FEATURE: LIFE SCIENCE
“For a physician, trusting an intelligence whose reasoning is unknown is a big leap”
questions to address, they say, in terms of legal liability – what if something goes wrong? Who is to blame? Another issue is that the sheer
Institute, a collaboration between health services and academia in Sheffield, UK. In addition to the digital twinning of a patient’s physical body, allowing for a holistic rather than organ-by-organ approach to treatment, the institute’s scientists are now exploring the possibilities of in vivo microscopy (IVM). IVM takes place within the patient’s
body while clinical procedures are being carried out, so they can provide a real- time data stream that reveals what’s happening on a cellular level. The tomographic images can be 2D or 3D, and the equipment that captures them might be coupled to the ports of an endoscope or introduced on a standalone basis. By combining IVM data with the holistic
perspective provided by tools such as the Virtual Patient and organ-level computer modelling, the Insigneo researchers aim to achieve a continuous view of human physiology that spans everything from sub-molecular activity to whole-organism health. This presents some amazing possibilities, such as finding correlations between a bodily symptom and a cellular process or connecting the dots between seemingly unrelated pathologies.
Diabetic retinopathy screening Diabetes is an insidious condition that takes a toll on many different areas of the body over time. Diabetic retinopathy is a case in point. It develops slowly but can lead to sight loss if left unchecked – in fact, it has been the leading cause of new- onset blindness in adults aged 20-74. The medical field’s best defence is to
screen for diabetic retinopathy in the hope of catching it early so it can be stopped or slowed down, and healthcare guidelines
such as those of the American Academy of Ophthalmology recommend annual screening for diabetes patients. But with so many people at risk of the condition, all this screening represents a huge volume of skilled work. AI has been used for many years to help ease the burden, with feature recognition tools spotting abnormal tissue in images from eye examinations. Nowadays, deep learning, where an
algorithm refines its abilities through repeated iterative feedback loops, is overtaking feature recognition as the ophthalmological field’s AI tool of choice. A 2018 review of deep learning AI performance in diabetic retinopathy screening found that “AI devices may significantly reduce the screening burden of DR worldwide, but additional knowledge gaps need to be addressed to ensure the effective use of this new technology”. However, it still has some way to go. A
study in Western Australia pitted a deep learning AI against four ophthalmologists in a clinical setting over a six-month period, and found that while the AI did pick up the two cases of diabetic retinopathy presenting at the clinic, it also flagged 15 false positives, while the ophthalmologists made none.
Navigating legal issues in healthcare- sector machine vision Like any sector, medical care comes with its own legal universe, one that’s been given a shake-up by the arrival of computer vision tools. Zach Harned, Matthew P. Lungren & Pranav Rajpurkar consider the emergence of AI and machine vision in healthcare from a legal perspective in a 2019 commentary, which we accessed as a preprint. There are big
20 IMAGING AND MACHINE VISION EUROPE DECEMBER 2023/JANUARY 2024
complexity of the technology tends to make it opaque to humans. Researchers do note the “black box” effect of AI software models, which can be so complex that even the designers who create machine vision tools for healthcare don’t know exactly how they work. For a physician, trusting an intelligence whose reasoning is unknown is a big leap, especially where human health is at stake – not to mention the malpractice claims that might arise. Harned et al. also touch on the important issue of interpretability. When images are perceived by a machine and the data is used to draw conclusions about healthcare, it’s critical for medical professionals to be able to see what the machines see, so they can ‘check their work’ and correct any errors. Fortunately, recent advances in interpretability are making it easier for physicians to do just that. “We conclude that the unique capabilities and functions of AI and machine vision, especially when conjoined with the aforementioned advances in their interpretability, create an opportunity to argue that the technology actually minimises physician liability,” say the authors.
Not agents, just very powerful tools It seems clear that while AI and machine vision hold unprecedented power to transform our healthcare, they remain tools, not agents, in the ongoing mission of keeping us well. The sci-fi spectre of the autonomous robotic arm wielding a scalpel is still the stuff of cinema – for now at least. In skilled and responsible hands, machine vision can carry out some of the aspects of care that drain our time and resources. It can also shed light on what’s happening, both in our bodies and in our lives, in ways that a subjective human brain cannot. However, we cannot lose sight of its limitations. If designed and used with the level of care and caution we expect of the medical profession as a whole, and we are able to check and question it, it has exhilarating potential. I
@imveurope |
www.imveurope.com
Shutterstock/greenbutterfly
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32
