International Perfusion Association
PerfusionGPT is an AI-powered chatbot based on ChatGPT-4, specifically designed to provide expert knowledge for perfusionists in cardiac surgery. It serves as a critical resource for both clinical decision-making and educational purposes in the field of perfusion.
In a study analyzing 37,473 VV-ECMO patients, machine learning was used to predict acute brain injury (ABI), revealing a 7.1% incidence of ABI and identifying pre-ECMO cardiac arrest as the most significant risk factor. The study’s machine learning models, however, showed sub-optimal performance in predicting ABI, attributed to the low prevalence of ABI and the lack of standardized neuromonitoring and imaging protocols in the ELSO Registry data.
A review of the top 25 cardiology and cardiovascular medicine journals according to the 2023 SCImago rankings revealed that all allow the use of generative AI in scientific writing within ICMJE limitations, but prohibit AI in authorship, image generation, and peer review processes, requiring authors to take full responsibility for their work. These guidelines are standardized and followed by all studied journals, emphasizing the importance of adhering to and updating these policies to uphold scientific integrity.
The study describes a simulated cardiopulmonary bypass (CPB) model integrated into a mock operative theater, aiming to enhance the training of novice perfusionists. Results from 81 participants indicated high fidelity and effectiveness of the simulation in replicating real CPB scenarios and skills, suggesting its potential for educational use in clinical perfusion training.
This study evaluates an AI-powered electrocardiogram (ECG) system with six deep neural networks (DNNs), trained to detect 20 diagnostic patterns in standard 12-lead ECGs, and compares its performance to current computerized interpretation methods. The results show that the AI system outperformed or matched state-of-the-art computerized ECG interpretation in all diagnostic categories, demonstrating its potential as a reliable clinical tool for detecting electrocardiographic abnormalities.
This study compares medical texts written by human experts with those generated by ChatGPT, revealing that while ChatGPT’s texts are fluent, they lack the specific, useful information typically found in human-authored texts. It also presents a machine learning method, with over 95% accuracy, for distinguishing between these two types of texts, underscoring the importance of responsible AI use in medicine.
Machine learning models, specifically extreme gradient boosting and light gradient boosting, demonstrated higher accuracy than conventional models in predicting 90-day mortality for patients undergoing venovenous extracorporeal membrane oxygenation (VV-ECMO). These advanced models, outperforming existing methods like RESP and PRESERVE, show potential for improving patient selection by identifying those less likely to benefit from VV-ECMO therapy.
This study used artificial intelligence to analyze preoperative CT scans of heart failure patients receiving left ventricular assist device (LVAD) implantations, revealing that greater adipose tissue areas are associated with higher postoperative complications and in-hospital mortality. It concluded that preoperative body composition, particularly adipose tissue, can predict poorer outcomes post-LVAD implantation, impacting postoperative quality of life and walking distance.
This qualitative study explored organizational readiness for sharing health data for AI development across various sectors, finding that motivation and capabilities are central to data-sharing efforts. The study highlights that while organizational values align with data-sharing priorities, incentives can further enhance cross-sector collaborations and overcome barriers, suggesting tailored incentives may improve the sustainable flow of health data for AI development.
This study developed an XGBoost machine learning model using intraoperative blood pressure time-series data to predict the risk of acute kidney injury (AKI) after Type A acute aortic dissection repair. The model, which outperformed others in accuracy, identified factors like intraoperative urine output and the duration of mean arterial pressure below 65 mmHg as critical predictors for postoperative AKI.
