Monitoring lung volume are critical in determining lung efficiency and assessing pulmonary function, which are useful for treatment planning and clinical follow-up. The objective of this study was to estimate human lung volume using bioimpedance measurement data by deep learning. In this study, the measurement voltages obtained from real experiments were used for simulate lung volume information. The neural network technique was used to estimate the left- and the right-lung volume by using the bioimpedance values.
Two and three network architectures were investigated by simulation. The result shows that the two-layer neural networks can efficiently estimate lung volumes with the error less than 0.78 ml and with the correlation higher than 0.99. Even adding noise to the degree of 40 dB signal-to-noise ratio, the performance was still satisfactory. A two-layer network model was then sufficient for this lung application.
