Reviving
GI Fluoroscopy.

An AI-driven educational archive for radiology training

RSNA 2025 Educational Exhibit · Loma Linda University Diagnostic Radiology

GI fluoroscopy exposure during radiology training has declined substantially. I wanted to explore whether AI could help turn existing teaching archives into a more interactive way to learn.

Inside the prototype1:44 demo · No audio

The problem

A disappearing
learning opportunity

GI fluoroscopy exposure during radiology residency has declined substantially, while reduced procedural exposure and limited mentorship can make hands-on learning harder to access.

At the same time, valuable legacy teaching cases can be difficult to use in modern educational workflows.

The project started with a simple question:Could we turn those cases into something residents could actually learn from interactively?

The idea

More than a case archive

Learn from real cases

Search and work through structured GI fluoroscopy cases with clinical context and paired imaging.

Practice the read

Interpret cases and practice structured reporting instead of simply looking at the diagnosis.

Get guided feedback

AI-generated feedback, teaching points, and follow-up questions encourage learners to explain their reasoning rather than simply reveal an answer.

How it works

01Learner question
02Relevant case
03Interpretation
04AI feedback
05Faculty review

The platform uses case retrieval to connect learner questions with relevant teaching cases. Learners review the imaging, submit their interpretation, and receive structured AI-supported feedback designed to reinforce reporting and diagnostic reasoning.

Faculty oversight remains part of the model to evaluate the quality and educational accuracy of AI-generated feedback.

Interactive learning

From looking to reasoning.

Search cases by difficulty, system, and keyword — Original platform screenshots from the exhibit · Select an image to enlarge

Clinical context and paired fluoroscopy, CT, and radiography — Original platform screenshots from the exhibit · Select an image to enlarge

Case exploration

Cases can be searched by factors such as difficulty, system, or keyword.

Structured reporting

Learners practice describing findings using standardized templates.

Socratic feedback

Instead of simply giving an answer, the system can ask follow-up questions that encourage diagnostic reasoning and self-assessment.

Teaching points

Cases conclude with key imaging features, diagnostic pearls, common mimics, and pitfalls.

Diagnostic reasoning prompts and space for the learner’s reasoning — Original feedback examples from the exhibit · Select an image to enlarge

Key teaching points for the case — Original feedback examples from the exhibit · Select an image to enlarge

Why I built it

Can AI help us
learn radiology better?

I built the first version during medical school because I was interested in a bigger question:

Can AI help us learn radiology better—not just automate radiology once we’re practicing?

The project combined two things I was already interested in: medical imaging and building better ways to learn.

I developed the web-based prototype in Python/Streamlit, selected and organized cases, reviewed the literature, and worked with radiology faculty to develop the educational framework.

What’s next

From prototype
to evaluation

Ongoing work

The next phase focuses on faculty validation and prospective evaluation of the learning experience.

Planned evaluation includes learner confidence, faculty assessment of AI-generated feedback, agreement between AI and faculty evaluation, and performance across different case difficulty levels.

Prospective educational validation has not yet been completed.

Project information