Incredible Algorithms - Imperial Inaugurals

Today I attended an inaugural lecture at Imperial College by Professor Bernhard Kainz, who works in Medical Image Computing. First things first he wanted to talk about how people imagine research. We often picture a simple path: a discovery is made, a company develops it, and eventually it reaches people. In reality, it’s far messier. Between an idea and real-world impact there are countless obstacles, experiments, failures, and iterations.

A brilliant example came from his work with medical imaging of babies in the womb. Capturing clear MRI scans is incredibly difficult because babies constantly move. Unlike adult scans where a patient can stay still, here the images are constantly shifting. So he explained how instead of relying on a single scan, researchers take many images from different angles and reconstruct them computationally. Algorithms essentially piece together fragments into a coherent picture.

But detecting disease introduced another challenge: most babies are healthy. That means there is very little training data showing what abnormalities actually look like. His team solved this in a really clever way - they generated realistic medical abnormalities themselves. By creating plausible malformations in scans, they could massively expand the dataset and help algorithms learn the difference between normal and abnormal cases. What I found especially interesting was how they organise these scans into a kind of multidimensional space, where similar conditions sit closer together. Populating that space with both real and synthetic examples allows the algorithm to understand subtle differences much better.

The final part of the talk explored something that really caught my attention: the role of AI in accelerating research itself. Scientists often have hundreds or thousands of hypotheses they would like to test, but simply don’t have the time. Professor Kainz explained that large language models can now help run many of the technical steps needed to test ideas, dramatically speeding up the process. Instead of spending years exploring a few possibilities, researchers can explore far more ideas much faster. 

One audience question I loved asked whether AI might eventually create its own scientific hypotheses. His answer was revealing: while that may be a direction AI moves toward, he believed current systems still lack true creativity. Since they generate ideas based on patterns in existing research rather than genuinely new insights he felt that hypothesis generated ended up being quite mundane and that currently at least, creativity is in the human's hands. Another question was about whether AI could explain why someone develops a disease. His answer was refreshingly honest: even scientists often struggle to fully explain that themselves. And understanding what is happening is already difficult enough so he said its currently too hefty of a task for it right now and that he believes that details about the disease is more important at the moment.

Walking out of the lecture, what struck me most was how powerful computer science has become in driving progress across entirely different fields. This is exactly why I love it so much. Computer science  unlocks possibilities. It automates the tedious work, lets researchers explore more ideas, and ultimately pushes humanity forward faster than ever before. Seeing that intersection between algorithms, innovation, and real-world impact was incredibly inspiring. I'm so excited for what the future will bring!