Study On Gaze Estimation And Disease Classification Model Based On Eye Movement Information

Eye movement information refers to some characteristics of eye movement when the subject obtains visual information from the outside world.Studying eye movement information is an effective means to study visual processing,and its research results can be applied to the analysis and diagnosis of various diseases.It can be seen from the relevant literature in recent years that the research on eye movement in the medical field has expanded from the earliest diagnosis of vestibular dysfunction and vertigo to the research on alzheimer’s disease,Parkinson’s disease,autism and other mental disorders.However,the research on eye movement information at home and abroad is still very weak.On the one hand,the research mainly focuses on a single disease without comprehensive analysis.On the other hand,traditional methods are often used in information processing,which is not conducive to the in-depth exploration of the hidden characteristics of eye movement information and limits its reliability and universality in medical pathological analysis of eye movement information.In recent years,the development and innovative application of artificial intelligence technologies have brought great changes to many fields.AI’s powerful information processing and feature extraction capabilities provide good conditions for medical information processing.In this paper,the advanced deep learning technology is applied to gaze estimation and disease diagnosis based on eye movement information.The main work is as follows(1)In order to obtain eye movement information related to diseases,six eye movement tests were designed and experimental software was developed on the basis of summarizing two classical eye movement test paradigms.Then minimum area bounding rectangle based on the convex hull algorithm was used to extracted 9 eye movement characteristics of medical significance form eye image.In order to detect the blink eye images,we designed a blinking detect model according on the bounding rectangle of pupil,the blinking images can be accurately detected from the eye-movement video.Finally,combining the characteristics of eye movement and deep learning algorithm,we find a knowledge expression method of eye movement(2)Gaze estimation based on eye image is realized by two methods.The first one method is polynomial fitting。This method establishes a polynomial between the eye map reference frame and the scene reference frame firstly.Then the pupil position information and fixation point position information were obtained through the 13-points calibration method.Finally,the least square method is used to solve the parameters of the polynomial。The second one is based on deep learning method.Firstly,a regression model of gaze estimation is designed by using convolutional neural network.Then the model learns the mapping relationship between eye images and fixation position through training model.In the training process,this paper makes the model perform better on a small number of training samples by means of data enhancement and change of learning strategies.Test results shows that the regression model of gaze estimation can predict the fixation position accurately(3)Neural network multi-classification model was used to classify the patients’ eye movement information,and a disease diagnosis model based on eye movement information and deep learning was constructed.Firstly,FC,LSTM and C-LSTM multi-classification models was designed.Then the eye movement information of 24 subjects from The Second Affiliated Hospital of Chongqing Medical University,It can be seen from the test results that the classification accuracy of the three multi-classification models on the test samples can reach over 75%.It is proved that this method is suitable for the analysis and diagnosis of disease.