Research And Implementation Of Aided Diagnosis Of Keratitis Image Based On Machine Learning

Keratitis is a common eye disease and it is one of the major blindness diseases in China.The wearing of contact lenses and the popularity of eye surgery have greatly increased the prevalence of keratitis.If patients with keratitis do not receive timely treatment or surgery,their vision will surely be affected.In severe cases,the cornea will even suffer irreversible damage.This will eventually lead to blindness,which will greatly affect people's lives.In recent years,with the continuous development of the field of machine learning,computer-aided diagnosis has become a new type of medical model,which provides a new direction for the improvement of the medical status of keratitis.Slit lamp examination is a common and important examination in ophthalmology clinic.It provides doctors with a visual representation of the true corneal shape of the patient,provides a diagnostic basis,and also provides data resources for computer-assisted diagnosis of keratitis.These make it possible to apply computer technology to the diagnosis of keratitis.This paper,based on the method of machine learning,studies on Automatic Diagnosis of Keratitis in Eye Slit Lamp Image collected by the Zhong Shan Ophthalimic Center.It mainly contains two parts,keratitis image classification and classification work.In the classification of keratitis images,this paper firstly adopts the traditional machine learning method,and combines the LBP and SIFT feature extraction methods with SVM,RBF,Random Forest,and k NN classification algorithms to construct eight types of keratitis classification.And it analyzes and compares these classifiers.Then,a convolutional neural network with excellent performance in the image classification field is adopted.The transfer learning method is used.Based on the two pre-training models of Inception v4 and residual network Res Net,the keratitis image classification models are trained and the two models are compared and analyzed.In the classification model training process,this article uses a variety of parameter optimization methods to improve the model's generalization ability,so that the accuracy rate is greatly improved.In the grading of keratitis images,this article uses a convolutional neural network to train the keratitis image grading models on the two attributes,the stage of keratitis and is the edge clear,and evaluates the severity of keratitis,and compares the hierarchical model.In the process of grading model training,this paper uses data enhancement and cost-sensitive methods to reduce the impact of unbalanced data sets,greatly reducing the missed diagnosis rate and improving the accuracy of the model.In the classification of keratitis images,the accuracy rate of convolutional neural network reached 92.24%,exceeding the maximum accuracy of traditional machine learning methods by 81.20%.This shows that convolutional neural network is more accurate than traditional machine learning methods in keratitis image classification.In the keratitis image grading,the accuracy of the network models based on Inception v4 was 89.38% and 86.73%,and the recall rate reached 92.59% and 93.55%,respectively,which greatly reduced the missed diagnosis rate.As a whole,the classification and grading results of the convolutional neural network model studied in this paper can be used as reference for ophthalmologists to diagnose keratitis.