Robustness analysis of soft Gaussian Mixture Model clustering for acoustic emission features in characterizing osteoarthritic knees

Abstract

Acoustic emission (AE) is a well-established non-destructive evaluation (NDE) method that currently holds enormous potential for the early detection of knee osteoarthritis (OA). Knee joints have intrinsic complexity, resulting in marked variability of the obtained AE signals. This problem complicates the distinction between the AE signatures of different knee conditions. In this regard, Machine learning (ML) algorithms, particularly the Gaussian Mixture Model (GMM), can solve this problem by identifying the overlapping data points generated from different knee joint conditions. Early studies had limitations in the generalizability of their findings due to the small dataset. Therefore, a comprehensive evaluation of the robustness of soft GMM clustering in handling overlapping data points is needed. The current study constitutes further efforts to bridge this knowledge gap by investigating the robustness of GMM clustering in detecting overlapping AE data from knee joints. This study presents a comprehensive statistical analysis of cluster properties before and after soft GMM clustering to identify and remove overlapping data points. The results of this investigation confirm the robustness of soft GMM in clustering AE features for the intelligent assessment of knee health.