The use of AI in medicine is developing so rapidly that almost every week there is a new and somewhat startling announcement. One of the major uses is to review radiology scans, such as CTs, to find worrisome findings that may be missed by human eyes. Researchers at Harvard Medical School and Massachusetts General Hospital have been using an AI tool they named “Sybil” to forecast lung cancer risk based on a single low-dose CT scan of a person’s lungs. But, can artificial intelligence predict lung cancer?
Sybil was able to predict lung cancer within one year to six years with remarkable accuracy. The researchers found that this performance remained consistent for various groups of people, despite sex, age, and smoking history.
In an editorial in the Journal of Clinical Oncology, two oncologists noted that the model’s overall performance was outstanding. They noted that the tool allows minimum work-ups and invasive testing for lung nodules that Sybil predicts won’t lead to future cancer. It also predicts which patients can wait longer between screening tests or even avoid screening entirely if Sybil predicts a very low risk of developing cancer.
The research project involved more than 13,000 CTs from 6,392 adults who underwent lung cancer screening from 2015 to 2021 at Massachusetts General, and 12,400 scans from more than 10,000 adults at a major hospital in Taiwan. Patients who participated in both studies did not have a personal cancer history or even a history of smoking. However, because the Taiwan part of the study did not include detailed smoking data, the ability to predict lung cancer in non-smokers remains more speculative.
This was interesting to me on a personal level because one of my children who is studying for her PhD in mathematics is involved in a class at her university that is studying CT scans of infants to try to determine what information can be found which can predict future health problems. We are apparently evolving rather rapidly from the AI used in rhumbas and other simple robotics to AI that can replicate and improve on medical decision-making. It’s a whole new world out there.
