Intelligent Prediction Of Infant Cognitive Development Based On General Movements

Cognition is an essential human function,and its development in infancy is crucial.Traditionally,pediatricians have used clinical observation or medical imaging to assess infants’ current cognitive development（CD）status.The object of pediatricians’ greater concern is however their future outcomes,because high-risk infants can be identified early in life for intervention.This opportunity has however not yet been realized.Fortunately,some recent studies have shown that the general movement（GM）performance of infants around 3-4 months after birth might reflect their future CD status,which gives us an opportunity to achieve this goal by artificial intelligence.Therefore,on the basis of these studies,combined with artificial intelligence,this thesis realizes the intelligent prediction of risk of infant CD disorder based on GM.The data eventually collected a total of 44 infants,including their GM videos at 3 to4 months of age,and their assessment scores for CD status at about 1 year of age,using the cognitive assessment section of the Bayley Infant Development Scale Second Edition.OpenPose model is well known in the field of human pose estimation.However,due to trunk-to-limb ratio differences between adults and infants,direct use of OpenPose for infants may lead to inaccurate estimates.In this thesis,we retrain the OpenPose model using our annotated infant pose dataset.The retrained OpenPose model was 4.4% more accurate than the original OpenPose model in predicting infant posture information（with PCKh@0.50 as the evaluation indicator）,which was helpful to extract more accurate infant posture information.In this thesis,OpenPose model for babies was used to extract joint position information of infants,and bilateral movement symmetry（BMS）features were extracted from the position information of symmetrical bilateral joints.The infants were then divided into low-risk and high-risk CD disorder groups based on their cognitive scores at age 1.Finally,with the grouping results of infants’ cognitive scores as the gold standard,the extracted BMS features were put into 8 different classifiers for dichotomous prediction.In the results,the area under the curve,recall and precision values reached 0.861,0.842,and 0.842 for two-group classification,respectively.It is possible to automatically predict future CD risk in infants using their GM videos recorded at the early stage of life.In addition,this study not only helps clinicians better understand infant CD mechanisms,but also provides an economical,portable and non-invasive way to screen infants at high-risk early to facilitate their recovery.In addition,this thesis has confirmed from an entirely new perspective that the developmental domains of infants,such as their motor and cognitive functions,are not independent of each other,but closely interrelated.