Predicting Bladder Pressure and Contractions with Dense Time-Series Data

Bladder dysfunction due to spinal cord injury can result in incontinence and the inability to effectively void the bladder. Electrical stimulation of nerves in the bladder during a contraction can inhibit bladder contractions (eliminating incontinence) or excite bladder contractions to ensure the bladder is completely voided. However, determining when a bladder contraction will occur remains an active area of research. Our goal is to infer bladder pressure from external urethral sphincter electromyography (EUS EMG) readings from experimental data using rats. Due to the extremely dense time-series data, traditional mathematical modeling techniques are not applicable. Instead, we employ statistical methods (such as LASSO) and machine learning methods (recurrent neural networks) to make predictions of bladder pressure from external nerve data. Furthermore, to address inter-individual heterogeneity between rats, we applied a multi-task learning algorithm in which each individual rat’s prediction was a separate task – producing more generalizable results. These bladder pressures were then used to predict the onset of bladder contractions with high sensitivity and specificity.