Try a short adaptive learning demo and see how adaptive learning can identify knowledge gaps, uncertainty, and learner progress in medical imaging education
Adaptive learning is an educational approach where the learning experience changes in response to what a learner appears to know, where they are struggling, and how their understanding develops over time.
In short. Adaptive learning in radiography education builds a changing model of each student's knowledge from multiple responses and image-based tasks. It then selects feedback, explanations, prerequisite material, or more advanced applications based on the concepts the learner understands, misunderstands, or applies inconsistently.
That sounds straightforward.
But what does it actually look like for a radiography student?
We have been building a series of short, interactive learning experiences for medical imaging education. Rather than giving every student the same video, slides, and fixed quiz, the aim is to build a better picture of the learner's developing knowledge and adapt the teaching accordingly.
We have created a small fluoroscopic image quality demo to show the direction we are taking.
Try the adaptive learning demo:
Open the demoIt is deliberately a short example.
The interesting part is not the number of questions.
It is what happens to our understanding of the learner as they answer them.
Traditional online assessment often works like this:
Question.
Answer.
Correct or incorrect.
Next question.
At the end, the student receives a score.
The problem is that a score can hide a great deal.
Imagine two radiography students both achieve 80%.
One has a strong understanding of fluoroscopic image quality but makes two careless mistakes.
The other has significant gaps in their understanding but correctly guesses several answers.
The final score may be identical.
The educational problem is not.
This is one of the questions we are trying to explore with adaptive learning:
How do we estimate what a learner actually knows rather than simply counting correct answers?
A single correct answer is evidence.
It is not proof of mastery.
When a learner answers a question correctly, several explanations are possible.
For that reason, our approach does not treat one correct answer as the end of the story.
The system can build evidence across multiple interactions.
The aim is to develop an evolving estimate of knowledge rather than awarding mastery because a student clicked the correct answer once.
This is probably one of the biggest misconceptions about adaptive learning.
Correct answer? Give a harder question.
Incorrect answer? Give an easier question.
That is branching.
It can be useful, but adaptive learning can go considerably further.
| Fixed quiz | Basic branching | Adaptive learning |
|---|---|---|
| Same questions for everyone | Harder or easier question based on one response | Selects content using patterns across concepts and interactions |
| Produces a total score | Changes difficulty | Estimates concept-level knowledge |
| Limited diagnostic value | Identifies right or wrong | Identifies misconceptions, uncertainty, and prerequisite gaps |
Consider a student learning about fluoroscopic image quality.
The student may understand spatial resolution but struggle with temporal resolution.
They may correctly identify image noise but misunderstand why it occurs.
They may remember a definition of magnification but struggle to apply the concept when looking at an image.
These are different knowledge gaps.
An adaptive learning system should try to identify what the learner is struggling with, not simply record that they selected option C instead of option B.
The next teaching interaction can then be selected because it provides useful learning or useful evidence.
The learning path does not need to be identical for every student.
Adaptive approaches are not only theoretically attractive. A systematic review and meta-analysis of adaptive e-learning for health professionals and students found adaptive environments particularly effective for improving skills, rather than only factual knowledge [1]. A broader meta-analysis of 50 controlled evaluations of intelligent tutoring systems reported a median improvement of 0.66 standard deviations over conventional instruction, roughly the difference between the 50th and the 75th percentile [2].
In assessment, computerised adaptive testing has a long history of tailoring item difficulty to the individual learner [3], and perceptual and adaptive learning modules have shown strong results for pattern recognition in image-based disciplines such as radiology [4]. The learner modelling approach described in this article builds on knowledge tracing, an approach first published three decades ago [5].
None of this guarantees that any particular adaptive product works.
It does mean the approach is grounded in published evidence rather than novelty, which is why we are building on those foundations and validating as we go. Full references are listed at the end of this article.
There are facts radiography students need to remember.
We are not trying to remove recall from education.
But medical imaging practice requires considerably more than remembering definitions.
A student may be able to define spatial resolution perfectly and still struggle to identify loss of spatial resolution in an image.
They may recall the factors affecting fluoroscopic dose but make a poor decision during a simulated procedure.
They may understand image intensifier theory but struggle to critique the resulting image.
