Artificial Intelligence (AI) has the potential to transform the healthcare industry of the future, although there are unique challenges in healthcare, including patient privacy and related ethical considerations, highly regulated nature of the healthcare industry, and the human element and trust that need to be addressed for successful AI implementation in health care practice. In the field of radiology, also known as medical imaging, numerous research studies have used AI to perform various diagnostic tasks, with an increasing number being developed as commercially available clinical tools regulated by the US Food and Drug Administration (FDA). These are important and highly relevant applications of AI with unique opportunities and challenges for safe and effective implementation in clinical practice. However, one should not lose sight of the many other potential applications of AI that are not primarily focused on a diagnostic scan interpretation task, and instead focus on optimizing and enhancing various healthcare processes. Such tasks have great potential to improve healthcare delivery in the near to intermediate term, and depending on the specific task, they may be technically less complex and have a lower regulatory burden, with a clear positive impact for patients and return on investment for healthcare organizations. In this article, AI applications for enhancement and semi-automation of diagnostic cross sectional medical imaging scans such as CT (computed tomography; also commonly referred to as CAT scans) and MRI (magnetic resonance imaging) are discussed.
• AI is increasingly used to support the scan acquisition process and image quality in CT and MRI.
• These technological advancements can improve patient care and satisfaction, address practical challenges such as manpower shortages, improve diagnostic quality, and increase productivity, with a tangible return on investment.
• Decision makers need to be aware and can leverage these technological advances to ensure a competitive, state of the art practice.
Artificial Intelligence Applications for Computed Tomography Scanning
CT is the workhorse of cross-sectional imaging technology, used for the evaluation of a wide range of urgent and non-urgent medical conditions. CT scans are acquired very rapidly, with a single body part scanned within seconds. This is one among other reasons why they are widely used in the emergency setting. There has been a steady increase in the utilization of CT scans. However, the constantly increasing demand and ensuing high volumes, coupled with other pressures such as rapid turnaround requirements in hospital emergency settings and at times very ill or unstable patients, poses unique challenges for CT operations. This has been compounded by a shortage of radiology technologists – the healthcare professionals who operate and perform the patient scans that are subsequently interpreted by expert doctors, i.e. radiologists. The technologist shortage is not new, but exacerbated by increased demand for diagnostic imaging services. In addition to having a core operational impact that in the worse case scenario could be to limit the operational time of a scanner, there is also a quality impact. An indirect effect of this shortage is to have CT scanners staffed by technologists that are not specialized or less familiar with CT, or outside locum technologists who may not be familiar with the scanners at a particular institution. This can sometimes result in a less than optimal operation and quality of the scans obtained. These challenges can at least partly be addressed using current technology and AI.
There is a movement among different major CT scan vendors in automating parts of the technologists’ workflow, freeing them up to perform other pertinent tasks such as patient facing tasks with an overall positive impact on operational efficiency and quality. Some CT scanners are now equipped with cameras and other technology that can assist in optimal positioning of a patient. Increasingly, the scan acquisition process is being automated, making it simpler to operate with a few clicks. These processes, in addition to their potential for increasing operational efficiency, make scanning less operator dependent can standardize and improve the quality of scan acquisition using machine intelligence.
Technology enhancements, including AI, can also assist by helping standardize and optimize patient positioning and acquisition parameters. As such, scans are increasingly personalized, optimizing radiation exposure to the minimum required for a diagnostic quality scan. The improved quality and consistency of the examinations will also result in better quality scans, which in turn, can enable a better diagnosis at the time of scan interpretation by radiologists.
Beyond the operational optimization of CT scanning, AI is increasingly used for improving the diagnostic quality of images by incorporation of AI algorithms into the image reconstruction process. After a patient is scanned in CT, the X-Ray attenuation data are used to make the CT images that the expert radiologists will look at, a process referred to as CT image reconstruction. Increasingly, AI algorithms are being used in this process by different scanner manufacturers. Currently, a major focus of such algorithms is to reduce image noise and increase image quality. Through this process, the algorithms can also enable the acquisition of a diagnostic quality scan with less patient radiation exposure. Such AI-assisted image reconstruction technology is already available to various degrees on many current commercially available CT scanners.
In the future, the image reconstruction realm is likely to further expand, with potential for incorporation of some diagnostic interpretive tasks at the CT console or suite. First, it is possible that AI may enable significantly enhanced or novel types of reconstructions that can further improve and assist in diagnostic interpretation, especially with certain types of advanced CT referred to broadly as spectral CT. On the direct AI-assisted interpretation side, one potentially impactful area is the immediate screening of acquired scans for quality and urgent conditions. Pertaining to scan quality, one can envision the machines increasingly analyzing and alerting the technologist at the time of scanning whether the scan is of good diagnostic quality or not. This would enable immediate mitigation and repeat scanning if necessary, avoiding a non-diagnostic scan or need for patient callback and associated patient inconvenience.
AI software, either third party software or that provided by a scanner manufacturer, can also be used to screen for unexpected urgent and or potentially life-threatening conditions, particularly in outpatient settings. This does not necessarily have to be done on the scanner console, but in the most impactful scenario, it would be done and the results available near-simultaneously and as soon as a scan has been obtained. This way, the technologist can be alerted before the patient has left the scanner suite and the Department. Flagging an unexpected urgent finding would enable immediate routing of the patient to their treating physician or the emergency department if necessary. At a minimum, this will reduce patient inconvenience and in certain cases, it could make the difference between timely diagnosis and potential prevention of an ensuing adverse or catastrophic event.
