Can AI be Used to Determine Cancer Recurrence?

Benedict Pignatelli

In 2013, an AI was developed in Japan to help out in a bakery. Its role was to differentiate different kinds of pastries from one another, and as such augment the role of the cashier, allegedly making the process quicker and more hygienic for all involved. By a bizarre turn of events, this AI was able to detect cancer cells.

According to the New Yorker, after the AI completed its career change, swapping a baker’s hat for a stethoscope, it boasted a 99% accuracy rate.

Now, AI can not only identify cancer cells but also predict cancer recurrence in patients in remission.

Why should AI be used in Cancer Research?

As a general rule, the faster the cancer is spotted, the better the chances it can be treated successfully. Evidence strongly suggests that the earlier the cancer is identified, the higher the percentage of full recovery. Artificial Intelligence – if working effectively – can learn and study at speeds far beyond anything a human could do.

A tumour not being seen due to oversight or human error could be mitigated with the help of AI. Machine learning (where computers learn various data patterns and algorithms and make predictions from them), could be revolutionary for early cancer diagnosis.

Cancer recurrence

One of the great ways we can see the potential benefits of AI in cancer treatment is preventing, or at least diagnosing, cancer recurrence. A constant risk when treating cancer is that it may come back, causing further health risks after the treatment is completed. Similar to spotting the initial cancer, the quicker these recurred cells are found, the better the chance of recovery.

Monitoring patients after treatment is thus vital to ensuring any cancer recurrence is acted on as soon as possible. Due to several factors, this process can often be delayed or imprecise.

In its efforts to survive, the cancer often tries to evade the immune system’s responses, and can even resist the effects of the treatment. A team from the University of Wisconsin have developed a test, utilising AI, to detect genetic abnormalities and to determine which patients’ cancer is more likely to return.

Although the results were only marginally more accurate than traditional testing systems, the tool has the potential to massively improve cancer recurrence detection if given the proper time and resources to learn.

A successful AI tool could detect cancer recurrence earlier in high-risk patients but also would mitigate a lot of the unnecessary, time-and-money-consuming follow-up trips to the hospital for those who are low risk. For instance, rather than having to take up clinical time in the hospital, an AI could do the job at the home of the patient.

‘This is an important step forward in being able to use AI to understand which patients are at highest risk of cancer recurrence’, commented Dr Richard Lee, a consultant at the Royal Marsden NHS Foundation Trust.

Pros and Cons of AI

The positives of AI in cancer treatment are obvious – with enough resources to train the AI properly, it could drastically reduce the time and effort that goes into spotting cancer, and cancer recurrence.

‘Cancer is an umbrella term for thousands of different types of conditions, yet treatment offered today is often generic and does not consider the need for differing therapies for different people,’ says Professor Toby Walsh, a leading expert in artificial intelligence.

Therefore having an AI that has accumulated vast amounts of medical knowledge and data, on a scale ‘impossible for humans alone to achieve’, would greatly benefit the process. Walsh went on to argue that using machines to encapsulate the knowledge of experts, and interpret the data quicker and better than the specialists, will create a new understanding of cancer and cancer treatments.

There are some concerns with AI in cancer treatment. The actual success of the AI has come under scrutiny. Essentially, the issue is many AI systems are not as intelligent as the name would lead us to believe. This is a flaw that will no doubt improve exponentially with time, however, it can be an argument that the investment in AI is not worth it at this time.

For example, researchers from Google AI have presented results for detecting cancer, championing the AI, and arguing it has better results than radiologists. Nature argued the results were exaggerated. While it had promise, the tool would need further validation and training.

In 2017, Sophia Genetics was attempting to partner with five different UK healthcare institutions to collect data and find what treatment is working best on what patient, using AI to work at machine speed, with the hope of reducing waiting times and providing cheaper and earlier diagnosis. Their main issue was a lack of organisation regarding their data.

The NHS was criticised for ‘not being proactive about using the technology in a proper way’. A key reason for this was its disorganisation, which stopped the AI’s progress like a brick wall. Until the NHS and other healthcare bodies become more digitised and centralised, it will be difficult to properly utilise the benefits of AI.

Another key issue then is the data gathered; it is important that GDPR compliance is adhered to, as failure to do this would not only be illegal but a serious security risk to UK citizens. Aside from the risk of cyber crime, there is the need to safeguard against corporations monopolising on patients’ data. Hospitals would be responsible for protecting the patients’ data and finding a way to use the AI without it being a risk.

Overall – Lots of Potential, Needs Work

There are people attempting to solve the dilemma, utilising the benefits of AI while still keeping patient data protected. Swarm Learning, for instance, has created an AI that can detect patterns in data without having to send any local data or patient information. Phil Quirke, Professor at the University of Leeds’s School of Medicine, commented on the importance of this tool, as it ‘improves our ability to apply AI in the future.’

To anyone who has watched their phone’s woeful attempt at text prediction, it is perhaps unsurprising that the AI ‘needs further validation’ before it can overtake radiologists. Machine Learning and AI still have a long way to go. Bringing completely new technology into national healthcare will, of course, take time and effort, but the potential is enormous.

There is no doubt about the benefits of AI and the possible future breakthroughs doctors could achieve when augmented by artificial intelligence. The speed and accelerated learning of AI means the most cutting-edge cancer treatment could be used all over the world, including in less developed countries where this would otherwise not be possible.

About the author: Benedict Pignatelli is a contributing writer from Dublin, Ireland. He studied World Religions and Arabic Language, and has an interest in Middle Eastern politics. He also writes fiction and was longlisted for the 2019 Bridport Prize.