As literary comparisons go, this feels apt. The curiosity with which many parts of the healthcare sector are adopting the advancement of technology (or at least observing with intent), is akin to Alice’s exploration of Wonderland. Grinning cats aside, curiosity is powering the use of AI in wide-ranging health applications – from drug discovery to personalised treatment – in our quest to drive efficiencies and progress. In the same breath, there’s a sense of urgency to reaffirm our place, as humans, in the age of technology.
The use of AI in medical imaging is nothing new. However, advancements in AI (deep learning techniques using large data sets) mean its role in imaging and diagnostics has taken new significance. With potential to increase productivity, recent mainstream media attention is no surprise. We’ve heard about companies and healthcare providers readily using AI to screen images, flagging up anomalies as priorities for radiologist input. When early diagnosis is crucial for improved outcomes, the benefit of speedy image analysis and reporting (think seconds versus days) is clear, with the reassurance that human eyes and experience make the final call. The knock-on effect of this positive impact on physician workload is more patients can be seen, quicker. That’s a leap forward – for patient outcomes, as well as the healthcare ecosystem.
But let’s be curious.
AI tools are only as good as the data they’re trained on. While this may be huge data sets, the existence of health disparities tells us that algorithms must have inherent bias. What this means in imaging is a risk of misdiagnosis, or delayed diagnosis, if a patient’s demographic is not adequately represented in training data sets, such as case images. While human safety checks are currently in place, the potential magnification of health inequity and consequences yet to be unearthed is a sobering thought, as technology continues to advance and become a mainstay of clinical practice.
More proof that better representation across all remits of healthcare is needed, as we navigate the boundaries of AI and its application in our world.
This thought leadership piece appeared in the May edition of PME. Read the full issue here.
