Classification of vibroartographic signals based on wavelet transformation and machine learning techniques

: pp. 40-52
Lviv Polytechnic National University, Department of Information-Measurement Technologies

Vibroartography is a method of medical diagnosis, designed for objective estimation of
human joint motor function in general and arthrokinematics of the knee joint in particular. The method is based on the analysis of signals of vibroacoustic emission. Vibroartography is
not so effective compared to methods such as radiography and magnetic resonance imaging
(MRI), but it is definitely a sensitive method for assessing the degree of knee joint dysfunction.
This paper presents the research results related to the design of a system for
vibroarthrographic signals computer processing. VAG signals are non-stationary, therefore,
for their analysis in this study, the discrete wavelet transform is used, which allows studying
not only the frequency components of the signal, but also their time localization. The novelty of
the proposed approach is based on the application of a discrete dyadic wavelet transform to
clear the biosignal from the impact of isolines drifts and random noise, as well as the use the
wavelet coefficients to form the diagnostically significant features. Scalogram analysis for 6
levels wavelet transforms allowed identifying a band from 78 to 780 Hz, where useful
diagnostic information is concentrated. Reconstruction of the signal in the specified band
resulted in the elimination of the destabilizing effects. After processing results of the wavelet
transform, twelve descriptors were chosen: standard deviation, mode and means of absolute
values for the four signal components. The results obtained for two classes scenario are the
following: accuracy of 94 %, sensitivity of 100% and specificity of 88%. For five classes
accuracy of 83%, sensitivity of 89% and specificity of 62% were achieved. Thus two classes
scenario demonstrated both high accuracy and sensitivity, while five classes scenario
demonstrated moderate results. The biggest overlap of descriptors is observed for the
neighboring classes. The main constraint in this study was a small number of signals – 26 in
each class. The duration of each recording is 6 seconds, at a sampling frequency of 10 kHz.
Records were separated into classes based on corresponding MRI images for each patient.

1. Wu Y. Knee Joint Vibroarthrographic Signal Processing and Analysis, Knee Joint Vibroarthrographic Signal Processing and Analysis, Springer, London 2015.

2. Bączkowicz D., Majorczyk E. Joint motion quality in vibroacoustic signal analysis for patients with patellofemoral joint disorders, BMC Musculoskelet Disord, 2014;15:426–433.

3. Bączkowicz D., Majorczyk E. Joint motion quality in chondromalacia progression assessed by vibroacoustic signal analysis, Pm&r 8 (2016) 1065–1071.

4. Dołęgowski M., Szmajda M., Bączkowicz D. Use of incremental decomposition and spectrogram in vibroacoustic signal analysis in knee joint disease examination // Przeglad elektrotechniczny, 2018, Nr 7, p. 162–166.

5. Krecisz K., Baczkowicz D. Analysis and multiclass classification of pathological knee joints using vibroarthrographic signals, Computer Methods and Programs in Biomedicine, 154, 2018, 37–44.

6. Rangayyan R. M., Oloumi F., Wu Y., Cai S. Fractal analysis of knee-joint vibroarthrographic signals via power spectral analysis, Biomed. Signal Process. Control. 8 (2013) 23–29.

7. Baczkowicz D., Majorczyk E. Joint motion quality in vibroacoustic signal analysis for patients with patellofemoral joint disorders, BMC Musculoskeletal Disord. 15, 2014.

8. Ferreira Moreira D. B. Classification of knee arthropathy with accelerometer-based vibroarthrography // Dissertation submitted to Faculdade de Engenharia da Universidade do Porto to obtain the degree of Master in Bioengineering, 2015, 112 p.

9. Song C. G., Kim K. S., Seo J. H. Non-invasive monitoring of knee pathology based on automatic knee sound classification. In Proceedings of the World Congress on Engineering and Computer Science, San Francisco, USA, 2009.

10. Meng Lu, Suxian Cai, Fang Zheng, Shanshan Yang, Ning Xiang, and Yunfeng Wu. Adaptive noise removal of knee joint vibration signals using a signal power error minimization method. In Computing and Convergence Technology (ICCCT), 2012 7th International Conference on, pp. 1193–1196. IEEE, 2012.

11. Wu Y., Yang S. Zheng F., Cai S., Lu M., Wu M. Removal of artifacts in knee joint vibroarthrographic signals using ensemble empirical mode decomposition and detrended fluctuation analysis. Physiol. Meas. 35 (3), 429–439.

12. Krishnan S., Rangayyan R. M. Automatic denoising of knee-joint vibration signals using adaptive time-frequency representations. Medical and Biological engineering and Computing, 38(1):2–8, 2000.

13. Rangaraj M. Rangayyan. (2002) Biomedical Signal Analysis. A Case-Study Approach. Jhon Willey and Sons Inc. 556 pp.

14. Smolentsev N. K. Basics of wavelet theory. Wavelets in MATLAB. – M.: DKM, Press, 2014. – 628 p.

15. MathWorks. Support. Documentation. Retrieved from

16. Rangayyan R. M., Wu Y. Screening of knee-joint vibroarthrographic signals using probability density functions estimated with Parzen windows, Biomed. Sig- nal Process. Control 5 (2010) 53–58.

17. Pihlajamäki H. K., Kuikka P.-I., Leppänen V.-V., Kiuru M. J., Mattila V. M. Reliability of clinical findings and magnetic resonance imaging for the diagnosis of chondromalacia patellae, J. Bone Jt. Surg. Am. 92 (2010) 927–934.

18. Samim M., Smitaman E., Lawrence D., Moukaddam H. MRI of anterior knee pain, Skeletal Radiol. 43 (2014) 875–893.

19. Tanaka N., Hoshiyama M. Vibroarthrography in patients with knee arthropa- thy, J. Back Musculoskeletal Rehabil 25 (2012) 117–122.