By Elaine Sanchez Wilson
Type “artificial intelligence” and “radiology” into a Google search, and one can observe how the sophisticated technology has been increasingly inserted into conversations about the future of the profession. During last year’s annual meeting of the Radiological Society of North America, for example, attendees showed they were eager to discuss their impact; more than 100 sessions explored the topics of AI and machine learning. Despite prominent computer scientist George Hinton’s bold declaration that AI would replace radiologists in five to 10 years, the radiology community isn’t holding the technology at arm’s length. Rather, many are embracing its potential to enhance their work.
For example, Srikar Adhikari, MD, MS, section chief of emergency ultrasound and director of the Emergency Ultrasound Fellowship at University of Arizona College of Medicine–Tucson, hoped to identify AI tools that could simplify time-consuming cardiac measurements. The department had a checklist: ease of use, accuracy, reproducibility, reliability, and impact on workflow.
“With the AI-powered Venue ultrasound system from GE Healthcare, users with basic ultrasound skills can use the AI tool and obtain cardiac measurements rapidly,” Adhikari says. “It increases efficiency, user confidence, and positively impacts the emergency department workflow.”
To Adhikari, radiologists are more than just image interpreters; in fact, there’s room at the table for both human and machine. “It will augment what they do by increasing their workflow efficiency and minimizing errors and variability,” he adds. “As powerful as these tools are, they still need human oversight to attain the levels of confidence that is required to make a diagnosis.”
Michelle Edler, senior vice president for imaging and care area solutions at GE Healthcare, agrees, saying that AI offers a host of benefits for a variety of stakeholders. “AI has the potential to transform every part of healthcare—from improved provider efficiency to increased diagnostic accuracy to more personalized treatment,” Edler says. “Because of this, and at this early stage, we see individuals across the healthcare system and at multiple levels benefiting from AI tools—from administrators to radiologists to technologists.”
GE Healthcare sees much potential in the intelligent scanner, including CTs, ultrasounds, and x-rays that are embedded with algorithms that can help provide high-quality care more efficiently, Edler shares. Specifically, the company’s x-ray quality application helps decrease manual work required to measure reject rates, increase capacity for diagnostic exams, and reduce unnecessary radiation exposure to patients.
“Previously, it took up to 230 mouse-clicks to access reject data and up to seven hours to calculate a department’s reject rate,” Edler says. “With the advanced analytics application, which will integrate AI and be embedded into x-ray machines, radiology departments can automatically identify and analyze the root cause of rejected x-ray images. This leads to faster, more targeted training and education, so technologists can quickly improve their skills and, therefore, patient care.”
Dolores Dimitropoulos, manager of medical imaging at Ontario, Canada-based Humber River Hospital, was a beta tester of GE’s x-ray app. Early on, the hospital’s quality team recognized the app’s potential toward continuous improvement in x-ray. Because the pilot experience was so positive and productive, the decision to purchase the app once it was officially released was a no-brainer, Dimitropoulos maintains.
“We had already began to implement educational activities for the technologists targeting specific views with high-percentage and high-volume acquisitions,” Dimitropoulos says. “Trialing with an individual technologist, self-awareness was powerful. The level of detail allowed her to review her own practice and seek support from others to improve her data within three months.”
Dimitropoulos recalls that prior to the app, at least two technologists had to travel monthly to each of the hospital’s seven DR x-ray rooms over three locations and two floors—as well as to six mobile x-ray units in four locations on three floors. That’s because they needed to extract the repeat/reject data, return back to the department, and document manually into a spreadsheet to calculate the overall repeat/reject rate. “The process was incredibly time-consuming, and the data produced was limited,” she says. “The outcomes in the past were more generalizations from which it is difficult to provide targeted education or other initiatives to improve our error rates.”
Since using the app, Humber River Hospital has enjoyed countless benefits, Dimitropoulos maintains. For one thing, users have instant access to data, with quick insights into repeat/reject trends down to the exam view. Plus, the time and labor associated with previous manual documentation has been eliminated. And, finally, the tool tracks the performance of the hospital’s 55-plus technologists, highlighting opportunities for self-improvement.
