Combination of wavelet transform and LSTM to detect low functional respiration

抄録

Automatic and long-term monitoring of respiratory is in great demand for lung diseases. It gets required greater in these years due to COVID-19 pandemic to reduce medical staff fatigue for checking patient conditions frequently for long time. Kobayashi et al., in our team, developed a device measuring respiratory condition by quantizing the displacement between the 6th and 8th ribs. We introduce long short-term memory (LSTM) neural network to classify patient respiratory signals into the two states of normal and low-functional respirations. The signals were checked by a medical doctor manually for classified into the two states. In the process, they were transformed to frequency-domain spectra with complex-valued wavelet transform, and then quantized the respiratory wavelet spectra due to the large number of spectra patterns. After that, the LSTM learned and classified the processed respiratory signals. The experimental results showed the feasibility to detect the two states.