Cardiopulmonary Bypass: Ready for Data Analysis Using Machine-learning Algorithms?

Background: Since the first successful cardiopulmonary bypass with a heart-lung machine, there have been continuous developments and optimizations in the field of devices and disposable materials. The data continuously recorded during cardiopulmonary bypass for documentation purposes can also be utilized for scientific research. Artificial intelligence, specifically machine learning techniques, is currently revolutionizing the therapy, diagnosis and prediction of diseases. We are evaluating the heart-lung machine and data recording parameters during cardiac surgery to determine whether machine-learning algorithms can be applied to analyze data from cardiopulmonary bypass automatically and thereby improve the outcome of patients.

Methods: We evaluate the technical aspects of the heart-lung machine (pumps and sensors) as well as data recording parameters during cardiac surgery using the following criteria: erroneous data, missing data, imprecise data, multimodal data, state estimation, harmony and integration. Our global objective is to identify missing parameters or incorrectly recorded data of the heart-lung machine to improve data recording and data management to use machine learning algorithms for data analysis.

Results: We found several parameters, such as the actual arterial flow in case of shunts in the cardiopulmonary bypass, continuous volume level in the reservoir and the detection of cardiotomy suction, which are not recorded during cardiopulmonary bypass. However, these are important parameters for data evaluation as well as to control the cardiopulmonary bypass. Due to several different data-management-systems for heart-lung machines and their primary goal of logging purposes, there is no coherent protocol for data recording. Based on our research, the sensor technology of the heart-lung machine should be expanded. This includes flow sensor for arterial line, sensor for volume level in the reservoir and sensor for detecting suction substance (blood/air).

Conclusion: In our view, current technique of the heart-lung machine and data recording during cardiopulmonary bypass is not well-engineered for evaluation of the generated data and analysis by machine learning algorithms. The data available so far can, of course, still be analyzed with reference to the missing data. To achieve a complete data recording and thus have complete data available for analysis, the sensor technology of the heart-lung machine must be improved.

Publikationsverlauf

Artikel online veröffentlicht:
13. Februar 2024