Q & A: AI assisted catheters

Q & A: AI assisted catheters: Anima Anandkumar is at the forefront of research into developing a new, reduced infection catheter using machine learning.

1.Why is this development so important? Are catheter-associated UTIs a major health hazard?

Catheter induced infections are a major health hazard with over 500000 cases annually – they are in fact one of the primary healthcare associated sources of infection.

2.How did you use AI in this research? What difference has AI made? Did it make your research quicker and more accurate?

To prevent bacteria from swimming upstream and infecting the human we are taking advantage of fluid dynamics. The ridges inside the tube cause vortices that the bacteria cannot swim past. However, for this to work, the shape of these ridges needs to be exactly right or the bacteria won’t be stopped. In the past this would have been done by trial and error: build a version of the catheter, try it out in the lab, observe and hypothesize what went wrong, and repeat it. This process would take weeks or even months. Instead, our AI model based on Neural Operators was able to simulate and understand the fluid behavior and directly propose an optimized design.

We 3D printed the design only once to verify that it worked. When testing the catheter in the lab; we measured the reduction in bacterial contamination by a 100-fold.

3.Did the shape of the AI-optimised catheter surprise you? Don’t bacteria clog up the inner tube of the catheter with all those spikes?

The shape is not much of a surprise since we gave it a range of options of shapes and angles to choose from. The hard part is choosing the correct angles to induce the fluid behavior we wanted which would have required a large number of experiments otherwise.

4. So far, you have tested the prototype in the lab. When may it be tested in a clinical setting?

We are currently working on making this a reality. As you would expect, taking this through clinical trials will take some time. My colleagues have a grant to do further studies to enable this

5. How much potential does AI modelling have in the medical field? Could it help design equipment like hip replacements, or even new drugs?

Yes – the catheter is only the first step. My goal is to build AI that understands all areas of physics and solve engineering problems across domains – this includes things like medical devices, but also being helpful for designing better airplanes, rockets and other areas. We have also done work in drug discovery – our works modeling quantum dynamics of molecules using AI (OrbNet and NeuralPlexer) was used to design a drug now in clinical trials at Iambic Therapeutics, a company a colleague of mine started based on our work together.