Auxiliary Diagnosis of Knee Joint Osteoarthritis Based on Vibroarthrographic Signals

Abstract

It is a significant task to automatically diagnose knee joint osteoarthritis at a low cost. Using the micro-vibration signal produced by the knee joint during flexion and extension activity should be a valuable implementation approach. In this study, subjects are diagnosed and annotated by expert physicians via knee joint X-ray examination. Then subjects participated in several stand up-squat down-stand up recycles and the vibroarthrographic (VAG) signals of the patella’s surface are measured through wearable sensors for diagnosis and grading of Knee disease. In 2-class classification, all subjects are classified into the knee osteoarthritis (KOA) group and the control group (CG). In 3 or 5-class classification, they are classified into several different abnormal groups with increasing heavier symptoms and the control group (CG). Patient features, time domain features, frequency domain features and wavelet features are well combined for auxiliary diagnosis. Several algorithms, XGBoost, LightGBM, CatBoost and Random Forest are used as classifiers. With the proposed feature grouping-crossing method, the accuracy of 2-class, 3-class and 5-class classifications are improved to 90.75%, 73.57% and 47.82%, respectively. The results show that a helpful auxiliary diagnosis framework of knee joint osteoarthritis is obtained through efficient signal acquisition pattern and the good machine learning method. The collection of VAG signals has the advantages of being non-invasively, non-radiation, low-cost and convenient. Based on the proposed method, we can identify and classify early osteoarthritis and the articular cartilage during outpatient clinic visits, which has enormous clinical treatment and health care prevention potential.

Note:
Funding Declaration: This research was funded by the Natural Science Foundation of Shaanxi Province (No. 2022JQ-661, No. 2021SF-189, No. 2019ZDLGY03-02-02), the Fundamental Research Funds for the Central Universities (Grant No. XJS222215, XJS222221) and by Military medical Everest project of The Fourth Military Medical University (No. 2018YYZT03).

Conflicts of Interest: None

Ethical Approval: This research is approved by the medical ethics committee of the First Hospital Affiliated of The Fourth Military Medical University (Xijing Hospital, approval no. KY20232197-C-1). All the subjects have the right to know the objective, significance, and purpose of the data collecting. All the subjects have signed the informed consent and their personal information would be fully maintained confidentiality.

Keywords: Vibroarthrographic (VAG) Signals, Knee Osteoarthritis (KOA), Ensemble learning, Feature Grouping-crossing