Use of Machine Learning Classifiers and Sensor Data to Detect Neurological Deficit in Stroke Patients

Abstract

Background: The pronator drift test (PDT), a neurological examination, is widely used in clinics to measure motor weakness of stroke patients.

Objective: The aim of this study was to develop a PDT tool with machine learning classifiers to detect stroke symptoms based on quantification of proximal arm weakness using inertial sensors and signal processing.

Methods: We extracted features of drift and pronation from accelerometer signals of wearable devices on the inner wrists of 16 stroke patients and 10 healthy controls. Signal processing and feature selection approach were applied to discriminate PDT features used to classify stroke patients. A series of machine learning techniques, namely support vector machine (SVM), radial basis function network (RBFN), and random forest (RF), were implemented to discriminate stroke patients from controls with leave-one-out cross-validation.

Results: Signal processing by the PDT tool extracted a total of 12 PDT features from sensors. Feature selection abstracted the major attributes from the 12 PDT features to elucidate the dominant characteristics of proximal weakness of stroke patients using machine learning classification. Our proposed PDT classifiers had an area under the receiver operating characteristic curve (AUC) of .806 (SVM), .769 (RBFN), and .900 (RF) without feature selection, and feature selection improves the AUCs to .913 (SVM), .956 (RBFN), and .975 (RF), representing an average performance enhancement of 15.3%.

Conclusions: Sensors and machine learning methods can reliably detect stroke signs and quantify proximal arm weakness. Our proposed solution will facilitate pervasive monitoring of stroke patients.

Keywords: machine learning; medical informatics; motor; neurological examination; stroke.

Figure 1

Flowchart of pronator drift test (PDT) software.

Figure 2

The pronator drift test: (a) the degree of drift in the weak arm and counter-arm of a patient was measured by the drift angle from the horizontal plane, and (b) the degree of pronation was assessed in front of the patient.

Figure 3

An example of a support vector machine with four support vectors in feature space.

Figure 4

The architecture of an radial basis function network.

Figure 5

A simplified random forest.

Figure 6

Differences of degree in PDT features between stroke patients and controls. Values are mean, standard deviation, and P value.

Figure 7

Plot of support vector machine (SVM) score function and decision by the SVM classifier: (a) positive scores of the SVM classifier for input (above the plane); two control cases were misclassified as patients, and (b) negative scores of the SVM classifier for input (below the plane); one stroke patient case was misclassified as a control case.

Figure 8

Random forest composed of decision trees as a pronator drift test classifier.

Figure 9

Weakness detection using a random forest, including feature selection.

Figure 10

Performance of stroke classifiers excluding/including feature selection.

Figure 11

Receiver operating characteristic (ROC) curve of pronator drift test classifiers and t test of area under the receiver operating characteristic (AUC) (SVM-exFS and SVM-inFS: support vector machine excluding and including feature selection; RBFN-exFS and RBFN-inFS: radial basis function excluding and including feature selection; RF-exFS and RF-inFS: random forest excluding and including feature selection).