With artificial intelligence, it is often very difficult to discern signal from noise, but it has long been used by
oncology experts in various ways, and these uses will only increase as more sophisticated applications of the underlying technology are discovered and clinically approved.
When talking about AI, it is important to clarify that the various algorithmic systems used for clinical purposes are very different from the types of generative AI tools that have become the
next big trend for technology speculators and have received controversy for industrial levels of plagiarism and hallucinating answers.
Both of those aspects are not found in AI used in cancer diagnosis and treatment, which is vital since a mistake of the kind commonly seen with tools such as ChatGPT would lead to cancer patients dying.
Instead, cancer treatment has long relied on AI in one way or another, and whilst historically it has primarily been a part of the diagnostic and radiographic process, its use may expand as the technology
becomes more versatile and more promising.
From Expert Systems To Bakery Checkouts
The use of AI in healthcare more broadly dates as far back as the 1970s with the development of what were described at the time as expert systems.
Whilst intended to provide an expert opinion through the development of specific rules and extensive specialist datasets, they tended to function best with doctors who understood the limitations of the software.
John McCarthy’s lecture on
MYCIN, the first AI system used in medicine that specialised in bacterial infections, recounts a story of how one system diagnosed a patient with cholera and recommended the antibiotic tetracycline without any other considerations.
Whilst it would kill the bacteria that caused cholera, the patient would have died before this was the case, and there is an assumption that any doctor using it will also need to know when they should apply their own common sense.
The technology has improved considerably since then, despite multiple so-called
AI winters where development of the technology stalled, and is far more capable of producing effective diagnoses of conditions when given enough high-quality training data.
This began, rather bemusingly, with an AI system
initially designed to function as a bakery checkout.
BakeryScanAI was originally used to speed up checkout times at luxury Japanese patisseries, but an oncologist working in Kyoto applied the technology to scan for potentially cancerous cells, working with remarkable accuracy and speed.
The vast majority of medical AI systems are not developed in the same way. Instead, they use neural networks and machine learning that are trained on vast amounts of data until they figure out which features of a medical image are a marker for cancer and which are not.
Where Would AI Be Most Commonly Used In Oncology?
There are a lot of areas where AI would see use, but the primary areas are advances in diagnostics, risk prediction, telemedicine and pharmacy.
The biggest use for AI at present is in the former, particularly in the field of oncology and radiology, where an AI can help human doctors study medical images and highlight potential biomarkers which could be signs of cancerous cells.
The primary aim is not for an AI to be more accurate than a human oncologist, although several models are believed to be as accurate as healthcare professionals in identifying disease, but what they can do instead is process far more images at a rate faster than a human doctor.
The aim, therefore, is to reduce the workloads of radiologists and ensure that people who need urgent treatment are identified as quickly as possible in the system.
Early diagnosis
The key to successful cancer treatment is that the earlier it is diagnosed, the more likely it is that less invasive treatments are available, and the greater the prognosis for patients.
One particular challenge in oncology that AI is helping to solve is predicting which treatments would suit a patient best based on the characteristics of their tumours, as well as genetic aspects which could affect potential treatments.
Another aspect where AI can help is that it can improve and clean up medical images, which means that less powerful CT and MRI scanners could be used, ensuring that the technology is more widely available and less expensive.
One complication to these applications, besides the opaqueness of many AI models, is the cost to develop a suitable specialised trained AI model, but this could potentially change with the rapid development of lower-cost open-source alternative AI systems that could potentially be applied to the field of oncology.
