MedImML - Computational Medical Imaging and Machine Learning

Developing machine learning and deep learning models that can analyze medical images and image-related data

By Matthew Davidson

Computational medicine has made huge gains over the past few years, thanks largely to advances in machine learning (ML). Still, more research and development is needed in order to realize the potential of machine learning for personalized, predictive medicine. MedImML is at the forefront of research into developing machine learning and deep learning models that can analyze medical images and image-related data, and bringing those techniques to clinical practice and patient care.They are using principles of mathematics and technology to create tools for many different medical applications, from cancer to aging to drug discovery.

MedImML’s approach is based on three pillars: research, education, and translation. Taking machine learning algorithms from the lab to the patient requires close collaboration between technological and clinical researchers and physicians. The project is based at the Mohn Medical Imaging and Visualization Center in the radiology department at Haukeland University Hospital, and therefore well-positioned to identify and understand the immediate, high-value medical needs, and to design tools that can integrate into existing hospital infrastructure and workflows.

To increase the common understanding of medical machine learning among clinical and technical disciplines, the project put a strong emphasis on educational activities. Two examples are the joint course between the University of Bergen (UiB) and the Western Norway University of Applied Sciences (HVL): ELMED219 , offered to medical students and engineers, and the recently established DLN Research School PhD course “A hands-on introduction to artificial intelligence in computational biotech and medicine”.

One example of MedImML’s research activities is using MRI images to study neurodegenerative disorders through the gap between biological age and chronological age of the brain and other organs. The biological age of your organs is not necessarily the same as your chronological age. Different factors like disease or trauma can make a 60-year-old’s brain look more like the brain of an 80-year-old. By training a machine learning model on repeated brain scans of over 40,000 patients, MedImML is able to measure this gap to try to identify neurodegeneration before it becomes clinically apparent. This tool is being designed to work directly on the medical images, without time-consuming pre-processing steps, and will be embedded in a system emulating the regular imaging workflow at the hospital.

MedImML is committed to responsible research and addressing two of the major concerns with AI: explainability and uncertainty. Deep learning models are often «black boxes», performing well under favorable conditions but poorly understood and prone to break down in unexpected ways. MedImML’s research aims to contribute to more robust uncertainty estimates for such methods, increasing user confidence and reducing biases.

The mathematical tools being developed in MedImML have wide application across the medical field. The researchers have only just begun to discover how these tools can be used to benefit patients. They are currently looking for collaborations with other machine learning researchers to continue advancing their research.

Project information

  • Category:
  • Duration:
  • Funding:
    Trond Mohn Foundation
  • Institution:
    University of Bergen, Western Norway University of Applied Sciences
  • Associated project

Project lead

Arvid Lundervold



Project lead

Alexander Selvikvåg Lundervold




Haukeland Universitetssykehus

External resources

Project website

Mohn Medical Imaging and Visualization Centre