How is Artificial Intelligence (AI) shaping the future of cancer control?

From the early detection of lung cancer to the speedy analysis of brain tumours, artificial intelligence (AI) is increasingly being used in cancer detection, diagnosis and treatment. “AI has a huge potential to revolutionise how we treat cancer,” says Dr Laszlo Radvanyi, President of the Ontario Institute for Cancer Research in a recent Let’s Talk Cancer episode.

AI is also suggesting increasingly personalised cancer treatment plans. The Hôpitaux Universitaires de Genève, UICC members, are the first European hospital to adopt IBM’s Watson for Genomics artificial intelligence. By combining genetic data from patients with medical literature, this technology suggests personalised cancer care for patients. 

"The analysis takes just 10 minutes, compared with the 160 hours required for manual analysis. Clinicians can then use the time to be more precise in what they choose for their patients.”
– Rodolphe Meyer, responsible for the IT department of the Hôpitaux Universitaires Genève (HUG) in an interview with Labiotech.

But Dr Radvanyi explains that the new technology begs many questions: Will it replace doctors? Will it be safe? Will it be for everyone? “A lot of people don't understand it. A lot of people are afraid of it”, he says. 

Oncologist, radiologist: professions of the past? 

One concern is that doctors will be replaced. Pathologists trained to scrutinise tumour slides to determine the type of cancer and stage find that AI can now tell you not only the type and stage, but can also molecularly classify the cancer and suggest treatment decisions and other clinical management decisions based on that. Despite this, Dr Radvanyi emphasises that human intuition remains essential.

"Doctors will be the ones that dispense the therapy and make the decision of what we're going to do with the patient, not the computer. We need to approach AI as a powerful tool to help us make those decisions.” – Dr Radvanyi

Healthcare is based on human interaction. AI models cannot compete with trust in doctors and a desire for patient-doctor relations. One study in the United States found that 60% of participants said they would be “uncomfortable with a healthcare provider who relied on artificial intelligence to do something like diagnose their disease or recommend a treatment” and they “would not want AI-driven robots to perform parts of their surgery.” 

In radiology, a career that seems particularly threatened by AI, more time would be given to interpersonal exchanges, for instance explaining the results and treatment options to patients and staff.

“There’s a need for compassion in communication that AI is unable to contribute."
– David Dranove, professor of strategy at Northwestern University's Kellogg School of Management in a piece for KelloggInsight.

Furthermore, in countries such as Scotland where many radiologists are nearing retirement, there is a risk of chronic staff shortages and AI can fill the gap, allowing health systems to continue to ensure timely diagnoses. This is crucial because the earlier the cancer is detected, the higher the chance of effective treatment and survival.

Urging caution – safeguards to science

Machine learning can go beyond human capabilities. It can also go beyond human understanding. Scientists speak of the “black-box problem” when they do not understand the inner workings of an AI system. 

If the diagnosis or treatment proposed by the AI model turns out to be wrong, doctors will struggle to solve a problem they have not understood, nor is it clear who should be held accountable. Moreover, patients often want to be walked through their treatment plan. If the doctor cannot do this, it can lead to a loss of trust between patient and doctor.

AI for some but not for all?

Will AI increase or reduce inequities in cancer care between countries? Some countries are leading the race in AI development. China and the United States alone account for 80% of global investments in AI. These models are trained on data from their people, which may not represent ethnic compositions in other regions. This matters because patients with the same cancer may need different treatment. Gene mutations can vary by ethnic group; genetic factors can influence the way an individual metabolises chemotherapy drugs. If the data is not representative of different population groups, it may suggest a flawed plan.

We need to get better at inclusivity and diversity in these large genomic data sets to understand cancers in all demographics, in all cultures, in all races. – Dr Radvanyi

AI may offer solutions that go beyond a country’s resources, such as recommending a treatment plan that is inaccessible to a patient. 

"If you develop a clinical decision support tool in the West that is powered by Artificial Intelligence (AI), you can’t really take it to Nairobi for it to work, because it’s a different clinical context and different data that was used to train it.”
– Dr Kingsley Ndoh, UICC Young Leader and Founder of Hurone AI in an article for ODBA

But AI could also help bridge gaps in areas that lack specialist knowledge and resources. Only 3% of the world’s healthcare workers are in Africa and the problem is only getting worse as more doctors search for work abroad to earn a better living. Six out of 10 Nigerian doctors plan to emigrate, according to the Nigerian Association of Resident Doctors. Patients in rural areas are particularly affected, needing to travel far to find a qualified doctor. Through chatbots and mobile apps, AI can help triage patients, facilitate remote appointments, and offer initial health advice based on symptoms.

What about inequities within countries? Evidence is growing that AI algorithms are more accurate for some groups. One study in the US found that an early-stage breast cancer test was less accurate for black people than it was for whites. The United Kingdom’s National Health Insurance (NHS) is exploring the use of AI to help dermatologists treat patients with cancerous lesions. But there are too few images of people with darker skin, for whom the tool becomes inaccurate.

AI reflects the quality of the data it is trained on. Algorithms are “imbued with the values, choices, beliefs, and norms of their developers and of those who assemble the datasets,” write the authors of the study “Artificial intelligence in health care: laying the Foundation for Responsible, sustainable, and inclusive innovation in low- and middle-income countries.” Even without AI, healthcare providers would often act according to preconceived ideas.  A study by the National Center for Biotechnology Information found that many providers have implicit biases that favour white people. Done well, AI could help overcome this human bias by providing more data-driven solution for care. 

“AI is not a silver bullet,” says Patricia Lee, “Future of Health” Deloitte, speaking at UICC’s 2022 World Cancer Congress. It is still far from being integrated in all aspects of cancer care and many have valid concerns about cost, inequities and biases. But through further research, collaboration between regions and the implementation of safeguards, AI could be a force for good in the diagnosis, treatment and prevention of cancer. 

Learn how AI will be feature of the World Cancer Congress