An Auxiliary Diagnostic System For Keratoconus Grading Based On Machine Learning

Keratoconus is one of the diseases that affect the health of the eyes.This disease can lead to the impairment of vision or the development of high myopia in the eyes of patients.The incidence rate of this disease is high among the refractive surgery population.China is a country with a large incidence rate of ametropia,and the demand for refractive surgery is huge.Keratoconus screening is an important examination before refractive surgery,and keratoconus is a contraindication for refractive surgery.Therefore,how to improve the accuracy of diagnosis of keratoconus is particularly important for refractive surgery in patients with refractive.Currently,the clinical diagnosis of keratoconus is usually manually by clinical ophthalmologists on the corneal topography of patients.This is not only time-consuming and laborious,but also the professional skills and experience of doctors are crucial for the accuracy of diagnosis.Therefore,a reliable computer-aided screening method for diagnosis of keratoconus is urgently needed.In response to the above issues,this paper proposes machine learning technology combined with corneal topography to screen for keratoconus and build auxiliary diagnostic system to solve the time-consuming and laborious problem of clinical artificial diagnosis of keratoconus.In this study,we first collected and annotated corneal topographic dataset,then conducted relevant experimental studies,and evaluated the effectiveness of our research methods.Finally,auxiliary diagnosis system for keratoconus was constructed.The main research contributions of this paper are as follows:(1)A model of residual neural network combined with corneal topography for screening keratoconus tasks is proposed.The model uses Res Net50 as the backbone network.The corneal topographic dataset was divided into two categories: normal cornea and keratoconus,and experimental research was conducted.The accuracy,sensitivity,and specificity of the Res Net50 model on the corneal topographic map dataset were 95.00%,92.45%,and 97.87%,according to the model performance evaluation index evaluation model.(2)A network model based on attention mechanism combined with corneal topography for screening keratoconus tasks was proposed.By integrating a convolutional attention module in a residual network,the specific method is to integrate a Convolutional Block Attention Module(CBAM)into a Res Net50 residual network to build a CBAM＿Res Net50 network model,in which the CBAM convolutional attention module can adaptively select important channel or spatial features based on the situation of channel attention and spatial attention.After that,relevant experimental research was conducted,and CBAM＿Res Net50 model was evaluated to obtain a classification accuracy of 98.00% and an AUC of 0.97 on the validated dataset of corneal topography.Compared with other models,its various performance indicators achieved the best.(3)Based on the previous research,auxiliary diagnostic system or keratoconus grading based on Py Qt5 was designed and implemented.The system is built using a cross platform GUI programming tool library Py Qt5.The system is divided into two modules.The first module is a training module,which is used to train the corneal topographic map dataset and obtain a training model;The second module is a prediction module,which predicts the input corneal topography.