How AI Can Reduce Repeat Imaging

Technology & Trends Article 4 Minute Read GE Healthcare Global

Repeat imaging exams run contrary to the dictum of the right scan (and dose)1 for the right patient and at the right time.2 Complications resulting from repeat imaging include delayed treatment to patients, increased overhead costs for providers, and higher radiation doses for patients. But now, artificial intelligence solutions are emerging as a means to reduce repeat images.

Discrepancies in images in radiology studies are unfortunately quite common, occurring in three to five percent of day-to-day studies.3 Some of these discrepancies, which are often caused by human error, cannot be reconciled, thus leading to repeat imaging, but AI could help address this challenge.

Human Error, Imaging Technologists

Patient position-error by imaging technologists accounts for the most significant share of medical imaging reexaminations.4 Deep learning–based imaging techniques are being used to populate associated metadata to position patients correctly.5 At the 2017 RSNA conference, the largest annual imaging meeting in the world, workflow AI algorithm features to improve patient positioning were prominent.6 Repeats can also be decreased when imaging technologists, or radiographers, receive timely feedback on their performance, including positioning.7 

Human Error, Patients

Patients contribute to repeat exams by moving, especially in MRI studies. This problem costs the average provider $115,000 per MRI machine per year due to poor scans.8 AI is providing new protocols to hasten imaging acquisition to reduce exam time at the root of repeats, such as the AUTOMAP platform.9 Researchers at Massachusetts General Hospital configured AUTOMAP to generate high-quality images in less time, allowing physicians to make real-time decisions while a patient was still in the MRI scanner. The quicker scan times also accomplished high-quality images with lower radiation dose on X-ray, computed tomography (CT) and positron emission tomography (PET) units.

Human Error, Radiologists

Radiologists are human and prone to bias; this truth can lead to medical errors that cause morbidity and mortality.10 Because AI uses wide-ranging tools to enable people to rethink how to integrate information, analyze data, and improve decisions,11 it has already proven to offer equal or better reading accuracy in redundant exams such as mammography.12 AI can now be deployed to research tens of millions of mammograms for lesions taken annually in the U.S. alone, freeing radiologists for other duties such as interacting with referring physicians and patients.

Inefficient Information Platforms

When a physician cannot find a prior image of a patient, they order a repeat image. This common problem plays well into AI’s ability to sort, organize, store, and retrieve vast amounts of information. A 2016 study of health information exchange (HIC) use found that imaging is one of the biggest line-items in Medicare’s budget, which is $10 billion a year.13 X-ray, mammography, and ultrasound accounted for 70 percent of the patient sample. The use of an HIC in a hospital system reduced 47 repeat imaging procedures in 90 days for a total savings of $32,000. Likewise, a Canadian study found that a diagnostic imaging repository (DIR) decreased repeat exams for biliary cancer patients between 15-19 percent in its network of 38 hospitals 100 clinics.14  And finally, an AI-driven program called patient-centered analytical learning machine (PALM) is reaching deep into the hospital care system to improve outcomes and reduce the demand for all services, including imaging.15  PALM AI metadata can monitor and alert physician and staff to a wide range of patient metrics. These range from flagging patients ahead of repository failure (i.e., images required) to predicting no-shows, which is another imaging challenge.  

With a projected shortage of radiologists on the horizon,16 AI offers viable and emerging solutions for the sector to better manage and meet patient care volumes.

References:

  1. Image Wisely: A Campaign to Increase Awareness about Adult Radiation Protection. Radiology. https://pubs.rsna.org/doi/full/10.1148/radiol.10101335. Last accessed August 14, 2018.  
  2. The right scan, for the right patient, at the right time: The reorganization of major trauma service provision in England and its implications for radiologists. Clinical Radiologisthttps://www.clinicalradiologyonline.net/article/S0009-9260(13)00035-4/fulltext.Last accessed August 14, 2018.  
  3. Error and discrepancy in radiology: inevitable or avoidable? Insights into Imaging. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5265198/. Last accessed August 14, 2018.
  4. Uncovering the causes of unnecessary repeated medical imaging examinations, or part of, in two hospital departments. The Radiographer.   http://www.minnisjournals.com.au/articles/Radiographer%20dec%2005%20nol.pdfLast accessed August 14, 2018.
  5. AI-Enhanced Medical Imaging to improve Radiology Workflows. Intel. https://ai.intel.com/ai-enhanced-medical-imaging-to-improve-radiology-workflows/. Last accessed August 14, 2018.
  6. Is artificial intelligence the future of radiology? 3 key takeaways from RSNA 2017. Advisory Board. https://www.advisory.com/research/imaging-performance-partnership/the-reading-room/2017/12/future-of-radiology.Last accessed August 14, 2018.
  7. Monitoring the Use of Extra Images on Chest Radiography Examinations. ScienceDirecthttps://www.sciencedirect.com/science/article/pii/S0363018818301452. Last accessed August 14, 2018.
  8. Artificial Intelligence: Radiology’s Next Frontier? Axis Imaginghttp://www.24x7mag.com/2018/05/artificial-intelligence-imagings-next-frontier/. Last accessed August 14, 2018.
  9. Toward Quantifying the Prevalence, Severity, and Cost Associated With Patient Motion During Clinical MR Examinations. Journal of the American College of Radiologyhttps://www.ncbi.nlm.nih.gov/pubmed/25963225. Last accessed August 14, 2018.
  10. Image reconstruction by domain-transform manifold learning. Naturehttps://www.nature.com/articles/nature25988. Last accessed August 14, 2018.
  11. Bias in Radiology: The How and Why of Misses and Misinterpretations. RadiographicsLP Busby and others. Bias in Radiology: The How and Why of Misses and Misinterpretations. Radiographics. Last accessed August 14, 2018.
  12. How artificial intelligence is transforming the world. Brookings. Darrell West. How artificial intelligence is transforming the world. Brookings. Last accessed August 14, 2018.
  13. Detecting and classifying lesions in mammograms with Deep Learning. https://www.nature.com/articles/s41598-018-22437-z. Last accessed August 14, 2018
  14. Use of Health Information Exchange and Repeat Imaging Costs. Journal of the American College of Radiology.  https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4722040/. Last accessed August 14, 2018
  15. Reducing repeat imaging in hepatico-biliary care through shared diagnostic imaging repository. Published online at VITAL. http://www.vitalimages.com/the-odds-of-avoiding-repeat-imaging-just-improved/.Last accessed August 14, 2018
  16. Embedded Analyst: AI Without Borders. FeibusTech.com. https://www.feibustech.com/blog/2018/7/29/downsizing-the-big-data-problem. Last accessed August 14, 2018
  17. Are You Ready for a Radiologists Shortage? AuntMinnie.com. https://www.auntminnie.com/index.aspx?sec=sup&sub=aic&pag=dis&ItemID=120391.Last accessed August 14, 2018