Image credit: Gerd Altmann, Pixabay
Love or hate it, AI is here to stay, and many cancer researchers are already adopting it into their workstreams. AI has potential to improve patient outcomes across cancer types, but it is arguably set to have the biggest impact in skin cancer first. Here, we speak with Dr Matt De Vries, who used melanoma cells to develop an AI tool that could help new drugs reach patients much more quickly.
Artificial intelligence (AI) is increasingly infiltrating our lives, with the technology improving at a staggering pace. However, determining where AI is helpful and where it risks becoming too pervasive is challenging. Could AI really pose a threat if it becomes too advanced? Or are we missing out if we don’t explore every possible use for it?
One area in which AI’s benefits seem clear is healthcare. Technologies that allow computers and machines to perform human tasks – often at a higher level – can increase efficiency, save costs and improve accuracy. The reduced administrative burden allows healthcare professionals to spend more time and energy on patient care. Patients also stand to benefit from AI-related advances in diagnosis and treatment.
Dr Matt De Vries, a recent PhD graduate at The Institute of Cancer Research, London, and his supervisor Professor Chris Bakal, Group Leader of the Dynamical Cell Systems Group at The Institute of Cancer Research (ICR), believe so strongly in the power of AI to revolutionise cancer care that they have co-founded a spinout company, Sentinal4D, to further explore its uses.
One of their priorities is to continue developing and testing their patented AI tool, which has potential to support clinical decision-making, particularly in personalised medicine. This tool was created using melanoma cells, and Dr De Vries – who is Chief Technology Officer at Sentinal4D – believes skin cancer is an area in which AI could have diverse applications.
He said: “I foresee us being able to use AI in many ways – from early lesion detection and risk stratification to treatment response prediction and drug discovery.”
Your support will help our researchers make more life-saving discoveries, giving people with melanoma precious extra time with their loved ones.
Using AI to make new treatments available to patients more quickly
In the UK, clinicians are already using AI to try to improve outcomes for the hundreds of thousands of people who receive a skin cancer diagnosis each year. From detecting skin cancer more efficiently to supporting clinical decisions and even helping to personalise treatment, there are many ways AI can inform and speed up processes along the care pathway.
ICR scientists are now looking at how AI can be used to better monitor treatment response. Using AI to build on decades of ground-breaking work by their predecessors on cell shape, they have developed a tool, MorphoMIL, that can show how cancer cells respond to new drugs by observing changes in their shape and structure, known as morphology. The team used almost 100,000 3D microscopic images of melanoma cells to teach the AI technology about the biochemical changes that occur during treatment with various drugs.
“Our goal was to train a foundational model that could capture a wide range of biologically meaningful 3D cell shapes,” said Dr De Vries. “Melanoma cells were specifically chosen for training the model because they exhibit diverse shapes in response to both chemical and genetic changes.”
The researchers also had success using MorphoMIL in other cancer types, and they believe that the platform can be scaled further to support research across other complex diseases. Dr De Vries explained:
“In principle, any disease in which cells or tissues undergo detectable changes in shape, regardless of the cause, could benefit from this technology. Morphology isn’t just descriptive; it can be mechanistic, predictive and actionable.”
The tool’s uses are also likely to increase over time. Initially, the team’s priority is to use it to accelerate drug discovery. By predicting how a drug will work and how diseased cells will respond, MorphoMIL will give an indication of its likely efficacy in different patient groups. The tool will also enhance patient safety in clinical trials by predicting the extent of side effects and interactions with other drugs.
“We hope that our new AI tool will significantly reduce, and possibly even halve, the time it takes to trial a new drug,” said Dr De Vries. “Not only will this save time and money, but it will also make much-needed new treatments available to patients more quickly.”
In the longer term, Dr De Vries and his colleagues believe that MorphoMIL will have an important role in personalised medicine. He said:
“Although this technology is not currently intended for direct use in the clinic, the aim is to eventually use it for patient diagnosis or treatment planning. With further development and validation, many elements of this technology could support clinical decision-making tools. The ability to predict a person’s response to different drugs based on their specific cellular profile would help doctors select the most effective treatment for them.”
Challenges and considerations
Although the potential benefits of AI are clear, it does have limitations, and researchers will need to work hand in hand with the technology rather than relying on it.
“I can’t see skin cancer specialists becoming redundant,” said Dr De Vries. “AI should be viewed as a tool to augment, not replace, clinical or biological expertise. Our work is designed to help researchers and clinicians extract more information from complex datasets, speeding up discovery and supporting decision-making. We have no intention of trying to substitute for human judgement.”
Scientists who incorporate AI into their processes and projects must bear several factors in mind:
Data quality
Many research tools are developed using machine learning. This subset of AI involves inputting large quantities of data into a system that uses a series of algorithms to autonomously learn and improve, without being explicitly programmed.
The quantity and quality of the data used to train the system will therefore determine its accuracy. “Rubbish in, rubbish out” is a key principle in data science, acknowledging that unstructured, incomplete or inaccurate data will result in a poor output.
Another major concern is bias in data. Dr De Vries explained:
“AI has the potential to amplify existing disparities if not carefully developed and audited. It's essential that all models are trained and validated on diverse patient groups to avoid embedding demographic or biological biases into the predictions. We need to ensure that our tools generalise well across populations and contexts, and that the data they’re trained on reflects the full range of biological variation. We can’t just use what is easiest or most available to collect.”
Importance of human input
Although studies suggest that AI can be more accurate than people in certain areas, including the diagnosis of skin cancer, it is still not 100 per cent foolproof. Expert oversight is required to help reduce the risk of missed diagnoses and false positives.
For this reason, it is essential that AI models can be interpreted. Dr De Vries said:
“The internal workings of many systems are unknown, which can make it difficult to understand why a particular prediction was made. In biology and medicine, where decisions can influence research direction or patient care, this lack of transparency is problematic.”
To overcome this issue in their research, he and his team designed their model to prioritise specific parts of the input data – a technique known as an attention mechanism.
“This allowed us to trace the model’s reasoning, generate hypotheses about biological processes and build trust in the results,” said Dr De Vries.
Integrating AI with other technology
Progress in cancer research is reliant on scientists using a combination of tools effectively to enhance their findings. AI is not the only area in which we are making rapid advancements, and scientists should be using it alongside other impressive technology to accelerate biology.
“I believe we are in the midst of three converging revolutions,” said Dr De Vries. “Although the most visible is the rise of AI, two quieter yet equally transformative shifts are occurring simultaneously.
“The first is in microscopy, where we can now image cells and tissues in 3D at unprecedented resolution and scale. The second is in biology, where we can now engineer complex 3D environments that better replicate conditions in patients. Together, these revolutions form a trifecta that is reshaping our ability to study life and disease.”
“AI will be an indispensable part of the toolkit”
It’s clear that AI use has both significant advantages and serious risks. However, because it can support researchers and clinicians in finding ways to better predict, diagnose, treat and maybe even prevent skin cancer, it’s important that we fully explore its capabilities. Alongside this, Dr De Vries says we must understand its shortcomings and remember that the human brain is still vastly superior to technology in many ways:
“Cancer is biologically complex, and defeating it will always require collaboration between humans and machines. AI won’t ‘solve’ cancer on its own, but it will be an indispensable part of the toolkit.”