Multiclass SVM for Bladder Volume Monitoring using Electrical Impedance Measurements

Abstract

Urinary incontinence is a common condition that impacts the quality of life from those who suffer from it. Electrical impedance measurements offer the potential for a non-invasive low-cost solution to monitor changes in the bladder volume. This work focuses on using a multiclass support vector machine (SVM) algorithm to classify the fullness of the bladder into three states; not full, full, and a boundary class. This paper applies this machine learning algorithm to both simulation and experimental data. The SVM model uses the recorded voltages from electrical impedance measurements as features, is trained and optimized using a Bayesian Optimization approach, and then 10-fold cross-tested to obtain a generalized error. This paper demonstrates that simulation data with a signal-to-noise ratio of 40 dB, and experimental data from a pelvis phantom, can be perfectly separated into the three classes defined above.