Quantifying Parkinson's disease motor severity under uncertainty using MDS-UPDRS videos

Abstract

Parkinson's disease (PD) is a brain disorder that primarily affects motor function, leading to slow movement, tremor, and stiffness, as well as postural instability and difficulty with walking/balance. The severity of PD motor impairments is clinically assessed by part III of the Movement Disorder Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS), a universally-accepted rating scale. However, experts often disagree on the exact scoring of individuals. In the presence of label noise, training a machine learning model using only scores from a single rater may introduce bias, while training models with multiple noisy ratings is a challenging task due to the inter-rater variabilities. In this paper, we introduce an ordinal focal neural network to estimate the MDS-UPDRS scores from input videos, to leverage the ordinal nature of MDS-UPDRS scores and combat class imbalance. To handle multiple noisy labels per exam, the training of the network is regularized via rater confusion estimation (RCE), which encodes the rating habits and skills of raters via a confusion matrix. We apply our pipeline to estimate MDS-UPDRS test scores from their video recordings including gait (with multiple Raters, R=3) and finger tapping scores (single rater). On a sizable clinical dataset for the gait test (N=55), we obtained a classification accuracy of 72% with majority vote as ground-truth, and an accuracy of ∼84% of our model predicting at least one of the raters' scores. Our work demonstrates how computer-assisted technologies can be used to track patients and their motor impairments, even when there is uncertainty in the clinical ratings. The latest version of the code will be available at https://github.com/mlu355/PD-Motor-Severity-Estimation.

Keywords: Computer vision; Finger tapping; Gait analysis; Movement disorder society Unified Parkinsons Disease Rating Scale; Uncertainty.

Conflict of interest statement

Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Copyright © 2021. Published by Elsevier B.V.

Figures

Figure 1:

Progressive PD impairments demonstrated by 3D gait (poses fade over time; left/right distinguished by color) with MDS-UPDRS gait score shown below each skeleton. Participants are taken from our clinical dataset. Scores 0 to 2 progressively decrease in mobility with reduced arm swing and range of pedal motion (i.e.reduced stride amplitude and footlift) while 3 becomes imbalanced with almost frozen gait. The average movements are the outputs of our proposed pipeline. The top four joints which contributed to gait score estimation are indicated.

Figure 2:

Gait Scores of Raters A, B, C, A* and their Majority Vote (M).

Figure 3:

The proposed framework: we first extract the identified participants’ 3D body mesh and subsequently the skeletons via VIBE. Based on this 3D skeleton sequence, our proposed OF-DDNet estimates the MDS-UPDRS gait score.

Figure 4:

The OF-DDNet training pipeline with Rater Confusion Estimation (RCE). We implemented this scheme for the gait MDS-UPDRS score estimation with 3 raters. Given a series of input skeletons, OF-DDNet generates prediction pi for the input. This prediction is multiplied by each rater CM A(r) to produce a prediction for each rater. The total loss is the sum of the individual losses between each pair of the rth rater.

Figure 5:

Per-class MDS-UPDRS gait score estimation performance of our method.

Figure 6:

Confusion matrix of our final model for estimation MDS-UPDRS gait scores.

Figure 7:

MDS-UPDRS gait score as a function of study participants. Orange dots indicate predicted scores by our method. Possible rating refers to all ratings the specific participant has received from the three raters. The participants are sorted in an increasing average rating order.

Figure 8:

Participants with ground-truth MDS-UPDRS scores of 1 plotted with their predicted score by our method and their age. The orange dots show the age of each participant. ANOVA test returned a p-value=0.22, confirming insignificant difference between the 4 groups concerning the participants’ age.

Figure 9:

Saliency for the same participants as in Fig.1 visualized on the input joints, measured as normalized gradient update. Saliency is highest at the ankles, heels, and toes, with values sometimes high at the arms and knees.

Figure 10:

Visualization of the confusion matrices of the learned CM estimates for each rater (Raters A, B, C) averaged over all folds.

Figure 11:

Per-class MDS-UPDRS finger tapping score estimation performance of our method.

Figure 12:

Saliency measured as normalized gradient update for the finger-tapping test on the input joints of four subjects with MDS-UPDRS finger tapping score shown. Thumb joints are connected with red edges. Saliency is highest at the thumb and index finger.