Advance signal processing and machine learning approach for analysis and classification of knee osteoarthritis vibroarthrographic signals

Abstract

Osteoarthritis is a common cause of disability among elderly significantly affecting their quality of life due to pain and functional limitations. This study proposes a novel, non-invasive, and cost-effective diagnostic technique using vibroarthrography (VAG) for early detection and grading of knee osteoarthritis (KOA) overcoming the limitations of traditional methods like X-rays, CT scans, and MRIs. Signal acquisition involved capturing of VAG signals from KOA patients using Thinklabs One digital stethoscope and a specialized knee brace within a frequency range of 20 Hz to 2000 Hz with a ± 3 dB tolerance at 44,000 samples per second. Various signal processing techniques, like time domain, statistical, PSD, wavelet, and Hilbert-Huang transform analysis, were used to study the resultant signal. Subsequently, a novel combination of self-organizing maps (SOMs) and K-means clustering was proposed to categorize VAG signals into distinct OA grade clusters. The resulting analysis identified distinct patterns in the time domain correlating with joint alteration severity. A SD/Mean ratio differentiated OA grades. Hilbert-Huang Transform established intrinsic mode functions relating frequency bands to OA stages, while wavelet and spectrogram analysis demonstrated increased signal complexity and variability with disease progression. The effectiveness of proposed clustering model was indicated by high mean Silhouette Coefficient (∼0.80) and low Davies-Bouldin Index (∼0.33) indicating distinct and accurate segmentation of OA stages. These findings clearly highlighted the potential of SOMs and K-means clustering in analysing VAG signals for classifying into different KOA grades. These results demonstrate the substantial potential of advanced signal processing, SOMs, and K-means clustering in uncovering complex patterns in VAG data, linking increasing knee sound signal complexity with OA progression. This highlights the potential of our approach in medical diagnostics, especially for chronic conditions like KOA, where early detection and ongoing monitoring are crucial.