When we can learn to apply AI to the tsunami of data currently collected, we can revolutionise the treatment of cancer. But first we have to look outside the box.
There is a cartoon that I keep at my desk. It is of a person telling a cat, “Don’t you dare step outside of this box” — the box being a litter box! I feel this is the way most of the medical community is approaching cancer therapies these days – thinking within the box.
We shoot the radiation where we can see the cancer. We shoot at what we can see, to eliminate it. What we can’t see gets left behind – often at the margins. Some of these residual cells are cancer stem cells, a rare subset that can initiate and sustain tumour growth. Leaving these behind is like leaving a time bomb ticking away – it’s only a matter of time until it blows up.
In 2017, my PhD student and I were investigating the physics of heavy ion therapy, where cancer is treated using beams of accelerated atomic nuclei. We were cataloguing all of the wide assortment of radiation that splashes around during this process, all the fragments of nuclei, using a powerful simulation toolbox from the world of high-energy physics. We realised that there was a substantial cloud of low-energy neutrons around the target volume. We had to check our simulation and analysis code multiple times – nobody had reported this before!
Not only was the cloud centred on the tumour, it extended quite a way beyond its perimeter. It prompted the question: what if we used stable, non-toxic isotopes, coupled to disease-specific biochemical vectors, to capture these neutrons and cause more damage to the tumour cells? What if we could use these neutrons to kill the cancer cells that are hiding in the back – the ones that we can’t see? The cancer stem cells that are waiting to launch a second wave of attacks and invade our organs again?
This was a radical idea and one that could potentially revolutionise the field of cancer therapy – but we needed to show that it worked in practice, not just in a simulator. We put together a small team of brilliant chemists, biologists and physicists who were all intrigued by the idea and devised a pipeline that was first tested at the Australian Centre for Neutron Scattering at ANSTO’s OPAL reactor, before being deployed at a particle accelerator in Japan.
When our first results came in, I had to sit down. They showed that our proposed method was a lot more powerful than we had expected. We had shown that with a third of the radiation and the addition of drugs that carried stable neutron-capturing isotopes to the cancer cells, we could achieve the same outcome both where the beam shone, and outside of the tumour. We had shown that by using all the low-energy neutrons that people don’t like and pretend don’t exist, we could deliver a second punch to cancer.
Here’s the thing about me and my team: we are not standing in the litter box – we are firmly outside of it. I think people should try joining us on the outside: there is plenty of room and it smells better!
I am a science realist, a naturalist. I firmly believe in the self-correcting nature of science. While I await the outcome of each and every one of our experiments with bated breath, a falsified hypothesis is only a precursor to a more robust conjecture. Challenges are exciting and often provide a new opportunity.
I think one of the biggest challenges that we face today in health and diagnostic medicine is big data – the enormous quantity that is being produced at an exponentially growing rate, and the application of artificial intelligence (AI) to sort and mine it for information.
Traditionally, we look for information – we know what we are looking for, where to look, and we either find it or we don’t. Today we have a tsunami of data – and if we are to utilise AI, then we have the problem of training it. How can we train an AI system to search for something when we don’t necessarily know the patterns or signals which may be significant – things that are yet to be classified and labelled? We could also use unsupervised learning for dealing with unlabelled data.
But what about biases and fairness? The data we put into the algorithm very much determines how robust, accurate and fair these models are. It is also important to remember that an algorithm can be both correct and biased and this is quite important when choosing data for training of a network. Even a simple calculator does not guarantee us arriving at the right answer – garbage in, garbage out.
Pursuing a career in science has always seemed a natural path to me: I am curious about the functioning of nature, love solving puzzles, and have always felt a deep sense that there is more to know and discover. My original career trajectory was geared towards research and development in the area of the engineering of radiation detectors and imaging systems. This discipline combines maths, quantum dynamics, particle physics, electronics and signal processing. I chose a PhD topic on medical imaging – new detector systems for positron emission tomography (PET), widely used for cancer diagnosis and biology research – and this ultimately led me in the direction of medical physics research.
Life is enabled by the transference of electrons in a water substrate. Our cells use the same principle of potential difference that we use in batteries to transport food across their membrane. Here, I had found an area that allowed me to combine all my interests, but also pushed me to collaborate with people outside my own field of expertise – biologists, chemists and clinicians. Suddenly my sandbox grew so much bigger!
It is a privilege to be a physicist (and I think this applies to all scientists). We do what we find most interesting and are paid to imagine, to think and conjure up hypotheses that may at first sound like science fiction. I cannot ask for a more fulfilling career.
Dr Mitra Safavi-Naeini
Dr Mitra Safavi-Naeini is a Senior Physicist and Research Lead in Human Health at ANSTO. She is a prolific researcher in the field of particle physics and medical radiation physics. Her two main research areas are radiotherapy (proton and heavy ion therapy, brachytherapy, photon therapy) and medical imaging quantification (with a specific focus on PET).