An Expert System for Quantification of Bradykinesia Based on Wearable Inertial Sensors

Abstract

Wearable sensors and advanced algorithms can provide significant decision support for clinical practice. Currently, the motor symptoms of patients with neurological disorders are often visually observed and evaluated, which may result in rough and subjective quantification. Using small inertial wearable sensors, fine repetitive and clinically important movements can be captured and objectively evaluated. In this paper, a new methodology is designed for objective evaluation and automatic scoring of bradykinesia in repetitive finger-tapping movements for patients with idiopathic Parkinson's disease and atypical parkinsonism. The methodology comprises several simple and repeatable signal-processing techniques that are applied for the extraction of important movement features. The decision support system consists of simple rules designed to match universally defined criteria that are evaluated in clinical practice. The accuracy of the system is calculated based on the reference scores provided by two neurologists. The proposed expert system achieved an accuracy of 88.16% for files on which neurologists agreed with their scores. The introduced system is simple, repeatable, easy to implement, and can provide good assistance in clinical practice, providing a detailed analysis of finger-tapping performance and decision support for symptom evaluation.

Keywords: Parkinson’s disease; UPDRS; atypical parkinsonism; automatic scoring; decision support system; finger-tapping; wearable inertial sensors.

Figure 1

Illustration of the inertial sensor system, with local coordinate systems of the thumb $\left(X_1, Y_1, Z_1\right)$ and index finger $\left(X_2, Y_2, Z_2\right)$ sensors. SCU-sensor-control unit.

Figure 2

Block diagram of the expert system for UPDRS finger-tapping score calculation.

Figure 3

An example of the normalized dominant component of the relative angular velocity ${\omega }_{\mathrm{rd}}$ for one MSA patient (ID: MSA11) with extracted markers.

Figure 4

Upper panel: Angle estimation. The dashed grey line marks the drifted angle sequence, and the solid black line corresponds to the angle sequence after drift removal. Red crosses show “zero posture” markers, and the dotted red line presents the polynomial fit used for drift removal. Lower panel: Angle amplitude decrement. The solid grey line shows the angle sequence, whereas black circles mark the angle amplitudes (highest finger apertures) per tap. The dashed red line presents the threshold $T{H}_{\mathsf{\alpha }}$ used for the detection of decreased amplitudes. The example is given for one MSA patient (ID: MSA11).

Figure 5

Calculation of hesitations and freezes: angular velocity ${\omega }_{\mathrm{rd}}$ (upper panel) and calculated $CS{A}_{\mathrm{T}}$ characteristic (bottom panel). The solid grey horizontal line marks the average $CS{A}_{\mathrm{T}}$ value. The dashed grey horizontal line corresponds to the upper threshold $T{H}_{50}=50\%$ of the $CS{A}_{\mathrm{T}}$ average value. The dotted black horizontal line shows the lower threshold $T{H}_{25}=25\%$ of the $CS{A}_{\mathrm{T}}$_- average value. Similarly, dotted grey vertical lines show areas that are classified as hesitations (an “H” mark) and freezes (an “F” mark). The example is given for one PSP patient (ID: PSP14).

Figure 6

Calculation of the frequency characteristic: scalogram of the obtained continuous wavelet transform (CWT) coefficients. The dashed black line marks the i-th sample. The CWT coefficients at the i-th sample are presented in the smaller upper panel. The red dashed line in the upper panel marks the frequency with the highest amplitude of the CWT coefficients for the i-th sample (referred to as ${f_i}^{\left(i\right)}$). The example is given for one PSP patient (ID: PSP14).

Figure 7

The block scheme of the decision support system. The system is divided into four processing blocks (bordered with dashed black rectangles). The inputs to the blocks are the calculated features: ${\alpha }_{av}$, $f_{av}^{\left(i\right)}{i}_{dec}$, and $H_{num},F_{num},$ respectively. Each block implements rules and assigns a subscore for the input feature. The final score $S_{FT}$ is decided based on the results obtained from all four blocks.

Figure 8

Dependency of: (a) calculated scores and the ${\alpha }_{av}$ feature; (b) calculated scores the and $f_{av}^{\left(i\right)}$ feature. Grey circles mark samples that are assigned to the $C_1$ cluster (“wider and slower” performance), whereas black crosses correspond to members of the cluster $C_2$ (“narrower and faster” performance).

Figure 9

Presentation of results using the confusion matrix. (a) Case I—The result obtained when all recordings are included. (b) Case II—The result obtained using only recordings on which both raters agreed. The cells on the diagonal of the confusion matrix show the overall success rate for each score (expressed as a percentage (%)), whereas the cells outside the diagonal show the error rate for the scores (expressed as a percentage (%)).

Figure 10

The result of the expert system comprising a graphical representation with detected irregularities, calculated features, and the final score. The example is given for one MSA patient (ID: MSA11), right hand, and one PSP patient (ID: PSP14), right hand.