Characterisation of arteriovenous fistula's sound recordings using principal component analysis.

In this study, a signal analysis framework based on the Karhunen-Loève expansion and k-means clustering algorithm is proposed for the characterisation of arteriovenous (AV) fistula's sound recordings. The Karhunen-Loève (KL) coefficients corresponding to the directions of maximum variance were used as classification features, which were clustered applying k-means algorithm. The results showed that one natural cluster was found for similar AV fistula's state recordings. On the other hand, when stenotic and non-stenotic AV fistula's recordings were processed together, the two most significant KL coefficients contain important information that can be used for classification or discrimination between these AV fistula's states.