An interpretable 1D convolutional neural network for detecting patient-ventilator asynchrony in mechanical...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Care

Background:

• Patient-ventilator asynchrony (PVA) arises from a mismatch between patient needs and ventilator support during mechanical ventilation.
• PVA is linked to poorer clinical outcomes, necessitating effective detection and correction strategies.
• Deep learning shows potential for PVA detection, but clinical application is limited by a lack of network interpretability.

Purpose of the Study:

• To develop an interpretable deep learning model for detecting common types of PVA.
• To enhance the clinical utility of AI in mechanical ventilation by providing transparent PVA detection.

Main Methods:

• An interpretable one-dimensional convolutional neural network (1DCNN) was designed to detect four types of PVA.
• Global average pooling (GAP) was used to highlight critical waveform sections for classification.
• Dilation convolution and batch normalization were incorporated to maintain model performance.

Main Results:

• The interpretable 1DCNN achieved performance comparable to state-of-the-art models for PVA detection.
• F1 scores exceeded 0.96 for detecting four PVA types across different ventilation modes.
• Highlighted waveform sections aligned with expert understanding of PVA, validating model interpretability.

Conclusions:

• The proposed 1DCNN effectively detects PVA with high accuracy.
• The model's interpretability allows clinicians to understand the AI's decision-making process.
• This technology can aid clinicians in better managing mechanical ventilation and improving patient outcomes.