Artificial intelligence: The evolution of radiology

Technology & Trends Article 3 Minute Read GE Healthcare Global

Artificial intelligence (AI) has emerged from the evolution of technology and machines, but this evolution hasn’t reached its completion yet, especially in radiology. Since the introduction of magnetic resonance imaging, technology has played an important role. The data obtained utilizes the processing powers provided by the computers and scanners they run on. Since that time, computers have become more important to both the world as a whole and radiology departments in particular.

Deep learning

Deep learning algorithms have been developed by researchers in recent years. These algorithms are taught by past imaging to do a variety of things. The radiology department specifies what function it wants the computer to learn. As the images are loaded, the computer may begin to recognize anomalies in the images. For example, a group of researchers developed an algorithm to predict abnormalities in knee MRI exams.1 This team then compared the outcomes of radiologists assisted by the algorithm and those who were unassisted by the algorithm. Their research found that the algorithm and radiologists combined readings may ave reduced the number of false-positives for ACL tears.

Scans affected by deep learning and AI

One of the many areas that deep learning has begun impacting MRI scans is in slice placement. In the past, radiologists have placed the slice, or area being imaged, within the scanner and selected the region of interest on the computer. They had to be sure to include what the physician wants imaged and to position the patient as well as they can. This can often take multiple attempts before the scan can be run, leading to long scan times. AI-assisted automated slice placement may lead to improved consistency and accuracy, as well as saving time. This is developed using deep learning algorithms to find specific anatomical landmarks to scan around.

Simultaneously, radiographers may find that certain algorithms may help to monitor the system’s performance. This would be due to a proactive and predictive analysis. Radiographers could use this data to adjust their scans, which could ultimately save time in part due to the reduction in alterations and re-scanning afterward.

MRI scans heat up tissue inside the magnetic field as a byproduct of the sequences the scanner uses. The rate at which the tissue is heated up is known as the specific absorption rate (SAR). Radiographers have to predict and monitor this rate to ensure that the tissue does not get too hot. With new AI software, they may receive assistance in calculating accurate specific absorption rates at a quicker pace. This software uses a machine learning algorithm to provide fast and accurate SAR estimates.

Finally, AI could provide the ability to manage imaging protocols from any device at every facility, everywhere. Since the development of technology that allows phones, computers and watches to simultaneously receive the same messages or calls, it seemed like only a matter of time until that same level of connectivity to be available in the medical field. Now, some software programs utilizing AI may increase clinical efficiency and workflow. This would allow medical professionals to make the changes to their protocols that they need without having to go to a specific computer.

Buyers Guide_950x250.jpg

Body scanning affected by deep learning and AI

The scanning process is not necessarily the only place that AI can help radiologists. In some cases, artificial intelligence programs may help with scanning of certain areas of the body, such as the brain, liver or heart. These areas can be challenging to image for a variety of reasons. For example, the liver and heart are both susceptible to motion, either cardiac or respiratory.

Deep learning algorithms may have led to AI being able to automatically label and visualize tissue structures in the brain. This could allow for not only automatic segmentation but also for the calculation of volume-based statistics. These statistics may even include the segmentation and calculations related to possible white matter hyperintensities in the brain. In the liver, automatic segmentation and volumetric measurements may lead to a more longitudinal tracking method of liver lesions. Finally, deep learning-based contour detection may lead to a more comprehensive cardiac analysis in MR, possibly at an accelerated rate.

As radiology departments push toward taking advantage of these advancements in AI to image the body with automatic segmentation, active monitoring, and protocol management, they should keep in mind the main goal: improving clinical quality.2 Thanks to recent advancements, this focus may be closer than ever to realization.

For more information, please read SIGNA PulseArtificial intelligence: shaping the future of healthcare.”

References

1. Nicholas Bien, et al. “MRNet: Deep-learning-assisted diagnosis for knee magnetic resonance imaging.” StanfordMLGroup.gihub.io. Web. 7 May 2019. <https://stanfordmlgroup.github.io/projects/mrnet/>.

2. Chris Austin and Mary Beth Mussat. “Artificial intelligence: shaping the future of healthcare.” SIGNA Pulse. Autumn 2018 Web. 10 May 2019. <http://www.gesignapulse.com/signapulse/autumn_2018/MobilePagedReplica.action?pm=1&folio=66#pg66>.