November 28, 2022

Artificial Intelligence and Incidental Findings: Work Smarter, Not Harder

RecommendThanks for recommending this page!

The page you recommended will be added to the "what others are reading" feed on "My ACR".

 BookmarkThanks for bookmarking this page!

The page you bookmarked will be added to the "my reading list" feed on "My ACR".

Eric Cyphers, BS, MS4, Philadelphia College of Osteopathic Medicine and Columbia University Department of Bioethics

headshotIncidental findings are masses or lesions detected by computed tomography, magnetic resonance imaging or other modalities that are discovered during imaging performed for unrelated reasons.1 While many incidental findings are benign, some have the potential to develop into malignancy, which necessitates timely follow-up that can improve a patient’s health outcomes.2 Incidental findings are documented on radiology reports but may go unnoticed or may be improperly followed up on due to workflow inefficiencies or lack of patient response to recommendations; this potentially hinders meaningful intervention, leaving the patient’s disease progression unchecked.

Radiologists, ordering physicians and primary care providers have been increasingly aware of the public health impact of proper follow-up, along with the potential professional and legal implications of delayed or missed follow-up. Advancements in artificial intelligence (AI) natural language processing (NLP) systems offer a solution to improve patient outcomes by preventing delays in reporting and follow-up for incidental findings that warrant a further evaluation. NLP is a facet of artificial intelligence that creates the ability for a computer to understand spoken and written language, much like the dictations found in a radiology report.3 Recently at Northwestern University, an NLP system was developed to identify radiology reports containing lung and adrenal findings that warrant follow-up.4 The NLP system was integrated into Northwestern’s electronic health record (EHR) system where findings are reported to the ordering physician or the patient’s primary care provider within the EHR.2 Physicians can then order follow-up studies as appropriate and track those studies to completion.

Of more than 570,000 images screened over 13 months at Northwestern, the NLP system flagged more than 29,000 reports containing lung follow-up recommendations alone.4 This clinical validation system demonstrated a sensitivity of 77.1%, specificity of 99.5%, and a positive predictive value of 90.3% for lung findings requiring follow-up.4 The integration of this system within the EHR has generated nearly 5,000 interactions with ordering physicians and has tracked over 2,400 follow-ups to completion.4 Although this implementation was the first of its kind and searched for only two categories of incidental findings4, NLP systems can be coded to search for a variety of incidental finding categories and descriptors based on established best practice advisories.

The American College of Radiology Data Science Institute® has published ahead-of-the-curve material to help radiologists understand the value of NLP systems5, and to streamline and standardize various applications of AI in radiology6. Overall, the addition of AI NLP systems to screen radiology reports for incidental findings and track their follow-up to completion represents an exciting advancement for patient-centered and value-based care in radiology.

References

  1. Incidental Findings. American College of Radiology. (n.d.). Retrieved from https://www.acr.org/Clinical-Resources/Incidental-Findings#:~:text=An%20incidental%20finding%2C%20also%20known,performed%20for%20an%20unrelated%20reason.%E2%80%9D
  2. Greeson, T. (2022, March 28). AI application employed to assure "incidental" radiology findings are not overlooked (via Passle). Passle. Retrieved Sept. 22, 2022, from https://viewpoints.reedsmith.com/post/102hliv/ai-application-employed-to-assure-incidental-radiology-findings-are-not-overloo
  3. Lutkevich, B., & Burns, E. (2021, March 2). What is natural language processing? an introduction to NLP. SearchEnterpriseAI. Retrieved Sept. 23, 2022, from https://www.techtarget.com/searchenterpriseai/definition/natural-language-processing-NLP
  4. Domingo, J., Galal, G., Huang, J., Soni, P., Mukhin, V., Altman, C., Bayer, T., Byrd, T., Caron, S., Creamer, P., Gilstrap, J., Gwardys, H., Hogue, C., Kadiyam, K., Massa, M., Salamone, P., Slavicek, R., Suna, M., Ware, B., Etemadi, M. (2022). “Preventing delayed and missed care by applying artificial intelligence to trigger radiology imaging follow-up.” NEJM Catalyst, 3(4). https://doi.org/10.1056/cat.21.0469
  5. AI in brief the AI lifecycle. Data Science Institute DSI. (n.d.). Retrieved Nov. 4, 2022, from https://www.acrdsi.org/DSIBlog/2022/10/18/AI-in-Brief-The-AI-Lifecycle
  6. Standardization and structure report requirements. Data Science Institute DSI. (n.d.). Retrieved Nov. 4, 2022, from https://www.acrdsi.org/DSIBlog/2019/10/14/21/12/Standardization-and-Structure-Report-Requirements