Radiologists have warmed to artificial intelligence, with the technology slated to improve inefficient workflows.
Radiology has emerged as a leader in artificial intelligence out of a pressing need.
“The primary driver behind the emergence of AI in medical imaging has been the desire for greater efficacy and efficiency in clinical care,” wrote Hosny et al. in the 2018 report “Artificial intelligence in radiology.”
“Radiological imaging data continues to grow at a disproportionate rate when compared with the number of available trained readers, and the decline in imaging reimbursements has forced health-care providers to compensate by increasing productivity,” they continued. “These factors have contributed to a dramatic increase in radiologists’ workloads. Studies report that, in some cases, an average radiologist must interpret one image every 3–4 seconds in an 8-hour workday to meet workload demands.”
Industry leaders in radiology are actively working to identify opportunities for machine learning, neural networks, and natural language processing to optimize radiology workflows.
In 2019, the National Institute of Biomedical Imaging and Bioengineering met with the Radiological Society of North America, American College of Radiology, Academy of Radiology and Biomedical Research Academy, and National Institutes of Health to discuss the future impact of AI on medical imaging. While the workgroup noted that interpretation (e.g., detection, segmentation, classification) garnered the bulk of attention, it emphasized in a Journal of the American College of Radiology article that “progress in AI research and development is occurring in all areas.”
The world of radiology has warmed to the idea of artificial intelligence. The challenge now becomes identifying opportunities for reducing inefficiencies in radiology workflows through AI integration.
Based on a review of current literature, here is a breakdown of areas where AI can improve the practice of radiology.
Detection & prioritization
Detection is the poster child for artificial intelligence in healthcare, but there’s even more the technology can add as a screening tool.
“Defining the boundary between a normal and abnormal image in a formal way is very complex and multifactorial. In this context, deep learning can potentially excel by learning a hierarchical normal representation of a specific type of image from a large number of normal exams,” Canadian Association of Radiologists (CAR) wrote in 2018.
With automated detection, radiologists view images based on reading priority which speeds reporting and improves patient outcomes. With the addition of retrieval services, the AI pulls similar images from a database for review when it encounters unusual or complex cases.
In its white paper, the association conceived of three models for using AI in the context of clinical workflows:
Triage: AI screens examinations for probability of disease and determines the order of interpretation
Replacement: AI is used where its results consistently outperform human radiologists
Add-on: AI supports existing clinical pathways by handling time-intensive activities
The act of isolating an area of interest in an imaging study remains a manual task and subject to variability. In a recent white paper, the European Society of Radiology (ESR) described automated segmentation as “crucial as an AI application for reducing the burden on radiology workflow.”
Deep learning shows the greatest potential to address this inefficiency.
“Given its ability to learn complex data representations, deep learning is also often robust against undesired variation, such as the inter-reader variability, and can hence be applied to a large variety of clinical conditions and parameters,” noted Hosny and colleagues.
“In many ways, deep learning can mirror what trained radiologists do, that is, identify image parameters but also weigh up the importance of these parameters on the basis of other factors to arrive at a clinical decision.”
Monitoring & registration
Monitoring the development of a tumor requires the comparison of numerous images to track progress (i.e., image registration).
“Whereas some change characteristics are directly identifiable by humans, such as moderately large variations in object size, shape and cavitation, others are not,” noted Hosny et al. These could include subtle variations in texture and heterogeneity within the object. Poor image registration, dealing with multiple objects and physiological changes over time all contribute to more challenging change analyses.”
Deep learning provides a higher level of consistency and does so at unmatched speeds. “With robust registration algorithms based on deep learning, the utility of multimodal imaging can be further explored without concerns regarding registration accuracy,” they added.
In radiology, the precision of clinical decision-making depends on the richness of data contained in an image.
AI technology is poised to help address challenges to high-quality image acquisition. The first is the variation in imaging protocols and modalities.
“We find a widening gap between advancements in image acquisition hardware and image-reconstruction software, a gap that can potentially be addressed by new deep learning methods for suppressing artefacts and improving overall quality,” wrote Hosny et al.
The second is reducing harm to patients through appropriate dosing. Published guidelines exist but are frequently ineffective.
“The choice of protocol, however, is subject to variability since it is frequently operator-dependent, and consequently, the radiation dose and the quality of the exam are subject to variability at both intra- and inter-institutional levels,” noted ESR.
“In this setting, AI can be an optimising tool for assisting the technologist and radiologist in choosing a personalised patient’s protocol, in tracking the patient’s dose parameters, and in providing an estimate of the radiation risks associated with cumulative dose and the patient’s susceptibility (age and other clinical parameters).”
As with clinical documentation, reporting is generally a source of frustration for radiologists.
What’s more, this “laborious and routine time-consuming task” is error-prone simply as a result of when it takes place. “As the report generation task falls towards the end of the radiology workflow, it is the most sensitive to errors from preceding steps,” warned Hosny et al.
Currently, a lack of standardization has led to variation in documentation among radiologists. AI likely holds the key to making imaging data more uniform structure.
“Owing to the often different formats in which data are recorded by medical professionals, Al-run, automatic, report-generation tools can pave the way for a more standardized terminology — an area that currently lacks stringent standards and an agreed-upon understanding of what constitutes a ‘good’ report,” stated Hosny colleagues.
Doing so not only improves diagnosis and communication between radiologists and clinicians but also paves the way for future population health and big data research. For its part, ESR has partnered with RSNA on a structured reporting initiative to promote the adoption of common data elements.
The radiology information cycle is multifaceted and complex. When well and properly trained, AI can ensure that radiologists produce highly valuable data to improve the health of individuals and populations. By reducing inefficiencies, radiologists can a broader and deeper impact on patient care.
Source : healthitanalytics
Jennifer Cantelli was born and raised in the busy city of Lancaster. As a journalist, Jennifer has contributed to many online publications including the The Crimson White and USA Today. In regards to academics, Jennifer earned a degree in business from Carnegie Mellon University and an master’s degree from Temple University. Jennifer follows the money and covers all aspects of state and federal economy.here at Times Records.