Estimation of early-stage malfunction using implantable artificial heart sound in animal experiments

We have developed an automatic diagnosis system of an artificial heart in order to ensure the safety of the patient implanted with the artificial heart. The automatic diagnosis system is composed of an electro-stethoscope system, adaptive noise canceller (ANC), and artificial neural network (ANN). The ANC effectively eliminates ambient noise from the sound signal of the artificial heart detected by the electro-stethoscope, and a filtered sound signal is separated into each frequency components by fast Fourier transformation. Each frequency component of an artificial heart's acoustic signal is fed into the ANN in order to make a diagnosis of pump condition. The automatic diagnosis system was evaluated in mock circulatory tests and a long-term animal experiment using a goat implanted with an undulation pump ventricular assist device (UPVAD). In mock circulatory tests, the ANN was able to detect pump failing conditions, which were occlusion of inflow and outflow cannula and deterioration of the ball bearing. In a long-term animal experiment, after training the ANN using UPVAD's sound signal in normal condition, the diagnosis system continuously monitored UPVAD's sound signal detected by the electro-stethoscope placed on the surface of the left thoracic cavity of the goat. The UPVAD was stopped by rupture of a diaphragm in the pump on the ninth day of operation. We were able to identify initial signs of malfunction of the pump on the eighth day, while the UPVAD was able to operate normally. In conclusion, the automatic diagnosis system for malfunction of the artificial heart has enough performance to detect early stages of malfunction of the artificial heart, and it contributes to ensure the patient's safety.