Automatic and Efficient Fall Risk Assessment Based on Machine Learning

Abstract

Automating fall risk assessment, in an efficient, non-invasive manner, specifically in the elderly population, serves as an efficient means for implementing wide screening of individuals for fall risk and determining their need for participation in fall prevention programs. We present an automated and efficient system for fall risk assessment based on a multi-depth camera human motion tracking system, which captures patients performing the well-known and validated Berg Balance Scale (BBS). Trained machine learning classifiers predict the patient's 14 scores of the BBS by extracting spatio-temporal features from the captured human motion records. Additionally, we used machine learning tools to develop fall risk predictors that enable reducing the number of BBS tasks required to assess fall risk, from 14 to 4-6 tasks, without compromising the quality and accuracy of the BBS assessment. The reduced battery, termed Efficient-BBS (E-BBS), can be performed by physiotherapists in a traditional setting or deployed using our automated system, allowing an efficient and effective BBS evaluation. We report on a pilot study, run in a major hospital, including accuracy and statistical evaluations. We show the accuracy and confidence levels of the E-BBS, as well as the average number of BBS tasks required to reach the accuracy thresholds. The trained E-BBS system was shown to reduce the number of tasks in the BBS test by approximately 50% while maintaining 97% accuracy. The presented approach enables a wide screening of individuals for fall risk in a manner that does not require significant time or resources from the medical community. Furthermore, the technology and machine learning algorithms can be implemented on other batteries of medical tests and evaluations.

Keywords: Berg Balance Scale; balance; diagnosis; elderly; fall risk detection; human tracking; telemedicine.

Figure 1

Schematic diagram of the BBS score and fall risk prediction system.

Figure 2

The 3D sensor (left) measures the distances of points in the scene from which a skeleton representation of the body pose is produced (right).

Figure 3

The multi-camera tracking system setup includes two depth sensors allowing the capture of the full range of patient motion, as well as enabling data merging to reduce noise and skeleton errors.

Figure 4

Spatio-temporal features are computed from the skeleton data in each recorded video frame.

Figure 5

Confusion matrix between the true risk of fall as determined by the physiotherapists and the predicted risk of fall (left). False negatives can be reduced by manipulating the thresholds (right). The MSE values are 0.25 and 0.29, respectively.

Figure 6

Schematic diagram of the E-BBS fall risk prediction system with efficient and adaptive ordering of the BBS tasks.

Figure 7

Accuracy vs. average number of BBS tests for different selector methods (Section 5.2) trained on the physiotherapist scoring. For each method, the plot shows values for 6 different confidence thresholds (90, 92, 94, 96, 98, and 100).

Figure 8

Accuracy vs. average number of BBS tests for different selector methods (Section 5.2) trained on the automatic BBS scoring. For each method, the plot shows values for 6 different confidence thresholds (90, 92, 94, 96, 98, and 100).

Figure 9

Accuracy vs. average number of BBS tests for the different initial subset of tasks. Results are shown for Selector Method 3 and training on the physiotherapists’ data. For each initial subset of the tasks, the plot shows values for 6 different confidence thresholds (90, 92, 94, 96, 98, and 100).

Figure 10

Occurrence matrices depicting the ordering of BBS tasks in the E-BBS. Columns indicate the order in the E-BBS sequence. Each row indicates a standard BBS task as defined in [7]. The matrix entry value indicates the proportion of times a BBS task was used in a certain E-BBS sequence position across the test set. (a–d) Occurrence matrices of E-BBS sequences as trained on the physiotherapist data and using the 4 Selector Methods 1 to 4, respectively.

Figure 11

(a–d) same as Figure 10, but trained on the automatic BBS scoring data.