YOUR PARTNER IN DIABETES
+27 11 712 6000
This debate has raged on from as early as the 1980sand there are strong arguments for both sides. I cannot help thinking of ‘Skynet’ in the ‘Terminator’ when I think of artificial intelligence (AI), so I naturally have some degree of suspicion at the thought of AI in health. However, when I started writing this blog, I realized that AI in radiology is like everything in life… it is all about the balancing act!
There are two subsets of AI: machine learning and deeplearning.Machine learning involves algorithms using patternrecognition to solve tasks, for example recognizing symptoms of breast cancerin lung scans. Deep learning involves neural networks using the human brain model and these algorithms are used for creating medical images. AI is already playing an important role in health and indeed in diabetes management. Examples of this include the amazing new technologies in our current insulin pump options or the fundal cameras paired with AI interpretation we use at the CDE for retinopathy screening.
AI technology has assisted the radiology departmentsin a variety of ways. AI algorithms are already being implemented to assistwith diagnostic and treatment decisions especially in evaluation of breast, prostate and brain tumours. It is noted that 45% of radiologistsexperience burnout, even before the COVID-19 pandemic. The dedicated radiologist is expected todo tedious administration, thus losing time to less-challenging tasks and leaving limited time for the more complex cases. AI can assist with the simpler cases leaving more time for challenging cases - this has assisted with productivity in radiology departments. In addition, there is a positive outcome for cost effectiveness with AI cutting costs and improving efficiency.
Then do we need the radiologist if AI is accurate,cuts cost and time?
The answer is a resounding yes!
The radiologist does not only identify abnormalities found in images; he/she integrates the history, physical exam, and blood results of a patient and is often involved in discussions with the treating physician. This subtle nuance cannot be learnt by AI.While the accuracy of AI is not being debated here, the accuracy versus a radiologist is indeed an issue. A recent report from the BMJ compared a United States Food and Drug Administration-approved AI tool with radiologists who passed the specialty radiology examinations. The radiologist’s accuracy was 84.8% versus AI of 79.5% as yet.
So, what is the role for AI in radiology? I think ultimately the role will be somewhat like the role of autopilot and a pilot flying a plane. AI will make life easier, reduce stress, improve time usage and improve the workload. However, the pilot will never be redundant. One still needs the human element to prevent disasters! The radiologists’ experience and non-linear thinking is needed to make the system work. The radiologist and AI together will become the gold standard of diagnostic care.
Geoffrey Hinton, winner of the Turing award and considered one of the pioneers of deep learning, said in 2016: “We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better thanradiologists.” I am pleased to say that 7 years later this is profoundly untrue. I rely on my colleagues in the radiology department for thorough analyses of patient imaging with my history and examination and discussion about the person which ultimately influences his or her management. Radiologists are invaluable in the patient care team and are often the unsung heroes behind the scenes.
References