Artificial intelligence (AI) refers to a computer system capable of simulating human learning, understanding, problem-solving, creativity, decision-making and autonomy.
AI has become a hot topic of public interest and has aroused much unease about potential abuses of the technology. General fears about AI technology include There are concerns about job displacement, ethics, fears that it will invade privacy and existential worries that AI could act independently of human control.
However, we already know this new AI technology can be a great boon when properly regulated and one example of this is the use of AI in medical diagnostic imaging, eg X-rays. Another example of the beneficial use of AI is in the early diagnosis of Parkinson’s disease. This whole area is reviewed recently by Henry Miller in American Council on Science Health.
The use of AI in medical diagnostic imaging is closely regulated to ensure balance between safety, security and innovation. Regulation is carried out by the Food and Drug Administration (FDA) in the United States and by the Health Products Regulatory Authority (HPRA) in Ireland. However, many unregulated applications of AI have been applied in areas outside of healthcare, such as ChatGPT. These developments are causing much unease and eliciting calls for new stringent regulation.
Miller sensibly cautions, however, that public policy should not paint all applications of AI with the same brush. Regulatory frameworks already working well to balance safety, security and innovation in medical diagnostic imaging should be left alone. Action should focus instead on implementing necessary oversights in areas where AI applications pose new and unmitigated risks.
[ In the age of conversational AI, ‘scholarly bullsh*t’ floats to the surfaceOpens in new window ]
AI medical imaging technology is regulated as rigorously as all other medical devices. This includes premarket review of safety/effectiveness and post-market review monitoring ongoing performance of all approved imaging devices. A quality management system must be maintained and strict labelling and detailed information provided to intended users.
It has been recognised for some time that AI can greatly enhance the efficiency and accuracy of information extraction from diagnostic medical images when used to supplement visual examination alone. The most basic consideration tells us that critical visual examination of many images in succession will quickly tire out the examiner and, as more and more images are examined, will lead to missing important information and coming to erroneous conclusions.
On the other hand, computer-based AI doesn’t tire out and will analyse its thousandth image with the same accuracy as it did with the first. In addition to limitless stamina, AI can also simultaneously analyse vast amounts of data and reach conclusions quickly as compared with the slow processing of visual intake by the human examiner.
Many of us have had a colonoscopy, a medical procedure to examine the large intestine (colon) and rectum for abnormalities, such as polyps, swollen tissues or cancer. The procedure involves inserting a flexible tube, the colonoscope, up through the rectum to examine the interior of the colon. Colonoscopies are necessary for diagnosing gastrointestinal disease and help in prevention and treatment of problems such as cancer.
[ William Reville: We must defend our academic freedom or we will lose itOpens in new window ]
Miller reports results on the performance of a new AI tool called “GI Genius” that detects abnormalities such as polyps and precancerous lesions in the colon in real time. A 2021 trial on 700 subjects undergoing colorectal screening by colonoscopy showed that GI Genius identified precancerous lesions in 55.1 per cent of patients compared with 42 per cent of patients assessed by standard colonoscopy. And GI Genius continues to improve its performance as more examples of colonoscopies are added to its database.
Parkinson’s disease (PD) is a degenerative disease of the neurological system. One million people in the US have PD and 60,000 new cases are diagnosed annually. Early detection of PD, allowing the earliest possible palliative interventions, is difficult but a new AI program may change that.
Breathing is affected in PD cases from the earliest stages and a recent study showed that AI can sensitively detect nearly 90 per cent of PD patients by analysing their breathing patterns from the movement of “breathing belts ” ie by measuring changes in the size of belts placed around the chest or abdomen of a sleeping patient. Since no other approaches can detect PD or assess its severity from breathing, AI can provide insights that are inaccessible otherwise.
William Reville is an emeritus professor of Biochemistry at UCC
