Estimation of Urine Volume and Urine Conductivity Using Electrical Bioimpedance Based on the Neural Network Method

Taweechai  Ouypornkochagorn; Phirakrit  Chiangchin; Napatsawan  Ngamdi; Tudsanee  Limpisophon; Anurak  Dowloy

doi:10.18502/fbt.v11i3.15885

Taweechai Ouypornkochagorn Department of Biomedical Engineering, Faculty of Engineering, Srinakharinwirot University, Thailand
Phirakrit Chiangchin Department of Biomedical Engineering, Faculty of Engineering, Srinakharinwirot University, Thailand
Napatsawan Ngamdi Department of Biomedical Engineering, Faculty of Engineering, Srinakharinwirot University, Thailand
Tudsanee Limpisophon H.R.H. Maha Chakri Sirindhorn Medical Center, Nakhonnayok, Thailand
Anurak Dowloy Faculty of Medicine, Srinakharinwirot University, Nakhonnayok, Thailand

DOI: https://doi.org/10.18502/fbt.v11i3.15885

Keywords: Urine Volume; Urine Conductivity; Bladder; Fat Content; Neuron Network.

Abstract

Purpose: Urine volume and urine conductivity monitoring allow better care for urinary tract infection disease. Urine volume and conductivity involve electrical bioimpedance change at the lower abdomen. In previous studies, bioimpedance has been only used for estimating the volume, and the estimation error significantly increases when the conductivity changes.

Materials and Methods: In this work, the neuron network technique is proposed to determine both the volume and the conductivity based on the measured bioimpedance data on a sixteen-electrode configuration. Nine architectures of neuron networks were investigated by simulation. Eleven body models were created, consisting of muscle, fat, pelvis bone, rectum, and bladder. Seven bladder sizes, eleven conductivities, and eight levels of Signal-to-Noise Ratio (SNRs) were simulated.

Results: The result showed that the neural network method could efficiently estimate with an average of 1.04% volume error and 2.85% conductivity error. The performance remained stable with a signal-to-noise ratio higher than 60 dB, but it may reduce 2-8 times at lower SNRs. The moderate fat content provided high performance. The performance would be worsened if the bladder size was very small and the conductivity was low. The performance was increased when the volume was moderate, i.e. 302 ml, and the conductivity was higher than 1.76 S/m. The 3-layer architecture with 1024, 512, and 2 neurons yielded the highest performance. The 2-layer architecture with hidden neurons higher than 512 provided a comparative performance with only 0.9-1.5% lesser performance.

Conclusion: Neural network technique can be used to estimate urine volume and urine conductivity with excellent performance.