PORTLAND, OR – An integrated AI approach can help produce video radiology reports that are patient-friendly, according to a presentation at the Society for Imaging Informatics in Medicine (SIIM) annual meeting.
Researchers led by presenter Luyang Luo, PhD, a postdoctoral fellow at Harvard Medical School in Boston, believe that video reports produced by its ReXplain AI-based software could ultimately enable improved patient outcomes.
The group was motivated in their project to help improve patient engagement and activity, which further leads to better outcomes, Luo noted.
“Although patients nowadays have better access to their medical data through patient portals, the reports are basically communications between professionals,” Luo said. “They’re not designed to be understood by patients.”
Luyang Luo, PhD, presented the group’s results in a session at SIIM 2025.
The researchers hypothesized that AI-generated video reports would be rated by radiologists as accurate and that they could be integrated into the radiology workflow.
ReXplain features three components. The first, a large-language model (LLM), simplifies the text of a radiology report and produces an important finding explanation. Next, a segmentation model localizes the important findings on a patient’s image from the PACS. Finally, an avatar is generated to provide the video description by converting the important findings from text to speech and then explaining them to the patient in a user-friendly interface.
To assess the quality of these video reports, the group tested ReXplain in a utility study involving five radiologists and 10 video reports from CT scans. After being asked to answer questions on the video reports, the radiologists overwhelmingly indicated that they either agreed or strongly agreed with the following statements:
- Connecting the explanation with the image helps improve understanding: 100%
- The comparison with normal CT scan improves understanding: 100%
- The video correctly identifies the important findings in the CT: 100%
- The avatar is natural and conversational: 98%
- The explanation of the report is easy to understand: 98%
- The video can be understood by a patient with an eighth-grade reading level: 98%
- The video sufficiently reviews the findings with the patients: 96%
- I am comfortable providing this video to patients to help them understand: 80%
- I am comfortable providing this video to patients before their visit: 60%
- The rendering helps understand the 3D structure: 50%
“ReXplain-generated video reports demonstrate huge potential to be integrated into the radiology workflow,” Luo said.
In future work, the researchers will gather feedback from patients before and after viewing the video radiology report.