Cardiac MR and AITechnology & Trends Article 3 Minute Read GE Healthcare Global
Heart disease is the leading cause of death in the US; over 635,000 people died due to heart disease in 2017.1 Cardiology patients may need to have an imaging study done to help the doctor determine what their heart looks like and attempt to predict cardiac events before they happen. In the past few years, research on how artificial intelligence (AI) may help these physicians through improving imaging methods, especially magnetic resonance imaging (MRI). Two applications of AI that may increase speed and accuracy of MRI results are automated segmentation and 4D flow.
MR scanners typically take one hour to image a cardiovascular patient, but this is followed by another extensive length of time to create the images.2 This may be done manually by the radiology with the help of specific software. They have to determine where the organ or tissue ends to create a contour, which can take quite a while when done manually.2,3 This process is called segmentation. Researchers hope to be able to speed up segmentation by employing AI. Programmers may soon make this process semi-automated.3 The images would still be acquired by a highly skilled radiographer with knowledge of cardiac imaging, but software tools may soon be capable of automated planning and acquisition with their help. Additionally, the computer may not be able to automatically map abnormal anatomy or anatomy that varies on a patient-to-patient basis, like the right ventricle.
Machine learning, also called deep learning is one of the commonly researched software programs utilizing AI that could be used in cardiovascular MR.3 This program uses data garnered from thousands of images of a specific body part to map the patient’s body automatically based on the data obtained from their MRI. One study found that one algorithm’s accuracy improved drastically between 2,000 to 20,00 images but had little improvement from 20,000 to 200,000 images, though researchers initially thought the program would need as many pictures as possible.4 Deep learning is a method that could be used in MR when scanning a variety of different areas of the body.
After segmentation, AI has the potential to collect and record measurements that are currently done manually. These measurement includes both the left and right ventricle ejection fractions.5,6 These measurements may sometimes be inconsistent based on the radiologists examining them, and automation may help with this issue. In particular, one study’s researchers fed a deep learning algorithm 93,500 images with annotations.6 The program was then able to identify and calculate the ejection fraction. On the left side, the difference between the automated and manually calculated fractions was only 3.2%. The right side produced a slightly higher difference of 7.5%. Radiologists associated with the study determined that the automated, deep learning program produced reports that were almost the same as those calculated by humans. AI could potentially further decrease the amount of time it takes to finish processing the images and measurements.
MR can visualize blood-flow within the heart while quantifying the ejection fraction.7,8 This is done through the use of 4D flow technology. 4D flow utilizes an algorithm that may help clinicians assess heart diseases. The program creates a cine (in motion) MRI of the heart and the flow of blood within it. AI then color codes the blood flow to show what blood is flowing in and what is flowing out. This may allow physicians to monitor heart beat and detect abnormalities. One software has been developed that uses automated valve tracking to quantify the information.8 Physicians, with the support of this program, may find themselves more accurately diagnosing patients.
When segmentation and flow are completed using AI, the processing time for images may be decreased. Radiologists may find their department is working more efficiently, because they may not need to monitor the program and create their own measurements. At this point, automated segmentation is still in the research phase. However, if it becomes available, it could lead to more accurate and less time consuming MRI results. 4D Flow MRI can be used today through a few different software programs and may improve accuracy as well. Physicians can more easily visualize the flow throughout their patients heart and monitor for abnormalities in blood flow. Hopefully, as AI advances and physicians can more accurately diagnose and predict heart conditions, less than 635,000 people will die from heart disease in the coming years.
1. “Leading Causes of Death.” CDC.gov. 17 March 2017. Web. 24 January 2019. <https://www.cdc.gov/nchs/fastats/leading-causes-of-death.htm>.
2. Piotr J. Slomka, et al. “Cardiac imaging: working towards fully-automated machine analysis & interpretation.” Expert Rev Med Devices. March 2017; 14(3):197-212. Web. 23 January 2019. doi: 10.1080/17434440.2017.1300057.
3. Suvadip Paul and Jessica Wetstone. “Getting to the Heart of it: How Deep Learning is Transforming Cardiac Imaging.” medium.com. 5 February 2018. Web. 24 January 2019.<https://medium.com/stanford-ai-for-healthcare/getting-to-the-heart-of-it-how-deep-learning-is-transforming-cardiac-imaging-22d34bf91a4e>.
4. Richard Dargan. “Can AI Add Value to Radiology? Informatics Experts Share Latest Findings with Overflow Crowd.” RSNA Daily Bulletin. 29 November 2018. Web. 24 January 2019. <https://rsna2018.rsna.org/dailybulletin/index.cfm?pg=18thu02>.
5. Matt O’Connor. “Deep learning method may produce faster cardiac MRI reports.” HealthImaging.com. 9 October 2018. Web. 24 January 2019. <https://www.healthimaging.com/topics/cardiovascular-imaging/deep-learning-produce-faster-cardiac-mri-reports>.
6. Wenjia Bai, et al. “Automated cardiovascular magnetic resonance image analysis with fully convolutional networks.” Journal of Cardiovascular Magnetic Resonance. 14 September 2018; 20:65. Web. 24 January 2019. doi: https://doi.org/10.1186/s12968-018-0471-x.
7. Shourjya Sanyal. “4 Ways In Which AI Is Revolutionizing Cardiac Care.” Forbes Magazine. 27 October 2018. Web. 24 January 2019. <https://www.forbes.com/sites/shourjyasanyal/2018/10/27/4-ways-in-which-ai-is-revolutionizing-cardiac-care/#2aeb503a7a43>.
8. Daniel Allar. “Automated valve tracking reduces time, variability in 4D flow MRI.” CardiovascularBusiness.com. 2 November 2018. Web. 24 January 2019. <https://www.cardiovascularbusiness.com/topics/cardiovascular-imaging/automated-valve-tracking-efficient-4d-flow-mri>.