Research questions: How to predict chronic kidney disease precisly and efficiently?
Objectives: Develop an AI hybrid deep learning model to predict early-stage CKD
Scienfic rationale for the research:
The prevalence and incidence of chronic kidney disease (CKD) has increased significantly in recent years. In early stages, CKD is hard to be aware by patients since it does not perform significnat symptons such as headache, fever and so on. So many patients, especially in China, are diagnosed at stage 3b or 4 CKD when they firstly attended to the hospital. It has brought huge burden for both patients and social economies. Previous studies have found a series of biomarkers, which included laboratory testing results, special characteristics of fundus images, genomics and proteomics data, for early detection of CKD. But the precision of those models are limited since many of them used single modal data with simple machine learning algorithm. Along with the development of computer science, the feasibility of developing predicdion model by using multimodal is significnatly improved than before. Our studies also found that the performance of prediction model was better when we adding data extracted from fundus images to the development dataset, which was previously full filled with tabulated data. In addition, genomic variation also palys an important role on development of CKD. For instance, previous studies from Columbia University reported the diagnostic utility of exome sequencing for kidney diseases. So, the genetic diagnosis method should be taken into account when developing the prediction model, especially for the risk screening in general population.
