As Originally Published in Radiology Today
While AI was initially utilized in academic settings for research, the realization that AI solutions can automate and/or standardize specific complex image interpretation tasks in high-value workflows has driven acceptance in clinical settings. For example, within the past few years, AI in CT imaging has gained considerable acceptance.
The number of pathologies identified by AI-based applications is increasing across radiology, especially in CT imaging. Cyril Di Grandi, cofounder and CEO of Avicenna. AI, says the market reached a turning point in 2022 with greater consolidation of data, driven by manufacturers’ increasing willingness to propose complete and coherent offerings.
“The year was also marked by a large number of studies demonstrating the eco- nomic and clinical justification of the use of AI in CT imaging,” he says.
Steve Worrell, CEO of Riverain Technologies, says AI solutions that allow physicians to put more focus on critical aspects to achieve better outcomes in less time—such as lesion detection and characterization, automated measurements, and segmentations on CT images—are in high demand. He believes acceptance will become commonplace once reimbursement for AI use becomes available.
“We are on the cusp of a transition from the early adopter to early majority phase,” Worrell says. “That AI solutions have begun to become mainstream is remarkable. There is a wide variety of solutions available for clinical use, with more innovation coming fast and furious.”
Elad Walach, cofounder and CEO of Aidoc, has seen extensive engagement and discussion around AI and its ability to address real-world challenges for medical facilities. He says radiologists are leading the way in these discussions because they understand the potential efficiencies that AI can bring to radiology practices, enhancing their ability to treat patients.
“The idea that radiologists are in the driver’s seat for hospital-wide AI adoption is something we observed again in our conversations at RSNA 2022 this year,” he says.
Bob Jacobus, CEO of AI Metrics, says decision-makers in radiology have reached their limit on “AI hype” and are increasingly demanding both a justification for purchasing and a clear use case that solves a radiologist’s pain point. For this reason, he believes that the bar is being raised, and many of the 2015–2021 vintage companies won’t survive much longer without a clear value proposition for the radiologist.
“We see the trend towards product utility and away from AI as a panacea or even a product, for that matter,” Jacobus says. “In our opinion, AI is not a product at all, but is instead a technique, and a very useful tool used as a means to an end. … We’re solving pain points such as radiologist shortages, burnout, callbacks, and highly variable reporting.”
Aidoc offers radiology software for its AI operating system, aiOS, which integrates AI algorithms into customizable clinical workflows, assisting physicians across service lines with delivering care and communicating follow-up actions for their patients. In Q4 2022, Aidoc increased its FDA clearances to 12 with the addition of two CT-based AI solutions—one for aortic dissection and another for all large and medium vessel occlusions.
“The data tsunami—along with a staffing drought—are exacerbating a healthcare crisis,” Walach says. “This is where AI can provide support in the healthcare environment today. We derive insights from complex data, starting with diagnostic imaging, to aid health care teams in optimizing patient treatment, which results both in improved clinical outcomes and economic value.”
In AI Metrics’ case, the “end” it is seeking is an interactive software system that enables radiologists to evaluate cancer patients faster, with greater accuracy and better communication to oncologists. “Advance cancer reads are a currently low- or no-margin work activity for radiology,” Jacobus says. “We transform this critical, complex task into a high-margin, high-value part of the radiologists’ day. AI Metrics shortens the read time by half and reduces the read complexity. The result is ‘un–cherry picking,’ and it doesn’t require any worklist orchestration at all.”
Di Grandi believes radiology AI will only continue to thrive if the solutions are funded and deliver clinical impact, coupled with an economic benefit to the hospital and payers. Additionally, he says products must help address the challenges associated with the increasing volume of associated examinations.
“We aim to cover two distinct but related areas—emergency imaging and rapid detection of serious or life-threatening pathologies,” Di Grandi says. “Our applications help reduce errors and speed up patient management. We are also developing a new range of AI applications for incidental pathology discovery. In this case, the AI application scans prior exams that were not initially intended to detect the pathology, allowing medical teams to route the patient to the best care management pathway.”
Riverain’s approach has been to focus on the clinical impact of AI for radiologists, keeping in mind that any solution that aids the radiologist must not add to the complexity of the interpretation process. “Additional interfaces, widgets, clicks—they all increase the time spent reading the study,” Worrell says. “Our solutions offer the ability for radiologists to be certain of their findings search. We call that capability Clear Visual Intelligence. Our patented vessel suppression technology gives the radiologist an unobstructed view of the thorax within their existing workflow. This enables radiologists to see past obstructions to detect cardio-thoracic diseases correctly and quickly with Certainty of Search.”
