Recently, a project on pancreatic cancer at the University Medical Centre in Amsterdam, aimed to develop a machine-learning algorithm that would be able to identify pancreatic cancer from medical images.
So far, so good, you might think. However, as with all machine learning (ML) projects, there was a catch.
– A guest blog post from the quantum community in Norway by Professor Pedro G Lind, Department of Computer Science at Oslo Metropolitan University, participating in two of its main research initiatives, namely NordSTAR and the Quantum Hub.
Somebody had to provide the data that could be used to teach the algorithm to distinguish between cancers and benign tissue. The hospital had plenty of imaging data, but it was not labelled, so how could the machine learn?
Answer: with the help of 20 pancreatic surgeons, who took the time to go through thousands of images and labelled them according to their expert knowledge. This provided a training dataset for the ML algorithm—and that algorithm can now identify cancer with more than 80 per cent accuracy, helping back surgeons in future, not only in improving diagnosis but also in saving hours of reviewing images.
“Data hungry” and complex algorithms
But again, I can’t stress this enough: the algorithm could not have been developed if those 20 surgeons had not recognised its potential and voluntarily dedicated their time to label the data.
This is an important take-home message: ML algorithms are typically “data hungry”, at least till properly validated and tested. Another “cost” of such algorithms is also their complexity, which makes them look like spooky black boxes to surgeons and clinical experts, doing “magic” with the data. And this is particularly true when ML is pushed to the cutting edge of new – still unconventional – technologies for computation, namely quantum computing.
Quantum algorithms for experts in life sciences
During the last couple of years, I’ve started working on quantum computing together with some of my colleagues at Oslo Metropolitan University in Norway. One of the projects we are now working on links intelligent health and quantum computing. We are looking at algorithms that will help to optimise and personalise cancer therapies for patients with particular types of cancer.
This is a relatively new field in cancer therapy. There are already some mathematical models that can simulate patients with particular characteristics to explore how they would react to specific therapies and doses. However, these models require considerable computing power to examine all the possible combinations of therapies, alongside patient data past and present.
We are looking at two options side-by-side: a conventional machine learning algorithm, and then a quantum implementation. The aim is to compare the two and assess the comparative power of the quantum model.
The discussion has just started …
While it is a promising interdisciplinary research activity, my colleagues and I are starting to experience some challenges in bringing quantum algorithms and its new programming “paradigm” to applied fields outside computer science and artificial intelligence.
There is still a lot of discussion about how best to go about this. Many of my colleagues in physics believe that to teach quantum computing, people first need to understand quantum physics, which would impose a natural barrier to surgeons and other scientific experts without a heavy background in physics or mathematics.
Personally, and probably fortunately for many life scientists wanting to dip their toes in the quantum waters, I’m not sure it’s necessary to have such a pre-knowledge in physics. I think that, for the same reason nowadays we don’t need to understand electronics to write and program conventional algorithms, we will also not need a deep investment in quantum physics to prepare quantum programmers.
Educating practitioners in the health sector
The real issue—and this is where the story from the Netherlands chimed in—is how we, computer scientists, should educate current practitioners, including surgeons and experts in healthcare and other applied fields so that they understand more about quantum computing and even – why not? – develop their own programs to solve practical problems.
Quantum computing is obviously not the only field where these matters. In fact, with the recent access to ChatGPT, which enables almost anyone to develop their own programs, e.g., python or other languages, teaching both conventional and quantum computing will soon face a need to change their paradigms and strategies. But, at the same time, this might be a great opportunity to start enlarging the community using quantum algorithms and exploring its potential in different applied fields such as in health sciences.