A Multimodal Stage Prediction Study of Chronic Kidney Disease based on Deep Learning

Chronic Kidney Disease (CKD) is a critical global public health issue. Its early stages are often asymptomatic and this insidious nature leads to delayed diagnosis in many patients until the disease progresses to advanced stages where treatment costs increase substantially. Therefore early intervention for CKD is crucial. However traditional singlemodal data struggle to accurately predict disease staging. Based on this this study will analyze complications highly associated with CKD and their variations across different CKD stages. Clinical blood test data will be augmented using the SMOTE oversampling method and appropriate clinical indicators will be selecte. Leveraging this model medical diagnostic text data will be enhanced to build a BERT model with a self-attention mechanism. By integrating these models a deep learning-based multimodal CKD staging prediction model is constructed. Providing a method that comprehensively considers multimodal data to predict CKD staging.