Artificial intelligence (AI) has started to influence the world on a daily basis. Whether this is through apps on our phones and devices or through search engine prediction, some would argue that it has begun to impact our lives in a positive way. In the radiology department, healthcare workers are interacting with AI more and more commonly. One of the recent applications added to certain magnetic resonance imaging (MRI) scanners, deep learning, is improving the technologist's workflow.
Artificial intelligence in the MR suite
Radiology departments across the country are gradually integrating more and more new technology. As the technology is upgraded, AI may be slowly introduced to the planning and scanning processes. This integration has already begun to affect technologists in a positive way when it comes to scanning.
Without the use of artificial intelligence, technologists have traditionally positioned the patient table within the bore after positioning the patient and coil. Once they had the platform where they wanted it, they had to check how the placement was on the computer. When they began scanning, the technologist had to manually select the scans that they wanted done.
Certain AI platforms may be able to aid with this process in the future. New software has been developed to automatically detect and select the slices. This is done through the use of a workflow assistant and has already improved the process for some radiology departments. Because the program automatically selected the slices, it could help to deliver consistent scans, which vary less from technologist to technologist.1 This could be especially helpful in longitudinal assessments.
The artificial intelligence process for MRI
The process behind this advancement is relatively simple. It does, however, require quite a bit of data. The algorithm is trained using images from previous scans, meaning both normal and abnormal scans must be used. The images are marked with landmarks for the computer to learn using deep-learning techniques. Deep-learning means that the computer "learns" using an algorithm, and it is a subset of machine learning.
Once the algorithm has learned what to look for, it may be used for things like detecting landmarks and automatically detecting the desired slices. One team, led by Tom Schrack as the Manager of MR Education and Technical Development at Fairfax Radiological Consultants in Fairfax, Virginia, that has used the program states that the slice placement is accurate whenever they select a landmark for the prescription.1
The impact of AI for magnetic resonance
The platform for this technology also comes with a number of intelligent applications and smart devices. As a result, radiology departments may achieve greater efficiency, access to care and patient outcomes. The automated slice-selection tool helps to reduce the amount of time necessary for the user to select their desired slices.
As a result, the radiology department may see a reduction in the amount of time that a scan takes. Schrack has experienced this workflow improvement firsthand.1 He states that the program has helped him to scan faster by saving time tweaking parameters. Schrack also believes that the tool may help to make decisions easier for the technologists, making the process a little easier.
Tom Schrack and his team have seen the potential of this automatic slice-selection tool. As a result, they are using it to not only suggest the slices for imaging, but also to improve their workflow. The deep-learning algorithm used to create this tool utilized a large amount of images to determine landmarks, which has helped the department to reduce the time spent positioning the scan. If more radiology departments saw the same improvements, it is possible that artificial intelligence would play as big a role in the radiology department as it does in everyday life.
New deep learning tool streamlines MR slice prescription. SIGNA Pulse of MR. http://www.gesignapulse.com/signapulse/spring_2019/MobilePagedArticle.action?articleId=1488815&app=false#articleId1488815. Last accessed 15 August 2019.