Machine Learning for Urodynamic Detection of Detrusor Overactivity.

Area of Science:

• Urology
• Biomedical Engineering
• Machine Learning

Background:

• Urodynamic Studies (UDS) are crucial for assessing neurogenic bladder in spina bifida patients.
• Variability in UDS interpretation necessitates standardized methods.
• Detrusor overactivity (DO) is a key condition evaluated during UDS.

Purpose of the Study:

• To develop a machine learning algorithm for identifying detrusor overactivity (DO) in Urodynamic Studies (UDS).
• To standardize UDS interpretation in the spina bifida population.
• To improve the assessment and management of neurogenic bladder.

Main Methods:

• A machine learning pipeline was created using data windowing, dimensionality reduction, and support vector machines.
• UDS files from 805 tests (546 patients) were analyzed in time and frequency domains.
• Models were trained to detect clinician-identified DO, incorporating vesical, abdominal, and detrusor pressure channels.

Main Results:

• Machine learning models demonstrated good performance in detecting DO, agreeing with clinician interpretations.
• The time-based model using all 3 channels achieved an AUC of 91.9%.
• The frequency-based model with 3 channels showed high specificity (92.9%).

Conclusions:

• A proof-of-concept machine learning pipeline effectively identifies DO in UDS.
• Machine learning algorithms can standardize UDS interpretation.
• This technology may enhance shared decision-making and patient care for neurogenic bladder.