“The X-ray Quality App has helped us empower and better train technicians, elevate x-ray as a modality, and ultimately improve the patient experience,” Dimitropoulos concludes.
Another company that is eager to tap into AI’s potential is Malvern, Pa.-based Siemens Healthineers. At last year’s RSNA annual meeting, the company showcased its Fully Assisting Scanner Technologies (FAST) Integrated Workflow with the new FAST 3D camera. The accessory uses artificial intelligence and deep learning technology to facilitate accurate isocentric positioning of patients, Siemens officials reveal.
Matthew Dedman, CT marketing director of Siemens Healthineers North America, says it’s a pretty significant portion of patients who aren’t at that exact isocentering. This can occur for a number of reasons: the technologist’s level of experience, the size of the patient—and even something as seemingly innocuous as the technologist’s height.
Regardless of the scan or the environment, accurate isocentering is a prerequisite to achieving high diagnostic image quality, Dedman says. After all, he explains, such accuracy will benefit both patients and CT providers because accurate positioning is a key input into the output quality.
What Siemens’ FAST 3D camera does, Dedman says, is take both 3D and infrared images of the patient. From that data, Siemens has an artificial intelligence algorithm that will identify different body regions. For instance, the algorithm can identify a head versus a chest versus an abdomen—and then based on the protocol that’s been selected, it will automatically position that patient at true isocenter per the exam. “Once they take the 3D and infrared image of the patient, with just a simple press and hold of a move button, the system will automatically move the table up and into the gantry and position the patient at isocenter,” Dedman adds.
Because if patients are not properly positioned, CT image quality will be compromised. “You’re typically going to see noisier images that are harder to interpret for the radiologist, and these images will likely take longer for the radiologist to interpret,” Dedman says. “We know that the workload on radiologists is ever-increasing, and they’re being measured on ever-decreasing turnaround times.” If it takes radiologists longer to read a study due to insufficient image quality, he says, their workflow will bear the consequences.
In other words, Dedman says, the seemingly simple task of positioning a patient can have strong implications if it’s not done correctly. “And those are the variabilities that we can eliminate now with the FAST 3D camera,” Dedman says. “We can deliver consistent reproducible patient positioning, which translates to consistent reproducible image quality.”
Moving forward, he says, Siemens is mulling ways to automatically deliver relevant quantitative information to the radiologist. Specifically, Dedman says, “We’re looking to enhance the work of the radiologist, as well as technologists, and make them more efficient in their daily care.”
Man vs. Machine?
Dedman observes that AI is still largely a broad term in the medical world, and certainly within the radiology world. As a result, healthcare providers who are hoping to improve their workflow processes with new technology should evaluate their goals internally, rather than adopt AI only for its buzz factor. “I think it’s a matter of looking at what clinical challenges in your care and in your practice you’re looking to improve and then seeing what AI solution out there could specifically benefit that,” he says.
Srikar Adhikari, for one, recommends that radiology departments consider well-developed AI tools with minimal errors that can replace time-consuming measurements or increase diagnostic certainty. “It’s helpful to have a tool that’s accessible at the bedside, and that fits into your workflow and improves your efficiency,” he says.
And is machine on its way to replace man? Not so, Dolores Dimitropoulos says. “[Radiology] is evolving as imaging becomes not just diagnostic, but more therapeutic, and demand for imaging invasive procedures is rising. AI will be radiologists’ virtual assistant, supporting them in improving the quality of the diagnosis for high-volume reporting, like chest, abdomen or extremity bone exams, and potentially highlighting abnormalities that may have been missed due to interruption, distraction, or screen fatigue.”
GE’s Michelle Edler agrees. “The implementation of AI in healthcare will lead to a world of ‘man plus machine’ not ‘man versus machine,” she says. “AI tools are designed to enhance and empower radiologists’ work and eliminate mundane or administrative tasks. As such, radiologists’ role will evolve with the technology—enabling them to become a more integral part of multidisciplinary teams and closer to their patients.”
Elaine Sanchez Wilson is the former associate editor of 24×7 Magazine. Questions and comments can be directed to 24×7 Magazine chief editor Keri Forsythe-Stephens at [email protected].