How Stable Diffusion Could Advance Medical Imaging

As the use of medical imaging grows, the work involved can be more specialised and means that getting enough training to help make the best use of medical imagining solutions is a challenge.

However, new developments may help make this task easier, including a new innovation that follows on from one devised and launched only last year.

As Venturebeat reports, the launch last August of Stable Diffusion, a text-to-image foundational model, by Stability AI, prompted an idea from Stanford University radiologist Christian Bluethgen. He asked whether in fact it was possible to combine a genuine medical need with the creating of high quality images using basic text prompts.

The result was his collaboration with Stanford Graduate student Pierre Chambon, a mathematical and computational engineering researcher, which led to a study designed to establish the capacity of stable diffusion to generate X-rays. To their delight, they found it achieved this task very well.

All this, Mr Bluethgen, will help with the training of medics who might otherwise see very few scans relevant to their specialist area. He observed: “When you are working in a setting with scarce data, your performance correlates with experience – the more images you see, the better you become.”

Such a development may be just one indication of how better use of data technology is enabling medical imaging to be used with increasing effectiveness as a diagnostic tool, with improved training further boosting the capacity of specialists to identify medical problems when they emerge.

A further development in X-ray imaging may have emerged at the University of New South Wales in Sydney. In a paper published in the journal Nature Communications, researchers have revealed how a new algorithm has been devised that can enhance images of hydrogen fuel cells. It notes that the very same technique could be used to improve medical imaging.