Oxford University Press's
Academic Insights for the Thinking World

Artificial intelligence in oncology

There is no denying the presence of computers in our everyday life, whether it’s through phones, personal virtual assistants such as Apple’s Siri and Amazon’s Alexa, or video games. Lately, the interest and development surrounding artificial intelligence (AI) has escalated, and the opportunities to embrace this within the healthcare industry seem to be growing. I’m not talking about conventional robots—although robotic assisted surgery is on the rise—but the inclusion of high-speed, deep learning computers to aid diagnosis and treatment.

The growth of medical knowledge is far from slowing and is expected to double every 73 days by 2020. In 2017, in excess of over 80,000 oncology papers were published, according to indexing service Web of Science. This sheer volume of research makes it impossible for physicians to stay up to date with the latest advances in the field and so clinical decision support systems are built to work alongside doctors to ensure the latest, highest standard of care is being given. For breast cancer alone, there are 69 different approved drugs for standalone treatment. Add this to the number of combination treatments available, and oncologists face a memory test on top of their routine tasks. The focus for AI separates into two sectors—to improve speed and accuracy of diagnosis, and to increase efficiency of drug discovery.

Developed by IBM, Watson Health is an AI clinical decision support system able to analyse data across many health sectors including cardiology, drug discovery, and diabetes. Watson for Oncology (WFO) focuses on rapid diagnosis and optimum treatment proposals for cancer patients. Researchers in India recently found that WFO treatment recommendations for breast cancer patients were concordant with those of an expert panel of oncologists. In 93% of the 638 cases presented for analysis, the tumor board agreed with the WFO proposals. The same tumor board also found concordant results in lung, colon, and rectal cancer. Discrepancies were found in cases where different treatment options were available in India compared to the training location of WFO in the United States. As with most cancers however, there is rarely a right or wrong answer, and treatment proposals are determined by what clinicians believe to be the best fit for the patient.

“The issue gets more complicated when moral and ethical decisions are concerned. Can we train computers to have emotion too?”

Yet, all these developments don’t come without their limitations. WFO pulls data from patients’ records, clinical guidelines, and studies as well as journals, books, and articles. In spite of this, the system is unable to make recommendations based on its own understandings from the data, and instead relies on preliminary training from real-life physicians and analysts at the Memorial Sloan Kettering Cancer Center in the United States. The initial recommendations for treatment were scored and feedback was input back into the system to improve concordance. Consequently, machine learning is essentially only as good as the data and knowledge you use to train it, so a lot of responsibility rests with the experts.

The issue gets more complicated when moral and ethical decisions are concerned. Can we train computers to have emotion too?

As a result of advancements in data processing, AI has progressed leaps and bounds in recent years, with regards to being able to read and process human emotions. However, understanding emotions and expressing empathy is entirely different to just being able to decipher emotion. As of yet, it is still unclear whether WFO is able to take these facts into account and whether recommendations are the best they can be morally.

AI is being recognized as an emerging area for research and Cancer Research UK’s Grand Challenge award includes AI as a competing category for £20 million of funding. There are other oncology AI systems that are currently being used in various hospitals around the world as well as an increasing number of new launches, such as Eureka Health Oncology, ACSO’s CancerLinq, and Healthcare NExT from Microsoft. All systems primarily focus on the same goals, and it’s hard to tell which will be the most successful, especially when there are lots of improvements to be made. For now, we should be aware that AI is increasing its presence in the field and can be used to increase efficiency and accuracy in cancer treatment.

Featured image credit: “Computer, Business, Typing” by Free-Photos. CC0 via Pixabay.

Recent Comments

  1. Sriganesh Srihari

    Interesting. Have you looked at maxwellmri?

Leave a Comment

Your email address will not be published. Required fields are marked *