AI has huge potential in the diagnostics space — from improving sensitivity and specificity, to enabling clinicians to assess and measure dimensions that were not previously feasible or practical.
More accurate and consistent diagnostic results will then herald earlier arrival at correct diagnoses as well as far better and faster methods of tracking disease and treatment response.
Altogether, the promise of AI in diagnostics is improved outcomes for patients, more certainty for clinicians, and reduced waste for the entire system, not to mention freeing up precious time so that doctors can spend more time with patients and experience less burnout.
However, this bright future comes with natural disruption: AI threatens many established practices and will cause evolution of the role of clinicians devoted to diagnostic specialties.
Hype and Fear
A lot has been written about AI in medicine, including predictions that AI is about to replace physicians. We see claims that algorithms are ‘beating’ clinicians at several tasks, stoking fear among physicians working in the areas that are most “threatened”.
Compounding this, hundreds of companies large and small claim to be transforming healthcare with AI. Indeed AI will fundamentally
change medicine, though the changes will come slower — and likely much less cataclysmically — than many have predicted.
Today, most healthcare applications of AI are focused on improving operations, typically with use cases such as automating scheduling and billing. In contrast, clinical outcomes and decisions have yet to be significantly impacted.
While this is disappointing to many who expected quick change, clinicians have been, as usual, waiting to see the evidence before judging AI as positively impactful.
For example, just because an algorithm can have higher sensitivity or specificity (rarely both) than a clinician, we don’t yet know whether that will result in improved diagnostic or therapeutic decisions.
Furthermore, the tasks for which AI is being applied are quite specific, covering only a very small portion of what the clinician does when assessing their patient’s condition.
At Arterys, we have automated tasks that physicians find tedious and frustrating, and we preserve their ability to visually verify that the output is correct. Furthermore, physicians can edit these outputs if needed. Thus, we are expediting volumetric measurement and tracking of anatomy and disease, without taking control away.
The difficult road ahead
There are many challenges to unlocking AI’s enormous potential. A well-known obstacle is the lack of availability of well-curated, diverse and accurate ground truth data. Datasets that reside in individual institutions tend to be limited by several biases, from the equipment used to the population served.
Another roadblock that is often mentioned is the lack of appropriate regulatory pathways for new technology. In many geographies, software that can be updated easily continues to be regulated under processes designed for hardware that have historically changed far more slowly.
Two additional areas that get less attention surround the implications on clinicians and institutions that adopt AI. First is the role evolution of clinicians, especially as they make decisions based on information that may not be easily explainable.
They will need to learn how to interpret this information. Furthermore, physicians and, institutions will need to devote significant thought when it comes to how they will consume and generate real world data.
Second, new challenges around the legal ramifications of AI systems will be sure to arise. Even when physicians continue to own and be accountable for clinical decisions, we will need to better define the liability around CDS or automated diagnostic information. We need to create ways to ensure clinicians are confident and comfortable with their decisions.
The industry will be well served to stop trying to ‘beat’ clinicians and replace them. Instead, we would be most impactful and most supportive to health care delivery if we build systems that address clinical needs and produce significant improvements. Collectively, we need to do a better job at understanding what automation is most beneficial, and what implications that automation has on patients and healthcare providers.
By Fabien Beckers, the CEO and co-founder of Arterys,
A cloud/deep learning startup that is disrupting the medical imaging space and building image-based precision medicine tools. Fabien has led the growth of the company from four co-founders to a team of 90 today. Under his leadership, the company has become a pioneer in cloud-based medical imaging software, offering the first FDA-cleared end-to-end cloud infrastructure for medical imaging. They key advantages of the platform being automatic aggregation of real-world data and ability to scale and distribute the processing of increasingly large, complex datasets. Fabien’s vision for the company is to accelerate data-driven medicine by building precision medicine tools based on the consistent quantification of medical image features in combination with molecular, genomics and patient history data. Fabien holds a PhD in Quantum Physics from the University of Cambridge and a masters of business from Stanford University.