Horse-drawn carriages were a primary mode of transportation from the invention of the chariot in about 2000 BCE1 to the advent of the first steam-powered automobile in 17692. Imagine the anxiety of buggy manufacturers, with the inevitable shift to horseless carriages. Forward-thinking industrialists seized the opportunity to adapt and flourish, transitioning for example to making wheels for automobiles rather than wagons. Yes, disruptive technologies can be game-changers. However, as uses of artificial intelligence (AI) in medical imaging emerge, it is becoming clear that this advance will bolster the vital roles of the radiology team, rather than replacing human jobs.
One concept/many applications
“Artificial intelligence is the science of training machines to perform human tasks”3 – teaching machines to mimic some human capabilities, but not imparting human-like qualities. Machine learning is a subset of that broader science. Rather than the machine being told what to do (as in earlier analytical algorithms), the machine is provided with examples. It learns from those illustrations what conclusions are appropriate, then applies that knowledge to new situations.
Deep learning is, in turn, a subcategory within machine learning, where an artificial neural network applies multiple layers of algorithms to extract information from raw data.
Bradley J. Erickson, MD, Ph.D., associate professor of biochemistry and molecular biology at the Mayo Clinic College of Medicine in Rochester, MN, described it.
"Deep learning is the hot new technology that is a more specific form of machine learning, with one major difference being that we do not have to calculate the important features in the examples that it should use for making decisions."4
AI encompasses the power to efficiently analyze and provide meaningful insight from vast datasets, an immensely beneficial trait in medical imaging considering the continually increasing number of images per exam.
It can also identify patterns within data that are not clear to humans. The concept can be applied to routine tasks such as scheduling, for workflow improvements, and precise interpretation of imaging data. For example, a deep learning algorithm applied in image reconstruction may detect subtle lesions that may not be obvious to a radiologist, or spot molecular markers that cannot be perceived by humans.
Why we still need the radiology team
AI systems under development perform single tasks – called narrow use AI. The American College of Radiology's Data Science Institute (ACR DSI) believes AI is most useful in healthcare in this context-specific image recognition such as hemorrhage on brain MRIs or nodule detection on chest CTs.5
ACR DSI's Bibb Allen, MD, FACR, CMO said, "If these use cases are developed in coordination with input from physicians, so that the inputs and outputs can be appropriately defined, AI will significantly improve that care we are able to provide our patients as part of our clinical workflow in disease characterization, disease detection, and standardization of reporting. It will be an evolution built on the successful development of AI use cases. Clinical practice integration in the long term will include linking multiple use cases together to solve increasingly challenging problems."5
Regulatory and legal systems need to catch up with AI in healthcare. How will insurance respond to use of AI? Who is responsible for a misdiagnosis involving AI? Without legal precedent, the liability thread could lead back to the data scientist who designed the algorithm.5
Most importantly, there is a common misperception that a radiologist spends his or her entire workday reading and interpreting images. While those tasks may be among the most intense and critical, the radiologist has many other responsibilities such as:5
- Tailoring exam parameters to the patient’s condition
- Explaining procedures to patients
- Correlating imaging findings to related tests and medical records
- Discussing diagnostic results with patients and family members
- Consulting with other physicians regarding diagnosis and treatment
- Providing certain treatments and therapies
- Performing interventional radiology
Naysayers do not fully recognize the advantages of AI tools in helping an integrated healthcare team identify the need for screenings, triage interpretation by patient need, characterize disease, and incorporate results into standardized reporting. AI's capabilities would have to become substantially better than – not equal to – its human counterparts before technology could assume all image reading and interpretation tasks.5
Roles they are a-changin’
Automobiles did not replace transportation jobs; the new technology opened up opportunities for moguls in the field. The same is true of autonomous vehicles (another use of AI, by the way). Likewise, AI, machine learning, and deep learning will not eliminate human functions in radiology. Those roles will, however, naturally need to adapt to the use of technology, with the overarching goal of increasing the value of radiology inpatient care.
Which radiology jobs may be threatened? Those of individuals who refuse to work effectively in tandem with AI.
Which facilities and providers will emerge as leaders in healthcare? Early adopters of AI technology.
- Chariot. Ancient History Encyclopedia. https://www.ancient.eu/chariot/ June 16, 2019.
- Who built the first automobile? History.com. https://www.history.com/news/who-built-the-first-automobile June 16, 2019.
- Artificial Intelligence and Machine Learning. SAS Institute Inc. https://www.sas.com/en_us/insights/analytics/what-is-artificial-intelligence.html June 16, 2019.
- Putting the AI in Radiology. Radiology Today Magazine. https://www.radiologytoday.net/archive/rt0118p10.shtml. June 16, 2019.
- AI Will Change Radiology, but It Won’t Replace Radiologists. Harvard Business Review. https://hbr.org/2018/03/ai-will-change-radiology-but-it-wont-replace-radiologists June 16, 2019.