Performance Comparison of a Neural Network and a Regression Linear Model for Predictive Maintenance in Dialysis...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Medical Device Monitoring

Background:

• Dialysis machine reliability is crucial for safe chronic kidney disease treatment.
• Component drift in sensors and actuators can compromise dialysis performance.
• Proactive monitoring is needed to prevent equipment failure and ensure patient safety.

Purpose of the Study:

• To investigate the efficacy of AI, specifically Long Short-Term Memory (LSTM) neural networks, in detecting drift in dialysis machine components.
• To compare the performance of LSTM models against traditional linear regression for anomaly detection.
• To validate the AI approach using real-world clinical data from dialysis machines.

Main Methods:

• Training LSTM and linear regression models on time-dependent signals from dialysis machine components (e.g., weight loss sensor).
• Utilizing normal operational data to establish baseline performance patterns.
• Validating models on real-world clinical data, including complaint cases indicating component degradation or failure.

Main Results:

• LSTM model demonstrated high accuracy in reconstructing normal signals (errors < 0.02).
• LSTM successfully identified anomalies in complaint cases, predicting failures up to five days in advance.
• Linear regression model was only capable of detecting significant deviations, lacking sensitivity for early drift detection.

Conclusions:

• AI-based methods, particularly LSTM networks, offer superior capabilities for monitoring dialysis equipment compared to traditional models.
• AI facilitates early detection of component degradation, enabling predictive maintenance and minimizing unplanned downtime in dialysis care.
• The developed AI model shows potential for integration into clinical and home dialysis settings for scalable and adaptable equipment monitoring.