Triaged Patient Friendly Radiology Report Follow-up Recommendations


Encourage development and implementation of triaged patient friendly radiology report follow-up recommendations. 




Non-Interpretive, Patient Facing

Define-AI ID



Patient Facing Panel

Panel Chair

Andrea Borondy Kitts

Panel Reviewers

Alexander J. Towbin, Melissa Davis


Creative Commons 4.0

Status Public Comment 

Clinical Implementation

Value Proposition

Patient-friendly recommendations for follow-up in a radiology report would enable patients to become more involved in their care, enhance the patient understanding of necessary follow-up, and potentially improve the frequency of follow-up recommendations being completed. In the current landscape, patients have access to patient portals on their personal devices and many have the ability to view their radiology reports prior to meeting with their referring clinician1,2. However, as most radiology reports with the notable exception of mammograms, are generated only for a physician or provider reader. Thus, patients may not be able to fully understand the radiology report nor actionable follow-up items. Adding automated follow-up recommendations that are patient-focused would empower patients to better understand and become involved in their care1,2. By understanding patients are a part of their healthcare team and providing action items in the form of follow-up recommendations will enable patients to prompt discussions with their referring clinician, share in their healthcare decision making, and be engaged in their management1,2,3. This will increase the frequency of patients and their healthcare team completing follow-up recommendations in a timely manner. 

  1. Lee CI, Langlotz CP, Elmore JG. Implications of Direct Patient Online Access to Radiology Reports Through Patient Web Portals. J Am Coll Radiol. 2016 Dec;13(12 Pt B):1608-1614. doi: 10.1016/j.jacr.2016.09.007. PMID: 27888949.

  2. Delbanco T, Walker J, Darer JD, Elmore JG, Feldman HJ, Leveille SG, Ralston JD, Ross SE, Vodicka E, Weber VD. Open notes: doctors and patients signing on. Ann Intern Med. 2010 Jul 20;153(2):121-5. doi: 10.7326/0003-4819-153-2-201007200-00008. PMID: 20643992.

  3. Johnson AJ, Easterling D, Williams LS, Glover S, Frankel RM. Insight from patients for radiologists: improving our reporting systems. J Am Coll Radiol. 2009 Nov;6(11):786-94. doi: 10.1016/j.jacr.2009.07.010. PMID: 19878886.


A 67-year-old male patient with left lower quadrant abdominal pain presents to the ER. After a brief initial physical examination is performed and labs are drawn, the patient is sent for a contrast-enhanced CT of the abdomen and pelvis. In addition to demonstrating uncomplicated colonic diverticulitis, a 4 mm solid nodule is seen within the right lung base. Additional incidental findings include several benign renal cysts, a subcentimeter hepatic hemangioma, a chronic and mild L1 compression fracture, and small bilateral hydroceles. The dictating physician describes these findings and in the impression includes a mention of the pulmonary nodule and recommends timely follow-up with an outpatient non-contrast CT chest for further evaluation.

Workflow Description

1. A natural language processing algorithm filters through each radiology report (including both findings and impressions sections), identifying phrases that explicitly recommend or clinically warrant imaging follow-up. These are stratified into “emergent”, “urgent”, and “routine” follow-ups.

2. A numbered list of these clinically warranted follow-ups is generated in lay-language in the patient’s preferred language, and is reviewed by the attending radiologist prior to release.

3. That list is sent both to the ordering provider as well as the patient/healthcare proxy, preceded by the short lay-language summary of the impression, generated by the same NLP algorithm. Include a statement for patients to encourage them to discuss all findings with their physician. 

4. Confirmation of receipt to both parties is provided to the radiologist and is timestamped in the medical record.

5. Regular reminders are sent to the dictating/responsible radiologist if these follow-up recommendations are not received/read. 

Considerations for Dataset Development

  • Procedures: Text radiology reports from all diagnostic modalities (X-ray, CT, MRI, US, Mammography, Nuclear Medicine)
  • No limitations in regard to patient demographics, study type, or ordering location should be imposed. 
  • Develop a data dictionary of the most common imaging findings and follow-up recommendations and the appropriate layperson language for use in training the AI algorithm 

Technical Specifications


Text Radiology Report


X-Ray, CT, MRI, US, Nuclear Medicine


Data Type Text
Modality X-ray, CT, MRI, US, Nuclear Medicine
Body Region Any
Anatomic Focus N/A

Primary Outputs

List of recommendations

RadElement ID N/A
Definition List of recommendations in lay-language, sent to both ordering provider and patient/healthcare proxy.
Data Type Text
Value Set N/A
Units N/A

Urgency of Recommendations

RadElement ID N/A
Definition Urgency of each recommendations in lay-language 
Data Type Categorical
Value Set Emergent, Urgent, and Routine 
Units N/A


Secondary Outputs

EMR Timestamp

RadElement ID



Timestamp in EMR when recommendation is read by patient and ordering clinician

Data Type


Value Set




Future Development Ideas

  • Automated scheduling of follow-up imaging appointments
  • Layperson summaries of full radiology report with illustrations and links to additional information