Artificial inteligence reading of cystometric traces provides good correlation with human diagnosis

This research examined whether machine-learning systems could classify detrusor overactivity from cystometric traces in a way that correlates with expert human diagnosis. It is a separate urology project from UroDiary.

The study

The retrospective, single-centre study used 517 adult cystometry traces collected during 2023. Two hundred traces formed the training set and 317 were used for testing, with expert consensus providing the reference diagnosis.

Two complementary approaches were evaluated. A VGG16 convolutional neural network classified images of the traces with 75% accuracy. A wavelet-based method using Daubechies features reached 84.2% accuracy, 86.3% sensitivity, and 82.6% specificity while also producing quantitative descriptions of contractions.

An Isolation Forest model was used to detect artefacts including coughs, open lines, and catheter movement. Classification took under 20 seconds per study.

What stayed with me

The result is most useful as assisted interpretation, not as a replacement for clinical judgment. Combining automated detection with expert supervision can make reporting faster and more reproducible while keeping accountability with the people responsible for the diagnosis.

The work received the Best Abstract Award at the 2024 International Continence Society Congress and was published in the World Journal of Urology in December 2025.