Knowing a fact and applying knowledge are related, but they are not always the same educational task.
This is why our adaptive learning experiences can examine knowledge through different question and interaction types.
The aim is to ask:
For medical imaging education, those distinctions matter.
The obvious answer is to provide more teaching.
The more difficult question is: what teaching?
Giving the learner the same explanation again may not solve the problem.
Adaptive teaching should try to respond to the pattern.
For example, a learner who repeatedly struggles with image noise may be directed towards a focused explanation and a new visual example.
A learner who understands noise but struggles when exposure factors are introduced may need the relationship between image quality and technique reinforced.
Another learner may already demonstrate secure understanding and gain little from repeating the same foundation material.
The goal is not to make learning easier.
The goal is to make the teaching more relevant to the learner's current understanding.
Image quality is a useful example because it connects physics, equipment, image interpretation, and clinical decision-making.
Students may need to consider concepts such as spatial resolution, temporal resolution, noise, contrast, and magnification.
These concepts do not exist independently in clinical imaging.
Changes to the imaging technique can affect the resulting image.
Clinical requirements affect decisions.
Image quality must often be considered alongside radiation dose.
That makes fluoroscopic image quality a useful area for exploring adaptive medical imaging education.
Our current demo is intentionally small.
It is not intended to represent a complete fluoroscopy curriculum.
It is a way of showing how a short learning experience can begin by responding to the learner rather than simply presenting content.
We want to be transparent about how our adaptive learning experiences make decisions, and equally transparent about what has and has not yet been validated.
Concept and prerequisite mapping. Each learning experience is built on a concept map authored by radiography educators. Every concept is linked to its prerequisites. Understanding image noise, for example, depends on prior understanding of exposure factors.
Question tagging. Every question and interaction is tagged to one or more concepts and to a cognitive level: recall, recognition, explanation, application, or decision-making. This is what allows the system to distinguish a definition gap from an application gap.
Evidence weighting. Responses are weighted by interaction type and difficulty. Correctly critiquing an image provides different evidence from correctly answering a recall question, and both contribute to the same concept estimate.
Mastery estimation. After each interaction, a Bayesian model updates a probability estimate for every concept involved, an approach known as knowledge tracing [5]. Estimates rise and fall with the evidence rather than being fixed by a single answer.
Guessing and uncertainty. The model includes explicit parameters for guessing and for careless slips. A lucky click does not establish mastery, and one careless error does not erase it.
Difficulty calibration. Items are assigned an initial difficulty by subject matter experts and refined as response data accumulates.
Mastery thresholds. A concept is treated as secure only when its estimate passes a defined threshold with consistent evidence across interaction types. Below that threshold, the system selects targeted teaching or a prerequisite review.
Selecting the next activity. The next question, image comparison, or explanation is chosen for its expected learning or diagnostic value, using the current concept estimates and the prerequisite map.
Human oversight. All clinical content is authored and signed off by qualified radiography educators. The algorithm selects and sequences learning activities; it never generates or alters clinical content. Educators see the resulting analytics and retain full professional judgement.
Data privacy. The public demo runs in your browser and does not ask you to create an account.
Current validation status. The model design draws on published research in adaptive e-learning, adaptive testing, and knowledge tracing [1][3][5]. Our own implementation is at the demonstration stage: content is expert-reviewed and internally tested, and formal validation studies of learning outcomes are planned. We will publish results, including limitations, as that work matures.
Ideally, it should not feel like an algorithm is examining them.
The student answers questions, interacts with images, and receives teaching.
Behind the learning experience, their interactions contribute to a developing picture of their understanding.
A learner struggling with one concept may receive additional support.
A learner demonstrating stronger knowledge may progress to a more complex application.
Different areas of knowledge can be considered separately.
In medical imaging, a student could be developing strongly in anatomy while struggling with positioning.
They may understand radiation protection theory but have difficulty applying it during a clinical decision.
A single percentage score can hide those differences.
Adaptive learning gives us the opportunity to make them more visible.
The same information that supports adaptation for students may also provide better information for educators.
Instead of seeing:
Average quiz score: 74%
An educator could potentially see:
Spatial resolution: secure
Temporal resolution: developing
Image noise: common cohort misconception
Magnification: strong recall but weak application
That is a very different conversation.
