Predicting stroke volume variation using central venous pressure waveform: a deep learning approach

Area of Science:

• Anesthesiology and Critical Care Medicine
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Accurate estimation of stroke volume variation (SVV) is crucial for guiding fluid management in critically ill patients.
• Central venous pressure (CVP) waveforms offer a potential, less invasive source of data for SVV prediction compared to arterial methods.
• Traditional methods for SVV estimation often rely on arterial pulse waveform analysis, which may not always be feasible or optimal.

Purpose of the Study:

• To evaluate the predictive performance of a deep learning model for estimating SVV from CVP waveforms.
• To compare the accuracy of the deep learning model's SVV predictions against a commercial arterial pulse waveform analysis system.

Main Methods:

• A deep learning architecture combining Long Short-Term Memory (LSTM) and feed-forward neural networks was developed.
• The model utilized 10-second CVP waveforms and demographic data as inputs.
• Performance was assessed by comparing predicted SVV with SVV estimated by the commercial EV1000 device using concordance correlation coefficient (CCC).

Main Results:

• The deep learning model achieved a high concordance correlation coefficient (CCC) of 0.993 (95% CI: 0.992-0.993) when comparing predicted SVV to EV1000-derived SVV.
• The model was trained and tested on a large dataset comprising 224 cases with over 1.7 million CVP waveforms.
• The model demonstrated robust performance in approximating SVV values.

Conclusions:

• Deep learning models can effectively predict SVV using readily available CVP waveforms.
• This CVP-based deep learning approach provides an accurate alternative to commercial arterial pulse waveform analysis for SVV estimation.
• The findings suggest a potential for improved, non-invasive hemodynamic monitoring in clinical practice.