Center for Artificial Intelligence in Medicine (CAIM)

People

"I would like to convert my research into a useful tool for clinicians."

December 2022

Amith Kamath wishes to facilitate faster radiotherapy treatment for patients with glioblastoma through AI-supported therapy planning. The CAIM Young Researcher Award winner appreciates the openness of the Bernese community around AI applications in healthcare, also welcoming ideas from people trained in other disciplines to tackle hard problems in medicine. He is currently pursuing his PhD at the Medical Image Analysis research group of the ARTORG Center for Biomedical Engineering Research and looks forward to translating his research into a clinical tool through the broad entrepreneurial support he is receiving in Bern – including the personalized business coaching by be-advanced as part of his CAIM Award win in the category “translation”.

Amith Kamath (second from right) wants to build Deep Learning models that can help radiooncologists to quickly and safely plan therapy for brain tumor patients. Here with his research partners Prof. Dr. med. Nicolaus Andratschke and Dr. med. Jonas Willmann, Department of Radiooncology, University Hospital Zurich (left), and his supervisor Prof. Dr. Mauricio Reyes, ARTORG Center, University of Bern (right). (© Mauricio Reyes for CAIM)

What drives you in your research?
My research is centered around evaluating the quality of radiotherapy delivered to patients with glioblastoma. Given the usually bad prognosis, people already diagnosed with this tumor currently must wait between one and three weeks until they can start treatment, due to the current workflows in radiotherapy planning. We expect that by using AI models to help draw boundaries around organs while simultaneously estimating the radiation dose and toxicity, radiotherapy treatment can be started earlier before the tumor has progressed further. We hope that someday our work can really add quality to people’s lives in this sense.

Amith Kamath discussing tumor contouring in radiotherapy planning with Dr. Ekin Ermis, Department of Radiooncology at Inselspital. (© CAIM, University of Bern)

The challenge in radiotherapy for glioblastoma is to be targeted while killing the tumor but sparing healthy areas of the brain. Mistakes made in these initial steps in the process can add up in subsequent steps, underscoring the importance of being precise. For example, if you irradiate healthy tissue in the brain, people can lose their functional abilities, for example speech or motor abilities. Our idea is to use deep neural networks to not only estimate, but also simulate inter-expert variations in boundaries that are drawn around tumors as well as healthy areas during radiotherapy planning. These simulations give us a better sense of the range of safe variations in how human experts manually do this and thereby understand the clinical impact of such variations in the process, leading to safer treatment.

What does winning a CAIM Young Researcher Award mean to you?
What matters to me most is that many of us were able to share our research and receive constructive feedback and comments in a setting like the CAIM Symposium. Beyond the award, the existence of such a vibrant community is very rewarding.
For the translational focus of the award, I was fortunate to receive prior exposure through the Innosuisse startup toolbox program “Business Concepts” in October this year. The entrepreneurial coaching opportunity I now have with the CAIM Award is the perfect continuation of that. It would be great to get the experts’ opinion on how we can convert our research into a useful tool or product for clinicians!

If there are ten users of what I build that will mean more to me than writing a PhD thesis that no one may read.

The unique thing here in Bern is the entrepreneurial support you receive, for example through the Swiss Institute for Translational and Entrepreneurial Medicine, sitem-insel. There is a lot of existing knowledge amongst the faculty about how a PhD project can be shaped into a product that can be used in clinics, which is quite exciting! If there are ten users of what I build, that will mean more to me than writing a long PhD thesis that no one may read.

Discussion on the potentials of Deep Learning in radiotherapy planning for brain tumor patients. (© CAIM, University of Bern)

How important is it for you to share your research?
Very much! My background is mostly in image processing and computer vision, and I think it’s great how welcoming the scientific community here is to researchers from other academic backgrounds. I don’t have a biomedicine background, and I believe people without medical schooling can make strong contributions to tackling hard problems in the medical space. Ways of thinking that are commonplace in another field could be novel to healthcare challenges and thus lead to innovative solutions.
I like the people I get to work with daily who motivate me by asking all the right questions. I like that my work is very visual: I find it easier to look at a set of images or a video than at a bunch of equations for an “Aha” moment. When some images are hard to interpret, I appreciate that clinicians are quite open to talk to technical folks like us. This readiness to work with each other and speak the same language is quite important in this line of work.
Therefore, I like to share our research with a broader global community. I use Social Media to exchange ideas with other scientists and our research lab Medical Image Analysis has started a “How to” video series for beginners in Deep Learning for medical imaging, summarizing some of the pitfalls and stumbling blocks in a humorous way (https://github.com/ubern-mia/bender). We are also currently preparing for a symposium on interpretability of AI models at CAIM in March ‘23, with the hope to get a lively discussion going on this important topic for safer AI adoption in medicine.

Amith is a computer scientist and holds a Master of Science in Computer Science from Georgia Institute of Technology, in the US. He worked earlier as a software developer at the MathWorks Inc., on the Image Processing and Computer Vision Toolboxes in MATLAB, a scientific computing programming language. Prior to that, he earned a Master of Science in Electrical Engineering at University of Minnesota, and a Bachelor of Technology from National Institute of Technology Karnataka, in Surathkal, India. Currently, he is pursuing a PhD in Biomedical Engineering at the ARTORG Center at the University of Bern, under the supervision of Prof. Dr. Mauricio Reyes.

His PhD thesis is on image segmentation and how one could use AI models to not only automate the otherwise time and effort intensive segmentation process, but further evaluate the quality of the contours in comparison to human-experts. His research focuses both on the robustness of using AI models to perform auto-segmentation, as well as computing radiotherapy dose predictions from these contours faster than current methods used in clinical practice. These results could help improve the speed as well as the quality and safety of radiotherapy plans for patients suffering from glioblastoma.

Bern Interpretable AI Symposium (BIAS): www.caim.unibe.ch/bias2023