Computer Aided Translation of Radiology Reports to Lay Person - For A CT Without IV Contrast


Encourage development and implementation of patient friendly radiology reports for CT head without contrast




Patient Facing Panel

Define-AI ID



Patient Facing Panel

Lead David Andrews

Panel Chair

Andrea Borondy Kitts

Panel Reviewers

Patient Facing Panel


Creative Commons 4.0

Status Public Comment 

Clinical Implementation

Value Proposition
Radiologist reports are written in technical medical language intended for referring clinicians. However, with the advent of patient portals and empowered patients, patients are likely to be reading the radiologist report “unchaperoned” by their referring clinician or the reading radiologist. The language and terms used for example “can’t rule out malignancy” and the inclusion of comments about normal and expected findings (muscles/tendons:grossly intact; colon full of stool) can be confusing and cause anxiety for patients.  Providing radiologist reports in easy to understand language, with key findings and what they mean, next steps and actions for the patient to take will help patients engage in their care. Patients will thus provide another layer of safety for follow-up care. These easy to understand summaries will also help referring clinicians as they sometimes  find radiologist reports confusing and hard to understand. 

A 55 year old male comes in for a CT head without iv contrast after referral by a primary care physician who noted ataxia after head trauma caused by a fall in the shower.  The patient has no other history consistent with his symptoms.  He is unable to work at his construction job and is eager to get the radiologist’s report as soon as possible so the cause of his problem and possible solutions can be identified. He also indicates he would like to receive the results via the patient portal or email from the referring physician.  After the CT is completed the radiologist interprets the results and prepares a patient friendly summary for the referring physician created from his report by AI.

Workflow Description
The AI algorithm identifies the type of imaging modality, the pertinent body part, and the test objective. It then creates the patient-friendly option. The patient-friendly template is automatically populated with an illustration area of the brain and as the radiologist notes the sizes and locations of different findings, they are automatically added to the illustration of the brain. Future scaling of the approach for other body parts should be considered when developing the algorithm.  Additionally, the text is translated into patient-friendly language and automatically populates the patient-friendly template. The patient-friendly summary includes a description of the findings such as the size and location of pertinent issues, an estimate of the probability of each of them being related to the patient’s symptoms, an illustration showing their relative size and location, and a description of the options for next steps for the patient. Examples of the next steps could include talking to your primary care physician about the need for a neurological consult and possible treatment recommendations.  The full radiology report, images, and the patient-friendly radiology report could be provided to the patient via the patient portal, or via HIPAA secure patient preference. 

Considerations for Dataset Development

Are there different types of reports that should be considered when training the algorithm? 

Radiologist reports written for head CT

What procedures might prompt this algorithm?

Head CT without contrast 

What should be included in the training set?

  • Identify most common terminology terms used in head CT reports and provide data dictionary of layperson language

  • identify examples of images

Technical Specifications


Radiologist Report


head CT



Data Type

Unstructured Text



Body Region


Anatomic Focus


Primary Outputs

Structured patient friendly report


Patient friendly summary of head will include summaries of normal and abnormal findings, an estimate of probability of cause(s) of the patient’s symptoms an illustration showing the size and location of the abnormal area(s) and recommended next steps, including patient actions.

Future Development Ideas

To further expand, utilize AI to automate direct image findings onto the patient-friendly narrative and illustration.