Research On Aided Diagnosis Of Diabetic Kidney Disease Based On Deep Learning

The 10 th edition of the Diabetes Map of the world shows that we have the largest group of adults with diabetes in the world,with an undiagnosed rate of 51.7%.People with diabetes have to control their blood sugar levels for a long time,but patients are more concerned about the various complications that arise from diabetes.Diabetic kidney disease is one of the most serious complications of diabetes,which is often difficult to detect in time as there are no acute signs in the early stages,and once it has progressed to a later stage,it is more difficult to treat than other kidney diseases,causing not only heavy pressure on families and society but also life-threatening problems.Therefore,it is worthwhile to establish a reliable prediction model for diabetic kidney disease to accurately predict the risk of the disease,which has profound practical implications.In this study,the data were first collected from patients’ physical examination report.and collated at the department of laboratory medicine.Data were cleaned and standardized using techniques related to data pre-processing(including data cleaning,data coding,and conversion,data dimensionless).Secondly,the currently available prediction models for diabetic kidney disease lack attention to the problem of data imbalance,resulting in research results that are detached from practical clinical applications.In this study,to improve this problem,different data balancing algorithms(including SMOTE,SMOTE ENN,SMOTE Tomek Link,Borderline SMOTE,ADASYN)were used to achieve the synthesis of a small number of classes of samples to reduce the rate of data imbalance.For the problem of the high dimensionality of the dataset,feature dimensionality reduction was completed using principal component analysis.Finally,a combination of multiple classification algorithms(including decision tree,random forest,support vector machine,BP neural network,one-dimensional convolutional neural network)combined with data balancing algorithms was used to extract and select risk factors for diabetic kidney disease,and the experimental results were analyzed through multiple model evaluation metrics,resulting in a reliable and accurate prediction model for the risk of diabetic kidney disease.The experimental results showed that the combined one-dimensional convolutional neural network-ADASYN model established in this study performed the best under the high-dimensional unbalanced diabetic kidney disease dataset,with the accuracy,recall,F1 value,and AUC value reaching 97.6%,99.4%,98.5% and 0.999 respectively,and finally extracted nine physical examination indicators such as urine leukocytes,hemoglobin,and creatinine as pathogenic risk factors for diabetic kidney disease.The combination of a onedimensional convolutional neural network and ADASYN to construct a disease risk prediction model can provide scientific guidance information for the early prevention and clinical diagnosis of diabetic kidney disease,which is of great significance to the development of intelligent medicine.