Seeing the Edges: How Curiosity (and a Few Thousand Dots) Can Help Children Smile

When people think about Artificial Intelligence (AI), they often imagine robots, self-driving cars, or maybe ChatGPT. But in our lab at the University of Adelaide, AI is helping us look at something much closer to home: children’s faces.

Recently, my team worked on a project to help doctors better understand a condition called Hemifacial Microsomia (HFM). It is a condition where one side of a child’s face doesn’t develop at the same rate as the other. For some children, it is mild; for others, it can significantly affect their jaw, chewing, speaking, and even breathing.

Doctors have been diagnosing and grading the severity of this condition for decades. But here’s the challenge: it isn't always easy. Medical scans are incredibly complex, assessments often depend on a specialist's individual experience, and sometimes, the subtlest differences matter the most.

This is where curiosity stepped in.

What If We Looked at It Differently?

In science, curiosity often begins with a simple, disruptive question: “What are we missing?”

When we looked at 3D scans of children’s jaws, we realised something fascinating. These scans contain a massive amount of data-tens of thousands of tiny points describing the surface of the bone. In the tech world, we call these “Point Clouds.”

But as we stared at these clouds of dots, we wondered: Are all these points actually helping us, or are they just noise?

Think about drawing a mountain. Where do you put the most detail with your pencil? It’s usually along the sharp ridges and the jagged edges, not in the big, flat, smooth areas. We asked ourselves: What if the most important information isn’t everywhere, but only along the edges?

The Power of Asking "Less"

That simple question led to a new way of analysing medical scans. Instead of forcing an AI system to "swallow" every single point in a massive scan, we taught it to be selective. We trained it to focus on the “edges”- the curved boundaries where the jaw bends and shapes itself. These are the areas that tell the true story of how a bone has developed.

And something wonderful happened. When we reduced the data from 20,000 points down to just 2,000 carefully chosen “edge” points, the system didn't just get faster, it became more accurate. By ignoring 90% of the "distraction," the AI could finally see the "truth." That is the power of a curious question: it allows us to stop looking at everything so we can finally see something.

Curiosity Isn’t About "More" - It’s About "Smarter"

This project taught us a lesson that students in any classroom can relate to: More information isn’t always better. Sometimes, a scientific breakthrough comes from learning what to ignore so you can focus on what truly matters.

When Data is Scarce, Creativity Steps In

We hit another hurdle: HFM is a rare condition. We didn’t have the luxury of "Big Data" or thousands of examples to train our AI. So, we asked another curious question: “How can we learn from what we don't have?”

Instead of waiting for more cases, we used a technique similar to mixing paint. If one jaw shape represents a “mild” case and another represents “moderate,” we realised we could "blend" them to create realistic new training samples. By treating data like colours on a palette, we taught the computer to understand the smooth, continuous transitions of human growth.

Science is a Living Process

One of the joys of being a researcher is realising that science is never "finished." It evolves. It improves. It constantly questions its own assumptions.

Medical diagnosis today looks very different from what it did 20 years ago. With AI and 3D modelling, we are discovering entirely new ways to visualise the human body. But the heart of the process remains the same:

Observe -> Question ->  Test -> Refine.

That cycle is exactly the same in a primary school classroom experiment as it is in a high-tech university lab. It all begins with the courage to be curious and the willingness to look at a cloud of dots and ask: "Is there a better way to see this?"