Automatic Measurement of Postural Abnormalities With a Pose Estimation Algorithm in Parkinson's Disease

Abstract

Objective: This study aims to develop an automated and objective tool to evaluate postural abnormalities in Parkinson's disease (PD) patients.

Methods: We applied a deep learning-based pose-estimation algorithm to lateral photos of prospectively enrolled PD patients (n = 28). We automatically measured the anterior flexion angle (AFA) and dropped head angle (DHA), which were validated with conventional manual labeling methods.

Results: The automatically measured DHA and AFA were in excellent agreement with manual labeling methods (intraclass correlation coefficient > 0.95) with mean bias equal to or less than 3 degrees.

Conclusion: The deep learning-based pose-estimation algorithm objectively measured postural abnormalities in PD patients.

Keywords: Camptocormia; Parkinson’s disease; Pose estimation.

Figure 1.

Automatic and manual measurements of the anterior flexion angle (AFA) and dropped head angle (DHA). A: Lateral photos of the patients standing in front of a wall with a vertical reference within the frame. B: The 2D positions of the left ear, acromion, great trochanter and ankle were automatically labeled (labeled with red dots and connected with red line). C: The DHA and AFA were calculated based on the position of the ear, acromion, greater trochanter (labeled with red dots and connected with red line) and vertical line (green line). The calculated angle was noted as orange double-headed arrows. D: The same photograph from Figure 1A. The locations of the acromion and the hip joint were marked with detachable stickers (annotated with red circle). E: The acromion and greater trochanter were manually labeled based on the position of the detachable sticker (labeled with light blue dots and lines). The location of the left acoustic canal was also manually labeled (light blue dots and lines). F: The DHA and AFA were calculated based on the position of the ear, acromion, greater trochanter (labeled with light blue dots and lines) and vertical line (green line). The calculated angle was noted as orange double-headed arrows. G: The same photograph from Figure 1A shows a vertical reference (light blue arrow). H: Binary transformed image with edge detection algorithm (presented at Materials & Methods). Automatically detected vertical reference (blue line) and degree of the slope are shown. I: The AFA (annotated with orange double-headed arrow) calculated with the adjusted vertical reference (shown in blue line). The vertical line of the photo is shown as green line. The AFA was calculated based on the position of the acromion, greater trochanter (labeled with red dots and lines) and vertical reference line (the blue line moved from the original location annotated with a black double-headed arrow).

Figure 2.

Scatter plot and Bland-Altman plot of the anterior flexion angle (AFA) and dropped head angle (DHA). A: Scatter plot showing the mean AFA (degree) of manual (x-axis) and automatic (y-axis) measurements. B: The X-axis in the figure denotes the mean value of the two measurements (manual vs. automatic), and the y-axis in the figure (Δ) is the difference between the two measurements of the mean AFA. The mean (plain line) and 1.96 standard deviations of the bias (dotted line) are denoted as horizontal lines. C, D: Scatter plot (C) and Bland-Altman plot (D) for DHA. SD, standard deviation.