Artificial Intelligence Applications for Magnetic Resonance Imaging
MRI is another commonly used advanced cross sectional imaging technique. MRI provides excellent soft tissue contrast and is used for the diagnosis of a variety of oncologic and non-oncologic conditions. A detailed discussion and comparison of MRI and CT is beyond the scope of this article but on the practical and technical side, it’s important to remind the reader of a few things. Operationally, MRI scans take much more time than CT scans to acquire. An average MRI scan can easily take 15-30 minutes, depending on the specific indications and body part examined. In certain practices, MRI scans may on average be longer than 30 minutes, and occasionally some scans may take longer than 1 hour to complete. As a result, there is great opportunity for improving patient comfort and operational efficiency and revenue through reduction of scan times, with an additional associated quality enhancement opportunity.
On the patient comfort side, it is needless to say that remaining motionless on a scanner table for long periods of time is not the most desirable or comfortable experience for anyone, but even more so for a patient who is ill, in pain, or has trouble breathing. Claustrophobia can also be a major issue for MRI scans. In that sense, accelerating and having significantly lower scan times can directly benefit the patient by making the process more comfortable for the patient. There is furthermore a secondary quality impact to reducing scan times in MRI is that sometimes is overlooked or not sufficiently emphasized. It is not easy for any patient to remain motionless for long periods of time and the longer the examination time, the more the likelihood of having some sequences motion degraded which in turn can compromise diagnostic quality and interpretation. Reducing overall scan times can therefore indirectly benefit image quality and diagnostic interpretation.
The other potential significant impact of reducing MRI scan times is increased productivity and revenue, particularly because MRI is typically a largely fixed cost operation. This can also have the benefit of reduced wait times, especially when capacity is an issue. In the emergency and inpatient setting, this has the potential to reduce length of stay with a positive impact across the enterprise.
Similar to what was discussed for CT, there are an increasing number of AI algorithms used for reconstructing or enhancing MRI images. In the image reconstruction space, AI has shown the potential to both improve image quality and shorten MRI sequence acquisition time. It is important to be aware of these tools and get specifics from a scanner vendor during purchase of a new scanner, since these can have significant productivity and return on investment. Decision makers also need to be familiar with the potential and multi-faceted benefits of such technology, and in certain cases, an upfront investment in a more sophisticated technology could yield significant long term benefits, both from a patient care and an institutional productivity and financial perspective.
In addition to the above, an interesting category of AI tools have emerged that typically are vendor neutral and work on the images after the initial image reconstruction. These can improve quality and through the process, enable shortening acquisition time (while preserving image quality). To get the associated efficiency, sequence acquisition parameters are changed that reduce scan time. Without AI algorithm assistance, these change would reduce signal and increase image noise and could make the quality less than desired or unacceptable, but the AI algorithm makes up for the image quality and signal to noise on the shortened sequence yielding images of similar quality while enabling shortened scan times. Because these are applied after the original image reconstruction, some refer to this process as image enhancement. Nonetheless, they also have demonstrated gains in quality and increasing productivity. One particularly attractive feature of vendor neutral MRI enhancement software is that they can be applied to different scanners, including older scanners, and do not require a hardware upgrade. This widens the potential use and benefits to various scanners within an institution that are not yet planned for an upgrade or replacement. They also require much less upfront capital investment. While the data is preliminary, such tools have been shown to increase productivity and can have a significant return on investment, especially in practices where capacity is limited.
Some of the other operational benefits of technology and AI discussed for CT also apply to MRI. MRI is also impacted by the technologist shortage, and perhaps even more susceptible given the specific and complex expertise required for optimal acquisition of MRI scans. Similar to CT, MRI vendors are increasingly using different technologies to streamline the scanning process. There is furthermore other promising developments in this domain, including one approach where a technologist with a high degree of expertise, sometimes referred to as a superuser, can monitor multiple scanners from one location. In the future, this model is likely to expand and be further enhanced with AI, dealing with the practical challenges of manpower shortage and expertise in MRI operations.
Similar to what was discussed for CT, AI-assisted image reconstruction may eventually lead to new or enhanced sequences beyond those focused on image quality and noise alone, helping improve diagnosis. There is also potential for the scanners of the future to monitor the quality of sequences, automatically adjust parameters, and suggest repeat sequences when key MRI sequences are suboptimal, avoiding patient callbacks and improving diagnosis. In the future, AI may also be used to match the scan to a patient's prior examinations, optimizing follow-up, and provide a preliminary image analysis including flagging of potentially urgent or unexpected abnormalities as was discussed in the earlier section on CT scans.
There have been impressive advancements in CT and MRI scan technology, increasing automating and enhancing the process. These have benefits for patients and also can have significant operational benefits by addressing practical challenges such as manpower shortages, increase productivity, and provide a tangible return on investment for healthcare institutions. It is important for decision makers to be aware of these advancements in order to ensure an optimal and competitive operation. In certain cases, upfront investments in technology may yield substantial long-term return on investment. Decision makers in leading healthcare practices will be well served by familiarizing themselves and considering current and upcoming advancements as part of their technology roadmap, with ultimate benefits to the patients they serve and the healthcare institution.
Selected review of the use of AI for operational streamlining in radiology
Pierre K, Haneberg AG, Kwak S, Peters KR, Hochhegger B, Sananmuang T, Tunlayadechanont P, Tighe PJ, Mancuso A, Forghani R. Applications of Artificial Intelligence in the Radiology Roundtrip: Process Streamlining, Workflow Optimization, and Beyond. Semin Roentgenol. 2023 Apr; 58(2):158-169. doi: 10.1053/j.ro.2023.02.003. Epub 2023 Mar 23.