抄録
This paper describes the elemental technology of a system that discriminates whether or not sputum is accumulated (whether or not suction is necessary) based on auscultation sounds using edge devices such as microcomputers, in order to support endotracheal suctioning (a procedure to suction sputum accumulated in the trachea), which is one of nursing tasks. Since a typical FFT library for microcomputers is used, the lung sounds are FFTed not for an entire breath but for a short span, and the machine learning and discrimination of normal and abnormal by frequency spectrum are performed to experimentally verify the discrimination performance. On a high-performance microcomputer SPRESENSE, lung sound data read from a wav files is FFTed over a short span (approximately 0.26 seconds), and the frequency spectrum is used to discriminate normal from abnormal using a neural network model that has been machine learned on a computer in advance. The results showed that the discrimination accuracy is reasonably good, approaching 90%.
