Identification of fidgety movements and prediction of CP by the use of computer-based video analysis is more accurate when based on two video recordings

Abstract

This study evaluates the role of postterm age at assessment and the use of one or two video recordings for the detection of fidgety movements (FMs) and prediction of cerebral palsy (CP) using computer vision software. Recordings between 9 and 17 weeks postterm age from 52 preterm and term infants (24 boys, 28 girls; 26 born preterm) were used. Recordings were analyzed using computer vision software. Movement variables, derived from differences between subsequent video frames, were used for quantitative analysis. Sensitivities, specificities, and area under curve were estimated for the first and second recording, or a mean of both. FMs were classified based on the Prechtl approach of general movement assessment. CP status was reported at 2 years. Nine children developed CP of whom all recordings had absent FMs. The mean variability of the centroid of motion (C SD ) from two recordings was more accurate than using only one recording, and identified all children who were diagnosed with CP at 2 years. Age at assessment did not influence the detection of FMs or prediction of CP. The accuracy of computer vision techniques in identifying FMs and predicting CP based on two recordings should be confirmed in future studies.

Early prediction of typical outcome and mild developmental delay for prioritisation of service delivery for very preterm and very low birthweight infants: a study protocol

“…Importantly, it meets the requirement for assessment of fidgety movement on at least two occasions to definitively classify GMs as absent fidgety. 67 …”

Technology-Aided Assessment of Sensorimotor Function in Early Infancy

“…Of the previously described clinical assessment tools, by nature of its application and off-line scoring, perhaps the most amenable to combination with a technology-based solution for infants is GM assessment. Indeed, studies have shown that the predictive power of GM assessment is preserved even when the assessment is automated by fitting motion tracking sensors to the extremities of the infants (in addition to video recording the infant), and then performing motion feature extraction on a highly filtered motion image using custom computer software ( 58 – 60 ). The sensitivity and specificity of GM assessment for the prediction of cerebral palsy has also been found to be greatly enhanced by combination with MRI at term equivalent age ( 61 ).…”

Computer vision to automatically assess infant neuromotor risk

“…In the last decade, scientists have developed systems for video-based assessments using optical flow which hold promise to be widely accessible as they can be implemented on mobile devices (Adde, Helbostad, Jensenius, Langaas, & Støen, 2013;Stahl et al, 2012;Støen et al, 2017). Such methods typically use frame differencing of a video to estimate movements by tracking the centroid of motion.…”

“…This technique can be extended for the measurement of movement of each limb (Stahl et al, 2012). Using this approach, the amount of movement and the frequency of movement have corresponded to clinical evaluations (Adde et al, 2013(Adde et al, , 2010. However, frame difference metrics from optic flow rely on centroid estimates that only measure gross movements as opposed to measuring the kinematic variables for individual joint or limb segment movements.…”