By Tobias Bolli, Junior Project Manager Academic Relations
Medicine has come a long way. Prior to the 19th century, all illnesses in the West were ascribed to an imbalance in bodily fluids, which were assumed to determine people’s health and even their character. Even though in the 21th century we have a clearer picture of how the body works, the boundaries of medicine are still being pushed. In the 3rd edition of our Connected Series on July 9, 2020 we took a closer look at Sino-Swiss collaborations in this field.
Our first speaker, Prof. Dr. Mauricio Reyes, Associate professor at the Artorg Center for Biomedical Engineering Research at the University of Bern, started off by introducing his collaboration with Malong Technologies, an up-and-coming AI start-up in Shenzhen pushing the boundaries of computer vision. Prof. Reyes detailed the mission of this joint-research effort, namely to create a robust and accurate AI system for MRI imaging. He stressed that far and above the most important aspect of such a system is patient safety.
Visualization of the human brain. Due to their data-rich structure it is very time-consuming to analyze MRI images manually.
Analysis of MRI images, when done by humans, is a very cumbersome task. So much so that radiologists often resort to simplified 2D images which still are very time-consuming to analyze. In order to free radiologists from this mundane and rather uninteresting work, Prof. Reyes is developing machine-learning algorithms to detect and quantify patterns that indicate the presence of a pathology, in particular strokes and brain tumors.
The algorithms are fed with the knowledge of human experts who know exactly what to look for when it comes to detecting pathologies. Algorithms then use this knowledge to process not only one particular image, but all sorts of MRI images. Besides being able to quickly process data and pin-point anomalies, the AI system is also capable of describing pathologies in terms of their size, volume, extent, etc. The idea behind the technology isn’t to replace radiologists, but instead to free up their time and empower them to do more interesting tasks according to Prof. Reyes.
Detecting early signs of what could later cause a stroke is particularly important in China, a country in which strokes are most prevalent and the most frequent cause of death. To conclude his presentation, Prof. Reyes explained the concept of “interpretability” and why it is so essential for AI technology. Data algorithms are often very powerful but are hard to understand due to their inherent complexity. Interpretability is about being able to understand what is going on inside this black box and thus avoid the potential pitfalls that come with this technology.
A world map depicting the frequency of stroke incidents. China is among the most affected countries and thus stands to benefit greatly from Prof. Reyes´ research.
Next, Prof. Tianwu Xie, researcher at the Institute of Radiation Medicine at Fudan University, talked about computer models he develops in collaboration with a team at the University Hospital of Geneva. It is a well known fact that new drugs have to be tested in preclinical trials first before they are approved for wide-spread use. Animals and humans are normally used as test subjects to get to that final stage. Prof. Xie pointed out that this established testing procedure comes with significant drawbacks, it is costly, time-consuming, and poses a risk to the humans and animals involved. Thus it makes sense to look for alternatives and computer simulations of the relevant biological systems are exactly that.
The end goal is to have complex simulations taking into account differences between individuals such as height, age, body mass, sex, etc.
Prof. Xie presented computer models he developed together with University of Geneva - the Sino-Swiss team was very prolific and came up with an entire library of computer models. Among them were of course models of mice, the most frequently used specimen in animal testing. Prof. Xie also introduced models of trouts, crabs and even flatfish which are sometimes used to evaluate radiation levels in rivers. Last but not least, Prof. Xie presented a myriad of human models ranging from fetuses to elderly humans of both sexes. The idea is to go beyond a mere default or average model and simulate individual humans with individual characteristics.
Some of the models have already been tested in preclinical trials, for example a virtual rat which was used to evaluate the health consequences of various radiation doses (which can lead to leukemia). Moreover, there were virtual trials studying drugs to combat liver cancer and Prof. Xie even came up with a simulation of a pregnant patient to study the effects of radiation.
Prof. Xie already put some simulations to the test, for instance a simulated rat for evaluating a new cancer therapy.
Q&A session
During the lengthy Q&A session our two speakers addressed a host of different questions. Prof. Reyes acknowledged skepticism about AI-assisted MRI image processing only five to seven years ago. Professionals feared that they would lose their jobs as a consequence of this technology. However, today’s radiologists and doctors are mainly seeing the advantages of this technology.
Prof. Reyes pointed out that there has been a 1000% (!) rise in diagnostic demand over the last decade, meaning doctors have less and less time to look after patients. Automated MRI images processing could help reverse that trend and free doctors up to spend more time together with their patients. Due to more accurate diagnosis, the technology would also improve treatment and thus create concrete health benefits.
Asked about the potential of AI for medical applications, Prof. Reyes cautioned that we have to keep in mind the limitations of current technologies. Even though they deliver powerful results, algorithms aren’t capable yet of understanding cause and effect. Once AI grasp this all-important concept, Prof. Reyes predicts big breakthroughs in his field.
The webinar panelist from top left to bottom middle: Prof. Tianwu Xie, our moderator Gu Libing, Head of Academic Relations at swissnex China as well as Prof. Mauricio Reyes.
Prof. Xie highlighted that there are numerous challenges for computer models, namely the fact that it is still very time-consuming to program them. He hopes that in the future neural networks will be able to construct such models themselves and thus speed up the process.
Asked about when we will be able to perfectly simulate biological systems, Prof. Xie stressed that there is still a long way to go - not only in terms of developing better algorithms, but also in terms of understanding the underlying biological processes. Both professors expressed optimism that AI will enable us one day to fight diseases which cannot yet be cured. A key in unlocking a solution is to mine more and better health data according to Prof. Reyes.
At this point, we want to express our thanks and appreciation for Prof. Reyes and Prof. Xie for their interesting and in-depth presentations and thoughtful answers during the Q&A session.
Please find a link to the slides and webinar recording below:
