Development and Validation of an Arterial Pressure-Based Cardiac Output Algorithm Using a Convolutional Neural...

Area of Science:

• Cardiovascular physiology
• Medical artificial intelligence
• Hemodynamic monitoring

Background:

• Arterial pressure-based cardiac output (APCO) offers a less invasive alternative to pulmonary artery catheters (PAC) for estimating cardiac output.
• Existing APCO devices have reported inaccuracies, limiting their clinical utility.
• Proprietary algorithms hinder research and development for APCO accuracy improvements.

Purpose of the Study:

• To develop and validate an open-source APCO algorithm utilizing convolutional neural networks (CNNs) and transfer learning.
• To enhance the accuracy of non-invasive cardiac output monitoring.
• To provide a more reliable tool for clinical hemodynamic management.

Main Methods:

• A retrospective study analyzed intraoperative bio-signal data from a university hospital cohort.
• A CNN model was trained using arterial pressure waveforms to predict stroke volume (SV).
• Transfer learning involved pretraining on commercial APCO device SV and fine-tuning with PAC-derived SV, with performance evaluated against PAC SV.

Main Results:

• The study included 2057 surgical cases (1958 training, 99 testing).
• The deep learning model achieved mean absolute SV errors of 14.5 mL (overall), 10.2 mL (cardiac surgery), and 17.4 mL (liver transplantation).
• The deep learning model significantly outperformed the commercial FloTrac device (P<.001) in accuracy.

Conclusions:

• The developed deep learning-based APCO algorithm demonstrates superior performance over commercial devices.
• This open-source approach offers a promising avenue for accurate cardiac output estimation.
• Further refinement of this algorithm could significantly aid clinical practice and optimize care for high-risk patients.