We verified that arteriovenous fistula (AVF) function can be evaluated from shunt sound using a body-conducted sound sensor (BCS) with a wide-band frequency characteristic. In the BCS, we compared two types of sensors: an air-coupled microphone (ACM) and an acceleration sensor (Type 8001). Evaluation for sensors were performed for 36 patients. We also developed a shunt-sound observation system and evaluated it using the spectrogram and power spectrum values obtained using a sound spectrometer. Furthermore, we verified that two indices of ultrasonography information, namely, the resistance index (RI) and flow volume (FV), can be estimated from the shunt sound and classified through machine learning. The BCS detected the shunt sound signal in the spectrogram over a wide frequency band. The sensitivities of the BCS in the non-amplified detection of stenosis sound signals were 0.93/0.92 (below 1 kHz/above 1 kHz, respectively), surpassing those of other sensors(ACM: 0.82/0.77, Type 8001: 0.82/0.25). Furthermore, the highest machine learning accuracy was obtained using BCS in the classification segments of RI > 0.60, 0.65, and 0.70 and FV < 350 ml/min. BCS is particularly suitable for detecting stenosis sounds of over 1 kHz and estimating RI and is considered to be suitable for functional evaluations of AVF.
View full abstract