Pediatrics Strabismus Discrimination System Based On Video

Infant strabismus is one of the common eye diseases in children.The current mainstream screening method for strabismus is the traditional offline doctor’s diagnosis and treatment.Other solutions are developed around specific fundus image acquisition instruments and basic digital image processing,which are not easy to promote.In addition,the strabismus detection research in complex scenes with occlusion and different illumination is still in a relatively backward stage.At the same time,because the discrimination of strabismus is complex,doctors need to check for many times and observe the trajectory of eye movement before diagnosis,so the accuracy of the examination scheme based on single frame images for strabismus screening is doubtful,,that is,offline diagnosis and treatment is still needed.Plus,infant eye is in a state of dynamic development,coordination ability is poor,offline diagnosis requires a doctor a lot of time to diagnose,however,the contradiction between the extreme shortage of high-quality infant ophthalmologists and the large number of children with eye disease makes medical personnel seek more intelligent solutions that can assist doctors in the diagnosis of pediatric strabismus.Therefore,a solution that can break through the limitations of professional strabismus examination instruments,untie the constraints of offline diagnosis and treatment,reduce the burden on doctors,and achieve the purpose of simple strabismus screening for infants and young children in early stage has great research value and practical value.Firstly,a network model for feature extraction and pupil center coordinate regression was constructed in this study.Inspired by the target detection algorithm and the key point location algorithm,the model inferred the pupil center position by the information around the eye socket and computed the accurate pupil center coordinate.It solved the problems of poor robustness and inaccurate positioning of the pupil center model under the complex conditions and provided the key features for the next strabismus discrimination and strabismus classification.Furthermore,a cascading oneclass support vector machine(SVM)was constructed as a model for strabismus determination and strabismus classification.The model constructed the feature vector based on the pupil center coordinate obtained from the pupil center location model and designed the rigorous strabismus discrimination and strabismus category classification process which effectively solved the problems of unbalanced data and outliers in multiple classification problems,such as classifier instability and inaccurate classification.In the meantime,it empowered interpretability to the classification results.Combining the pupil positioning and the strabismus classification models,an integral strabismus discrimination and classification algorithm was constructed.It effectively solved the problems of pupil center positioning in complex scenes and the problem of squint determination under the condition of unbalanced data and interference from outliers in the solution.Next,this study proposes a solution that uses a portable,networkable mobile terminal to shoot a video of a child with strabismus turning their eyes,and then combines the resulting pupil positioning model and classification model for feature extraction and squint detection to construct an easy-to-use visualization application system.This system can be effectively applied to the judgment of strabismus in complex scenes,and plays an important role in assisting doctors in medical diagnosis.