Knee Acoustic Emissions as a Digital Biomarker of Disease Status in Juvenile Idiopathic Arthritis

Abstract

In this paper, we quantify the joint acoustic emissions (JAEs) from the knees of children with juvenile idiopathic arthritis (JIA) and support their use as a novel biomarker of the disease. JIA is the most common rheumatic disease of childhood; it has a highly variable presentation, and few reliable biomarkers which makes diagnosis and personalization of care difficult. The knee is the most commonly affected joint with hallmark synovitis and inflammation that can extend to damage the underlying cartilage and bone. During movement of the knee, internal friction creates JAEs that can be non-invasively measured. We hypothesize that these JAEs contain clinically relevant information that could be used for the diagnosis and personalization of treatment of JIA. In this study, we record and compare the JAEs from 25 patients with JIA-10 of whom were recorded a second time 3-6 months later-and 18 healthy age- and sex-matched controls. We compute signal features from each of those record cycles of flexion/extension and train a logistic regression classification model. The model classified each cycle as having JIA or being healthy with 84.4% accuracy using leave-one-subject-out cross validation (LOSO-CV). When assessing the full JAE recording of a subject (which contained at least 8 cycles of flexion/extension), a majority vote of the cycle labels accurately classified the subjects as having JIA or being healthy 100% of the time. Using the output probabilities of a JIA class as a basis for a joint health score and test it on the follow-up patient recordings. In all 10 of our 6-week follow-up recordings, the score accurately tracked with successful treatment of the condition. Our proposed JAE-based classification model of JIA presents a compelling case for incorporating this novel joint health assessment technique into the clinical work-up and monitoring of JIA.

Keywords: acoustic sensing; juvenile idiopathic arthiritis; machine learning; signal processing; wearable sensors.

Figure 1

Joint acoustic emission overview. (A) A healthy knee articulates smoothly due to its smooth cartilage and appropriate amount/constituency of synovial fluid. This smooth articulation creates a noise-like JAE (blue). In JIA, thickened/inflamed synovium with excessive joint effusions, cartilage loss and/or bone erosions may be observed. These changes are hypothesized to create a JAE with several large spikes (red). (B) To record the knee JAEs, two contact accelerometers were placed on each child's knees. They viewed and replicated the movements in an instructional cartoon during JAE recording such that their movement speed and range of motion was controlled. (C) The resulting JAEs were split into their approximately ten component cycles. Forty-nine features were calculated to describe these cycles. The features, subject numbers, and clinically determined disease status were fit to a feature matrix. (D) Using logistic regression and LOSO-CV, the probability of each cycle belonging to JIA were calculated. The average of those cycle probabilities is used as a “joint health score” to indicate the severity of JIA. If the majority of cycles for a given subject had a probability of JIA ≥ 0.5, that subject was classified as having JIA.

Figure 2

Representative time-domain and spectrogram plots of JAEs from a sample healthy control (A), a subject with active JIA (B), and that same subject after 6-weeks of successful treatment (C). The 12 s of the JAEs represent approximately 4 flexion/extension cycles. Spectrogram of the subject with JIA contains more high power and high frequency components compared to that of healthy and post-treatment subjects.

Figure 3

Assessing the performance of the logistic regression classifier on subjects (A) and cycles (B,C). (A) There was little overlap in the computed joint health score of the healthy control group and the group with JIA. A sub-group from the JIA group after effective treatment had JIA scores heavily overlapping with the healthy control group at follow-up. (B,C) The logistic regression model overall classified the individual cycles accurately 82.7% of the time. The model achieved adequately high sensitivity (84.5%) and specificity (80.4%). HC, healthy control.

Figure 4

Feature importance and model performance based on number of features and cycles. (A) Features are ranked based on their weighted coefficients as output by the trained logistic regression model. The most important feature was the mean spectral spread. (B) The model was trained on a feature set containing just one and up to 20 of the top features and the accuracy was assessed based on including those features and number of cycles recorded from a subject. The colors represent the average accuracy across all subjects for all permutations of cycle selection for a given set of testing parameters. The maximum accuracy of 80.6% is seen in the top right corner when trained on the 20 most important features and tested on all cycles of a given subject.

Figure 5

Longitudinal joint health score tracking. The average joint health score, which describes the probability of having JIA, dropped from 0.84 ± 0.08 to 0.19 ± 0.09 after successful treatment of the condition in 10 subjects. The individual subject scores are denoted by the black squares and dashed lines. The mean and standard deviation of the actively inflamed subjects with JIA is shown in red, and the purple marker indicates the mean and standard deviation at follow-up. This drop in joint health score was statistically significant (p < 0.001).