Moti tion Analysis to Quanti tita tati tively Assess th the Le - PowerPoint PPT Presentation

Moti tion Analysis to Quanti tita tati tively Assess th the Le Level of of Pain in Anima mals Annabelle Eyler Hood College February 1, 2020 Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 1

2019 SURIEM Program at Michigan State University Program Director: Dr. Robert Bell, Michigan State University Project Investigator: Dr. Mark Reimers, Michigan State University Research Members: Storm Chin, Morehouse College Joyce Quon, California State University, Los Angeles Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 2

Assessment of Pain • Different forms of pain in animals, like humans • Currently lacking descriptors of pain • Distinguishing different qualities of pain Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 3

Pain Through Fluidity of Movement • Correlation between pain and movement characteristics • Evaluate pain by help of motion patterns and fluidity • Determine defining characteristics to quantify pain Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 4

How Motion was Tracked • Computer vision software DeepLabCut • Deep convolutional neural network • Requires annotations and training time § Build and train top two layers of network Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 5

Our Variables • Control and injured rats • Injured rats were denervated on front right paw § Phantom limb pain • Different tests § Open Field § Novel Object Recognition Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 6

Accuracy Problems Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 7

Labeling Problems • Baseline errors • Human error • Social definition difference between labelers • Set standards lacking details • DeepLabCut provided trajectories • Mislabeling • Difference in pixels per feature Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 8

Frame Error Sample Calculation • Area occupied by rat: (30,95) • x-axis SE: 1.78 pixels, y-axis SE: 2.30 pixel Using the Pythagorean Theorem: distance 2 = 2.30 2 + 1.78 2 distance = 2.91 pixels Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 9

Distance From Mean For this specific frame example, the distance from the mean is 2.91 pixels. Overall average distance was 3 pixels from mean, which is a 4.31% pixel error Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 10

Processing DeepLabCut Predictions Nose Left Ear Right Ear Rump Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 11

Processing Raw Data • Problem: Inaccurate predictions from DeepLabCut • Possible Solutions § Remove incorrect frames § Check confidence of each frame § Data smoothing Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 12

What is a Moving Average? • A series of averages of different subsets of the data frame • Varies depending on window size Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 13

Before Interpolation After Interpolation Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 14

Motion Analysis Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 15

Total Distance Traveled • Used DeepLabCut output csv file • Created code to calculate Euclidean distance of each feature between two frames • Executed for all frames of a video, added up for total distance Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 16

Velocity, Acceleration, and Jerk • Each calculate the difference between previous calculation for all frames in video • Velocity vector stores the difference between two given distances of one feature • Jerk (derivative of acceleration) measures change in acceleration through video, frame to frame Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 17

Novel Object Recognition Code • Creates a radius using annotated corners • Flag frame if nose label enters any radius • Determines number of flagged frames, converts to seconds • Returns time that rat was near any object Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 18

Results Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 19

Distance by Feature Control Injured, no treatment Distances: Nose Left Ear Right Ear Rump Trends: Nose vs. rump and left vs. right ear Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 20

Nose Distance vs. Rump Distance Trends: • Control- move body less than nose • Injured- higher ratio of movement of nose to rump compared to control Generally, the control rats have the similar ratio of nose to rump movement Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 21

Future Work Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 22

• Rats reacting to visual stimuli • Pupil diameter with body motion • Recording brain activity with brain activity Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 23

Brain, Body, and Pupil Activity • Similar peaks seen • Attempt to find correlation Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 24

Future Work • Application of code created, ability to apply to mainstream motion analysis § Accessible to other computer vision software programs • Data smoothing to study humans • Generalize to determine level of engagement in humans Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 25

References • A. Mathis, P. Mamidanna, K. M. Cury, T. Abe, V. N. Murthy, M. W. Mathis, and M. Bethge. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nature Neuroscience , 21:1281-1289, 2018. • T. Nath, A. Mathis, A. C. Chen, A. Patel, M. Bethge, and M. W. Mathis. Using DeepLabCut for 3D markerless pose estimation across species. Nature Protocols , 14: 2152-2176, 2019. • S. Piana, P. Alborno, R. Niewiadomski, M. Mancini, G. Volpe, and A. Camurri. Movement Fluidity Analysis Based on Performance and Perception. CHI Extended Abstracts '16 , 1629-1636, 2016. Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 26

Acknowledgements I would like to thank Michigan State University and their 2019 Summer Undergraduate Research Institute in Experimental Mathematics (SURIEM) Program for hosting me for this research experience, which was financially supported by the National Science Foundation Award #1852066 and the National Security Agency Grant #H98230-1-0014. I would also like to thank Hood College, my home institution for the support in being here. Eyler Motion Analysis to Quantitatively Assess the Level of Pain in Animals 27