COMPUTERIZED SAMPLING AND DEVELOPMENTAL ASSESSMENT OF INFANT SPONTANOUS MOTOR PATTERNS
1Friedman Hagit PhD, 2Bar-Yosef Omer PhD MD, 3Gordon Goren PhD MBA, 3Forkosh Oren PhD, 3Schneidman Elad, PhD
1Department of nursing, Faculty of Social Welfare & Health Sciences, Haifa University, 2Pediatric Neurology Institute, Edmond and Lily Safra Children’s Hospital, Chaim Sheba Medical Center affiliated to Tel Aviv University, Ramat Gan, Israel . 3Department of Neurobiology, Weizmann Institute of Science, Rehovot, Israel,
Neonates require about 6 month acquiring independent complex motor behavior, such as reaching. Prior to the development of complex motor skills infants motor repertoire comprises of reflexes and spontaneous movements.
Spontaneous movements are internally driven. They are present during most of infant's awake time, consisting of sequential movements, involving different body parts, with only a rough pattern. It was suggested that these movements are the building block of future complex movements.
Comprehension of the development of early movements would expand the understanding of the transition from spontaneous to goal directed movements. This data could also serve as a tool for early identification of impaired motor development.
Previous characterization of spontaneous movements denoted a change of pattern involving kinematics of each limb and interaction between the limbs.
Previous methods used to analyze spontaneous movements included qualitative analysis of video recordings, motion capture apparatus or accelerometers.
These methods had limitations such as lack of quantitative data, use of complex devices and direct contact with the infant
The approach presented here, is a novel method for motion capture based on a combination of 3D video recording by Kinect® and a tracking algorithm. It is a non interventional method and does not require the attachment of markers to the infants body.
This method is currently designed to capture movements of the extremities of the 4 limbs and head.
Future plan is to serially analyze kinematics of movements of infants from the age of 2-24 weeks. Quantifying changes in intra and inter limb movement patterns and their association to head movements.
We used Kinect® sensor, Prime Sense.
Figure 1. Kinect sensor
An in-house driver and software were written, using Microsoft Kinect SDK. We recorded both color images (Fig. 2(a)) and depth field (Fig. 2(b)). Both spatial and temporal synchronization was performed. Due to filming angles, a depth-rectification was also implemented (Fig. 2(c)).
Figure 2. Kinect recording system. (a) Color image. (b) Original depth field. (c) Rectified depth field.
An in-house semi-automatic tracking algorithm was written in Matlab. The user marked the required end-points, left/right arms, left/right legs and nose. Marking were performed in 10 automatically identified relevant frames.
For each end-point a patch of size 20x20 was taken (Fig. 3(a, b))
In each consecutive frame the most correlated patch in the environment was found in the color image (Fig. 3(c))
The highest point in the depth field in the surrounding environment was chosen, signifying the hands, feet and nose. (Fig.3 (d))
Forward and backward tracking was performed to optimize the utilization of the user-marked data.
Figure 3. Tracking algorithm. (a) Color image with the right-arm patch marked in a box. (b) The right-arm 20x20 RGB pixel patch. (c) Difference field between the patch and the surrounding area, lowest difference marked in ‘x’. (d) Depth field surrounding the found location in (c); highest point (shortest depth from the camera) marked in ‘x’.
After obtaining the 3D time-series of each limb’s end-point, we calculated several parameters:
Average tangential velocity of each limb’s end-point, smoothed with a 5Hz filter.
Movement Units (MU): defined as epochs which started with increasing velocity crossing the 0.15 m/sec threshold ad ended with decreasing velocity crossing the same threshold.
Pearson correlation coefficient between limbs: we calculated the correlation coefficient between velocity time-series of pairs of limbs.
Figure 4. Tracking results. (a) 2D projections of the limbs traces. (b) Tangential velocity of the four limbs as a function of time for a 30 sec episode. Movement units are marked by yellow. (c) Movement unit durations.
Correlation of limb’s MU (pearson correlation coefficient)
RA-LA 0.42 RA-RL 0.27 RA-LL 0.26
RL-LL 0.69 LA-LL 0.41 LA-RL 0.51
The method presented uses a simple device with no body markers that enable an accurate analysis of movements kinematics
Based on the stable results by this method, we plan to start a research plan of serial recording of movements along infant’s first 6 month of life