With the infiltration of artificial intelligence (AI) into medicine, it is possible that radiologists may have more to either gain or lose with the technology.
AI in Radiology
Even insiders have varying opinions regarding the extent to which radiologists should welcome the technology. According to Dr. Ronald Summers, an AI researcher and radiologist from the National Institutes of Health, there are some AI techniques that are quite good, adding that some of them should already be included in the practice. Dr. Summers also questions why they are allowing such data to just sit over the table.
The lab of Dr. Summers was able to develop imaging programs that are computer-aided and that can detect osteoporosis, colon cancer, diabetes, and other medical conditions. None of these have seen wide adoptions, which Dr. Summers attributes to medicine's culture, alongside other factors.
Since the 1990s, radiologists have been using computers for enhancing images and flagging suspicious areas. However, advanced AI programs are capable of taking this a step further, as they can offer diagnoses and even draft written reports regarding the findings. Such algorithms are typically trained with millions of X-ray scans and other images from clinics and hospitals.
The Food and Drug Administration has granted approval to over 700 AI algorithms for helping physicians. Over 75% of these algorithms are for radiology, though only 2% are applied in radiology practices, based on a recent estimate.
ALSO READ : AI Can Reveal How Old You Are Using Chest X-Ray; How Can Radiography Be Used as Biomarker of Ageing?
Skepticism of Radiologists
Despite the promising potential, radiologists have various reasons for being skeptical of these programs. These include the lack of transparency regarding their work, limited testing in actual settings, and questions regarding the patient demographics used for the AI's training.
Dr. Curtiz Langlotz, a radiologist who also runs an AI research center, says that if they are unaware of the cases that the AI was tested on or if these cases match the types of patients that can be seen in actual practice, this could raise questions regarding whether the technology could really work.
Back in early 2020, the FDA conducted a workshop to talk about algorithms that could work without the oversight of humans. Briefly after this took place, professionals of radiology warned regulators that they think that it is premature for the FDA to think of clearing or approving these kinds of systems.
However, in 2022, European regulators approved the first fully automated software that can review and write reports for healthy and normal-looking chest X-rays. Oxipit, the company behind the software, is applying to the FDA in the US. Such a need is urgent in Europe as some hospitals face long backlogs due to radiologist shortages.
Across the US, it is likely for such automated screening to be years away, mainly because radiologists are not yet comfortable with entrusting even routine tasks to such programs.
However, according to Chad McClennan, the CEO of Koios Medical, which provides an AI tool for thyroid ultrasounds, says that radiologists usually overestimate their accuracy. In fact, their company's research found that doctors who looked at the same scans were in disagreement with each other over 30% of the time on whether they should perform a biopsy.
The National Cancer Institute also notes that roughly 20% of breast cancers get missed during usual mammograms.
AI May Aid Radiologists
Experts say that in the near term, the technology could function like a plane's autopilot system that performs vital functions under a human pilot's supervision. According to Dr. Laurie Margolies of Mount Sinai Hospital, this approach could reassure both patients and radiologists.
The first rigorous and large trials that involved testing radiologists who were AI-assisted against those who work independently offer hints of possible improvements. A Swedish study covering 80,000 provided initial results that show that a lone radiologist who worked with AI was able to detect 20% more cancers in mammograms. This is in comparison to what two radiologists who did not work with AI found.
The study also found that making use of AI rather than a second review decreased the workload of humans by up to 44%. However, the lead author of the study says that it is crucial for a radiologist to have the final say in every case.
However, if an algorithm ends up missing cancer, this would yield tremendously negative trust for the caregiver. This was noted by Dr. Kristina Lang from Lund University. Moreover, in such cases, the question regarding liability is also part of the legal issues that still require resolution.
Check out more news and information on Tech & Innovation in Science Times.
