We have talked to Professor Koen Van Leemput, who is heading the Medical Image Computing research group at DTU Health Tech, about his research and how it can make a difference for patients.
What is medical image computing?
In medical image computing, we extract relevant information from large collections of medical scans, for example magnetic resonance or compute tomography images.
We essentially use mathematical models, artificial intelligence (AI) and computers to help medical doctors do difficult, boring or repetitive parts of their jobs faster and potentially more accurately.
I think it is an interesting field because it gives someone like myself, who has a background in signal and image processing, a chance to work on very hard technical problems. Getting to collaborate closely with experts from completely different fields, ranging from clinical medicine to medical physics and basic neuroscience, is also very rewarding to me.
How does your research benefit patients?
Traditionally medical scans are analyzed by human experts visually eyeballing the images. However, this is extremely tedious to do, it easily misses subtle disease progression over time, and it is not nearly as reproducible as one would hope.
Furthermore, the acquired amount of imaging data keeps increasing, making it ever more challenging for humans to review all the data. By automatically extracting precise quantitative information from medical scans, AI has enormous potential value for detecting disease effects earlier, optimizing therapy planning, evaluating a patient's treatment, and in the development of new therapies (e.g. in drug trials) and in clinical research.
What do you hope to achieve with your research?
Since I started working in this field 25 years ago, we have progressed enormously. Many good software solutions now exist that can efficiently analyze imaging datasets acquired in large scientific studies. What is currently missing is the ability to analyze images that are acquired in routine clinical settings. These scans are of a much lower quality due to time constraints, they come in all types and forms because hospitals use different scanners with different settings, and they often contain large pathologies (e.g., tumors) that you do not see in for example neuroscientific studies. Bringing our computational tools into clinical practice, benefiting individual patients, is really the next frontier.
Ultimately, we would like to develop practical tools that work effectively in the real world. At the end of the day, the real joy is seeing the field actually progress. Nothing compares to seeing a method you developed do things that simply were not possible before.
Image: Brain scan (Koen Van Leemput)