Lung cancer screening programs are a notable example of where vessel suppression technology can improve outcomes. “The chest [lung and heart] is one of the most challenging anatomical regions to read due to a multitude of diseases and reading complexity, leading to a high exam reading burden for the radiologist,” Worrell says. “Missed lung cancer is the second most frequent cause of malpractice, with risks increasing along with radiologists’ workload. ClearRead CT gives radiologists an unobstructed view within the existing workflow so they can focus on what matters, to detect, precisely characterize, and report findings.”
AI and deep learning can reproduce complex tasks almost perfectly. For certain tasks, Di Grandi says, this technology allows the creation of algorithms capable of comparable performances to those of a human expert. Thus, these tasks can be automated and executed in parallel with existing workflows to increase the capacity of existing radiology teams.
“The management of neurological emergencies is one the most compelling use cases for AI,” he says. “For a patient with a stroke, time is the most important variable in their chance of survival without significant health consequences. Therefore, a completely automated tool that alerts all the clinical teams in charge of a patient’s care in a few seconds will clearly improve multidisciplinary communication and optimize outcomes.”
Aidoc technology analyzes medical imaging to provide comprehensive solutions for flagging acute abnormalities across the body, helping radiologists prioritize life-threatening cases and expedite patient care. “Our current algorithms include aortic dissection, intracranial hemorrhage, vessel occlusion, pulmonary embolism, and cervical-spine fractures, as well as incidental pulmonary embolisms, free-air, and rib fractures,” Walach says. “This technology helps doctors identify potential points of concern so they can offer more timely treatment and care to patients.”
The company’s AI layer in CT imaging acts as a safety net for the physician, Walach adds, constantly analyzing health care data and imaging in the background to help detect critical issues. Once an issue is identified, the system immediately notifies the appropriate physician or care team.
For example, if a patient undergoes a routine chest exam to follow up on an existing condition or disease, approximately 3% of those patients will have an incidental pulmonary embolism, which could be life-threatening. However, incidental findings are frequently handled too late or missed altogether. With Aidoc’s always-on AI, the technology immediately detects a suspected pulmonary embolism and notifies it as a priority for the radiology team to review. Care teams are activated more quickly to determine diagnosis and urgent treatment can be provided.
According to the ACR, “clinical deployment of AI is still in its early stages.” An ACR Data Science Institute AI Survey published in the Journal of the American College of Radiology in 2021 notes that only 30% of radiologists use AI clinically in current practice, which indicates that much needs to be done to facilitate wide-scale acceptance and adoption of AI.
“A significant challenge for AI is trust,” Worrell says. “Clinicians must have high confidence in any tool that aids their workflow. Scalability—the selected AI solution must have the ability to operate at necessary size, speed, and complexity—is also required. An AI solution that cannot handle the volume of imaging it is expected to process will only frustrate the end user.”
Walach says hospital leaders’ decision overload associated with enterprise-wide software investments often presents a challenge for wide-scale AI adoption. Radiologists can help lead the way in this respect as they are in a unique position to steer hospital teams on AI strategy by pro- viding insight into return on investment (ROI), increased collaboration, and enterprise technology that ensures patients receive appropriate treatment.
“Another challenge is that facilities cannot simply manage 100 single-point solutions, and this is a frequent obstacle to wider adoption,” he says. “Additionally, there is still some initial skepticism from medical practitioners who don’t have firsthand experience working with the power of AI. AI should be looked at as a tool that empowers physicians and a technology that will help health systems rise above some of today’s challenges to operate at an even higher standard.”
Commercial deployment of Avicenna. AI’s applications grew significantly in 2022, with more than 140 hospitals now equipped with its software in 14 countries worldwide. In a market of tens of thousands of potential customers, however, Di Grandi says AI deployment is just beginning.
“The market is still predominantly driven by early adopters who are investing time and resources to evaluate the impact of AI in clinical practice,” he says. “Going forward, determining the true ROI of AI solutions and how they are financed will be the key to large-scale deployment.”
AI is playing a significant role in getting health care to the point where it is delivering data-driven, scientifically driven, precision medicine. AI is fundamental to achieving these goals due to the complexity of the data and the impracticality of addressing the problem outside AI.
Di Grandi believes the future of this technology will be an augmented radiologist/clinician who will use AI in an integrated way in their practice to optimize their time and performance. “Some routine exams will not even be reviewed by physicians anymore, and even some ‘abnormal’ exams will initially be prepared by AI to allow the physician to optimize their time and maximize their added value on complex cases,” he says.
According to Walach, the application of AI is the biggest paradigm shift in radiology technology since PACS was introduced half a century ago. “Just as PACS revolutionized the field of radiology, AI is improving workflows and collaboration while reducing waiting times, helping radiologists make better diagnoses and improving patient care,” he says. “While some view health care AI as the future of tech, the future is already here. AI simply enables new levels of optimization in their work, allowing them to perform more efficiently and safely.”
By Keith Loria