AI won’t replace radiologists, but it will change their work. Here’s how

Ohad Arazi

Source Link

From “Terminator” to “Black Mirror,” we’re inundated with the idea that machines are slowly taking over, set to eventually replace humankind entirely. This sort of apocalyptic chatter within the medical field, especially in radiology, reinforces the notion that radiologists are in danger of becoming obsolete, along with every other profession. This could not be further from the truth. The number of radiologists has been growing in the double digits for decades, and radiology is predicted to be among the fastest growing fields of the next decade. Some countries are even facing a radiologist shortage.

Still, AI will reshape how radiologists work, shifting their detection of medical conditions from an active to a proactive approach. Understanding these changes can give a better picture of how work will change for radiologists in the near term.

CORONAVIRUS, HEALTH, COVID19, PANDEMIC

What is the World Economic Forum doing to manage emerging risks from COVID-19?

One of the greater misconceptions about radiologists is that they essentially just analyze images. Such simplified assessments, “dramatically oversimplify what radiologists do,” according to Curtis P. Langlotz, MD, PhD, department of radiology at Stanford University. As Langlotz points out, a comprehensive catalog of radiology diagnoses lists nearly 20,000 terms for disorders and imaging observations and over 50,000 causal relations. Algorithms that can help diagnose common conditions are a “major step forward,” he notes, but an experienced radiologist is looking for numerous conditions all at once. Only some of these assessments can be performed with AI.

The tasks a radiologist performs on a regular basis far outweigh the capacity of our current technological functions. Such work includes patient-facing work (such as ultrasound, fluoroscopy and biopsy) to consulting with other physicians, multi-disciplinary work (such as tumour boards), training, and audits.

Current radiology AI systems perform single tasks, undertaking specific image recognition, such as nodule detection on a chest CT scan. These narrow, numerous, and necessary detection tasks are required to fully diagnose the image findings. AI can play a substantial part in improving the diagnostic workflow, even replacing a human in some of the more mundane tasks, such as scheduling. But unless we miraculously invent a complete end-to-end system that includes qualified oversight over the entire diagnostic pathway, AI will not replace radiologists entirely anytime soon.

AI’s impact on radiology can be best compared to the introduction of autopilot to commercial flights. As modern flight systems developed, many of the traditionally human-led safety checks, such as collision-avoidance systems, became automated. Pilots were able utilize the autopilot, allowing it to handle tedious or repetitive tasks, but what happens when he/she encounters an unforeseen malfunction? Or when there’s a storm on the horizon? Well, the pilot is present and ready to take over manually.

This is the same dynamic that will mature in radiology: the human element will be given more freedom to communicate safety issues or react to useful new input. What should be particularly important to remember, however, is that there has been a zero decrease in the number of commercial pilots. In fact, the contrary is true: Airlines reported a shortage of trained pilots in recent years.

Changes ahead

In the immediate term, AI will help human radiologists muscle through the bottleneck forming caused by the pandemic. One of the greatest drivers of coronavirus-related mortality is the inability to quickly detect potentially severe cases and provide critical care to those who need it the most, which contributes to the massive capacity problem hospitals are facing globally, especially in countries that have been hit hardest by the pandemic, such as Italy, Brazil, and the U.S.

While AI can’t yet fight any virus head on, it will play a central role in creating the efficiency needed to tackle this challenge, especially post-COVID-19. As coronavirus cases accumulate, non-urgent cases are being delayed, including extremely important procedures like mammography screening, scanning for bone-health conditions, and cardiovascular imaging. This will create considerable pressure throughout the pandemic as demand skyrockets, and healthcare providers will struggle to process large imaging volumes in a very short time, for both urgent and non-urgent cases. The only way to move forward and prevent a health catastrophe will be through effective prioritization of cases and streamlined processes. After the influx of COVID-19 patients slows down, AI will serve as the human radiologist’s best friend in the effort to effectively prioritize cases.

Efficiency is of particular importance in the field of radiology, given the shortage of devoted professionals globally. A radiologist shortage might sound like a dull headline to people outside of the medical field, but it has concrete implications for everyone: slower processes means people have to wait longer for scans and subsequent treatment, and many of those patients require early detection to beat their conditions. AI can alert radiologists to acute conditions in patients in an efficient and timely manner, accelerating the time it takes to address cases

A transformation for the field

One of the most exciting developments we can expect is the shift from active to proactive detection of medical conditions. Rather than just looking into the specific medical condition for which the patient arrived and requested medical care, AI will empower radiologists to discover additional conditions that were previously undiagnosed or even unknown to the patient. This could lead to the discovery of vertebral compression fractures or cardiovascular events. Through the early detection of such conditions, enabled by AI, radiologists will be able to employ the right treatment for patients at risk for osteoporosis, for example, a game changer for patients and radiologists alike.

We can also look forward to the utilization of AI to extract discrete data from medical images that’s reproducible, quantifiable and extensible. Achieving exact and accurate measurements means that specific pixel data can be leveraged to unlock innovative application tools and big data analytics. Combining this with other pertinent data sources, such as genetic sequencing, will give birth to a diagnostic power play: leveraging insights to go beyond just prevention to drive customized therapeutics, and personalized care.

It would be easy to underestimate developmental possibilities as advances in computer technology continue. Nonetheless, Radiologists will be among the first to be affected by such changes. As AI becomes more immersed in the more laborious, time-consuming functions of radiology tasks, it will begin generating new, unanticipated results that can then be used as sources of medical discovery. Not only will this evolution move forward radiology, but a host of other medical fields and research domains will flourish.

Radiologists won’t be replaced. However, by embracing AI and adapting to these changing times, they will see their jobs transformed and their patients’ quality of care improve. Aided by AI, the field will continue to thrive.