Vegt received a grant from the UMCG Cancer Research Fund, with which he acquired an addition to the platform. “This addition allows us to build, train and validate our own deep learning based algorithms.” The data storage for this will be facilitated with a premium version of the Virtual Research Workspace that has GPUs and ensures data protection. Currently, a lot of work is done to realize this still ongoing project.
The (IT) solutions that are used in this project, are available for researchers within the UMCG as well.
Virtual Research Workspace: An environment for secure collaboration amongst different researc hers (working within and outside the UMCG) for analyzing data simultaneously and/or sharing privacy-sensitive data.
The Machine Learning Lab provides support to the whole UMCG community considering machine learning matters, connects a machine learning community and offers biweekly research meetings to discuss machine learning-related topics on a technical level.