It may help an educator decide what needs to be retaught, where simulation time could be focused, and which learners may benefit from additional support.
The purpose is not to replace the radiography educator.
It is to give the educator a clearer picture of learning.
Here is how the decision logic plays out for one learner in the fluoroscopic image quality demo. The exact wording and images vary in the live demo, but the underlying logic is the same.
An opening image quality question. The learner is asked which change would most improve a degraded fluoroscopic image.
What the model records: First evidence arrives across several image quality concepts at once.
An incorrect temporal resolution response. The learner attributes motion blur to poor spatial resolution.
What the model records: The temporal resolution estimate drops, and the response pattern suggests the two concepts may be confused with each other.
A diagnostic follow-up. The system presents a paired image comparison designed to separate the two concepts.
What the model records: This item is selected for diagnostic value. Whichever way the learner answers, the model learns something specific.
A targeted explanation. A short teaching sequence addresses temporal resolution directly, using a moving structure example rather than repeating the original content.
What the model records: Teaching is a response to the identified confusion, not a generic incorrect answer message.
An application question. The learner applies the concept to a pulsed fluoroscopy scenario.
What the model records: A correct response here is stronger evidence than the earlier recall item, and it is weighted accordingly.
Updated concept estimates. Temporal resolution moves from uncertain to developing, while spatial resolution remains secure.
What the model records: The learner continues to new material rather than repeating what they already know.
We are at the beginning of this work.
The fluoroscopic image quality demo is deliberately a small example, but it shows the direction Virtual Medical Coaching is taking with adaptive medical imaging education.
Rather than simply asking whether a student passed a quiz, we are interested in a more useful question:
What does this learner appear to understand, and what should we teach next?
Try the demo:
Open the demoI would genuinely be interested to hear what radiographers, medical imaging educators, and students think.
Adaptive learning in radiography education uses learner interactions to build an evolving picture of understanding and adjust teaching, feedback, or learning pathways accordingly. Instead of every student receiving the same fixed sequence, the experience can respond to areas of strength, uncertainty, or difficulty.
One correct answer should not automatically be treated as proof of mastery. Adaptive learning can examine patterns across multiple questions, related concepts, question difficulty, and different types of applications. Consistent evidence provides a stronger indication of understanding than one isolated correct response.
No. Recall can be assessed, but medical imaging learning also involves recognition, application, image interpretation, and decision-making. Adaptive learning experiences can use different interaction types to explore whether a learner can apply knowledge rather than simply remember a definition.
A normal quiz usually presents a fixed set of questions and calculates a score. Adaptive learning uses learner responses as evidence about developing knowledge. That information can influence subsequent questions, feedback, explanations, or learning activities.
It is a simple form of adaptation, but adaptive learning can be more sophisticated. The system can consider which concept is being assessed, patterns across previous interactions, possible prerequisite gaps, and whether the learner can apply knowledge in different contexts.
Yes. A correct answer may result from knowledge, partial understanding, recognition, elimination of incorrect options or guessing. This is why mastery should ideally be estimated using a pattern of evidence rather than a single answer.
Medical imaging students must connect anatomy, physics, image quality, positioning, radiation protection, and clinical decision-making. A student can be strong in one area and struggle in another. Adaptive learning can help identify those differences and provide more targeted learning support.
No. The aim is to give educators better information about learner progress and knowledge gaps. Educators remain responsible for teaching, professional judgement and supporting students. Adaptive learning can provide additional evidence to help them decide where teaching time may be most useful.
No. The current demo is intentionally short and is designed to show the direction Virtual Medical Coaching is taking with adaptive learning. Future learning experiences can contain larger question banks, different interaction types, and more detailed learner and educator analytics.
The Virtual Medical Coaching fluoroscopic image quality adaptive learning demo is available here:
Virtual Medical Coaching is developing adaptive teaching and learning for medical imaging to support student mastery, cohort insight, and educator...
Discover how VR and AI converge in medical education to deliver immersive simulations, real-time adaptive feedback, and data-driven personalized...
Learn how immersive VR transforms education by boosting engagement, retention, and personalized learning
Subscribe to our news, updates and more.