The digital radiologist who detects cancer and rare diseases

Ida Hagström, associate professor of computer vision and medical image analysis in the Department of Electrical Engineering at Chalmers, works on analyzing medical images using machine learning techniques.

– This means that I train computers to find cancer for example. It is through algorithms that computers perform tasks, in this case interpreting medical images.

With the help of deep learning, Ida Häggström trains computers by showing them large amounts of data while displaying the results. In this way, the computer learns to recognize patterns in images so that it can read, for example, X-ray images or nuclear medicine images.

– For example, I studied more than 17,000 images of about 5,000 lymphoma patients and created a learning system where the computer was trained to find signs of cancer in the lymphatic system.

Such a system could ease radiologists' workload, provide a second opinion, or prioritize patients with the most urgent need.

Ida Hagstrom doesn't think computer vision will completely replace radiologists, at least not in the near future.

– I don't think human expertise will be completely replaced for a while, but some aspects of the work will be enough.

Another advantage that Ida Hagstrom sees in computer vision is that it contributes to greater equity in care.

– For care to be more equal, a digital radiologist in the form of a trained AI system that can work anywhere is suitable. So, everyone, no matter what hospital you're in, patients have access to the same expertise and can have their images reviewed within a reasonable amount of time.

Even in the case of unusual diseases, an AI system may be more appropriate. A good evaluation relies on the radiologist seeing thousands of images, but in the case of unusual pathology, one may only see a few images. If you create an AI model, you can give it access to more information.

Ida Häggström works with Sahlgrenska Academy and Sahlgrenska University Hospital on several research collaborations, focusing on medical image analysis using machine learning techniques.