Deep-kidney: an effective deep learning framework for chronic kidney disease prediction.

Area of Science:

• Nephrology
• Artificial Intelligence
• Machine Learning

Background:

• Chronic kidney disease (CKD) is a significant global health concern.
• Late manifestation of CKD often leads to severe kidney damage, highlighting the need for early detection.
• Current diagnostic methods may not identify CKD sufficiently early to prevent advanced disease progression.

Purpose of the Study:

• To develop and evaluate advanced predictive models for the early detection of chronic kidney disease (CKD).
• To forecast the potential occurrence of CKD 6 to 12 months prior to clinical manifestation.
• To compare the efficacy of deep learning models, including CNN, LSTM, and a novel deep ensemble approach, for CKD prediction.

Main Methods:

• Development of three predictive models: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a Deep Ensemble model.
• The Deep Ensemble model integrates CNN, LSTM, and LSTM-BLSTM classifiers using a majority voting technique.
• Performance evaluation was conducted using two distinct public datasets to assess prediction accuracy.

Main Results:

• The Deep Ensemble model demonstrated superior performance compared to individual CNN and LSTM models.
• The deep ensemble model achieved high prediction accuracy: 0.993 for 6-month predictions and 0.992 for 12-month predictions.
• These results indicate a strong capability for early and accurate CKD risk identification.

Conclusions:

• The proposed deep ensemble model offers a promising tool for the early prediction of chronic kidney disease.
• Utilizing deep learning, particularly ensemble methods, can significantly improve the timeliness and accuracy of CKD detection.
• Early prediction through advanced AI models has the potential to improve patient outcomes and reduce the burden of advanced kidney disease.