Abstract
Anterior knee pain is one of the most frequently reported musculoskeletal complaints in all age groups. However, patient's complaints are often nonspecific, leading to difficulty in properly diagnosing the condition. One of the causes of pain is the degeneration of the articular cartilage. As the cartilage deteriorates, its ability to distribute the joint reaction forces decreases and the stresses may exceed the pain threshold. Unfortunately, the assessment of the cartilage condition is often limited to a detailed interview with the patient, careful physical examination and x-ray imaging. The X-ray screening may reveal bone degeneration, but does not carry sufficient information of the soft tissues' conditions. More advanced imaging tools such as MRI or CT are available, but these are expensive, time consuming and are only suitable for detection of advanced arthritis. Arthroscopic surgery is often the only reliable option, however due to its semi-invasive nature, it cannot be considered as a practical diagnostic tool. However, as the articular cartilage degenerates, the surfaces become rougher, they produce higher vibrations than smooth surfaces due to higher friction during the interaction. Therefore, it was proposed to detect vibrations non-invasively using accelerometers, and evaluate the signals for their potential diagnostic applications.
Vibration data was collected for 75 subjects; 23 healthy and 52 subjects suffering from knee arthritis. The study was approved by the IRB and an Informed Consent was obtained prior to data collection. Five accelerometers were attached to skin around the knee joint (at the patella, medial and lateral femoral condyles, tibial tuberosity and medial tibial plateau). Each subject performed 5 activities; (1) flexion-extension, (2) deep knee bend, (3) chair rising, (4) stair climbing and (5) stair descent. The vibration and motion components of the signals were separated by a high pass filter. Next, 33 parameters of the signals were calculated and evaluated for their discrimination effectiveness (Figure 1). Finally the pattern recognition method based on Baysian classification theorem was used for classify each signal to either healthy or arthritic group, assuming equal prior probabilities.
The variance and mean of the vibration signals were significantly higher in the arthritic group (p=2.8e-7 and p=3.7e-14, respectively), which confirms the general hypothesis that the vibration magnitudes increase as the cartilage degenerates. Other signal features providing good discrimination included the 99th quantile, the integral of the vibration signal envelope, and the product of the signal envelope and the activity duration. The pattern classification yielded excellent results with the success rate of up to 92.2% using only 2 features, up to 94.8% using 3 (Figure 2), and 96.1% using 4 features.
The current study proved that the vibrations can be studied non-invasively using a low-cost technology. The results confirmed the hypothesis that the degeneration of the cartilage increases the vibration of the articulating bones. The classification rate obtained in the study is very encouraging, providing over 96% accuracy. The presented technology has certainly a potential of being used as an additional screening methodology enhancing the assessment of the articular cartilage condition.
