NHS doctors build AI that spots kidney failure 6x faster

Tracing the outline of a diseased kidney across 50 to 60 MRI slices by hand takes about an hour of intense screen staring.

Doctors at Sheffield Teaching Hospitals NHS Foundation Trust in the UK just handed that tedious chore over to an algorithm, and the results arrive in less than a minute.

Skipping an hour of tedious tracing

The software is built specifically for autosomal dominant polycystic kidney disease, or ADPKD. It is a common inherited condition causing fluid-filled sacs to enlarge the kidneys, affecting between 30,000 and 70,000 people in the UK alone.

To predict how long a patient has before their kidneys give out, doctors measure total kidney volume. Traditionally, analysts sit down with an MRI scan and manually trace the organ on dozens of stacked 2D image slices.

Lead nephrologist Professor Albert Ong and principal scientist Jonathan Taylor developed an algorithm trained on hundreds of MRI scans from a pan-European study. Radiographer Richard Thomas had manually outlined all the kidneys in that dataset to teach the system how to trace properly.

The resulting tool matches the precision of human expert analysts while completing the process roughly six times faster than manual methods. Sheffield's 3D Lab has actually been using it quietly in clinical workflows since 2022.

Buying precious time before dialysis

ADPKD typically pushes patients into kidney failure around age 50, which is relatively young for organ failure. Once a patient reaches that stage, damage is irreversible, leaving dialysis or a kidney transplant as the only remaining options.

Automating the volume measurement gives specialists a fast and accurate projection of when failure might occur. That timeline helps doctors step in with targeted treatments before the worst happens.

Ong also pointed out that the software can easily scale beyond elite academic centers. Clinics worldwide that lack dedicated image analysts can now calculate precise kidney volumes without specialized in-house staff.

Why it matters

We think Sheffield's approach is a textbook example of where clinical AI actually shines. Instead of trying to reinvent the entire diagnosis, the team took the single most labor-intensive bottleneck in the existing workflow and wiped it out.

For patients and clinic staff, the payoff is immediate. Slashing an hour-long manual chore down to under sixty seconds frees up specialists to spend time with patients rather than tracing pixels. It also brings expert-level prognostic precision to community clinics that never had the luxury of dedicated imaging staff.

For the wider medical tech industry, this puts real pressure on commercial renal imaging platforms. A free-standing or embedded tool that has operated stably in an NHS 3D lab since 2022 shows that targeted, high-utility automation is ready for prime time right now.

The main caveat lies in generalization. The model was trained on MRI data from a single European research consortium. Before it rolls out globally as promised, clinicians will need to see whether it holds up equally well against varied scanner hardware, noisy real-world imaging artifacts, and atypical cyst geometries across diverse populations.