Publications by authors named "Pierre Elias"
Women with cardiac disease have worse neonatal outcomes compared to women without cardiac disease; risk factors are not well-defined. We hypothesized that structural heart disease, as assessed by echocardiography, is a non-invasive metric for abnormal hemodynamics and an unfavorable maternal-fetal environment. We assessed the association between echocardiographic markers of structural heart disease in women with cardiac disease and a primary endpoint of adverse neonatal outcomes operationalized as neonates with small-for-gestational-age birth weight, preterm delivery, neonatal intensive care unit/transition care unit admission, or neonatal/fatal demise.
View Article and Find Full Text PDFEarly detection of structural heart disease is critical to improving outcomes, but widespread screening remains limited by the cost and accessibility of imaging tools such as echocardiography. Recent advances in machine learning applied to heart rhythm recordings have shown promise in identifying disease, although previous work has been limited by development in narrow populations or targeting only select heart conditions. Here we introduce a deep learning model, EchoNext, trained on more than 1 million heart rhythm and imaging records across a large and diverse health system to detect many forms of structural heart disease.
View Article and Find Full Text PDFBackground And Aims: Classification and risk stratification in aortic (AR), mitral (MR), and tricuspid regurgitation (TR) remains a significant clinical challenge. This study aimed to develop an artificial intelligence (AI) system to assess valvular regurgitation and stratify MR-progression risk.
Methods: Using transthoracic echocardiograms (TTEs) at two sites (internal development/test, external test), the DELINEATE-Regurgitation system was developed to classify AR, MR, and TR severity using colour Doppler videos.
Background: The 12-lead electrocardiogram (ECG) remains a cornerstone of cardiac diagnostics, yet existing artificial intelligence (AI) solutions for automated interpretation often lack generalizability, remain closed-source, and are primarily trained using supervised learning, limiting their adaptability across diverse clinical settings. To address these challenges, we developed and compared two open-source foundational ECG models: DeepECG-SSL, a self-supervised learning model, and DeepECG-SL, a supervised learning model.
Methods: Both models were trained on over 1 million ECGs using a standardized preprocessing pipeline and automated free-text extraction from ECG reports to predict 77 cardiac conditions.
Objective: Study the association between cardiac biomarkers and echocardiography parameters of ventricular performance in neonates with neonatal encephalopathy (NE).
Methods: Prospective observational study (2016-2021) of neonates undergoing therapeutic hypothermia (TH). Neonates with brain injury had repeated echocardiography and biomarkers measurements on day of life (DOL) 2, 3, 4, and 10.
The primary practice of healthcare artificial intelligence (AI) starts with model development, often using state-of-the-art AI, retrospectively evaluated using metrics lifted from the AI literature like AUROC and DICE score. However, good performance on these metrics may not translate to improved clinical outcomes. Instead, we argue for a better development pipeline constructed by working backward from the end goal of positively impacting clinically relevant outcomes using AI, leading to considerations of causality in model development and validation, and subsequently a better development pipeline.
View Article and Find Full Text PDFSimple models vs. deep learning in detecting low ejection fraction from the electrocardiogram.
Eur Heart J Digit Health
July 2024
Article Synopsis
- This study investigates whether simpler models using standard ECG measurements can effectively detect left ventricular systolic dysfunction (LVSD) compared to complex deep learning methods.
- Analyzing a dataset of nearly 41,000 ECGs, researchers found that a random forest model and a logistic regression model both achieved high accuracy in detecting LVSD, with performance comparable todeep learning models.
- The findings suggest that simpler ECG models are not only effective but also easier to implement and interpret in clinical settings, making them potentially more suitable for widespread use.
Article Synopsis
- A novel deep learning (DL) system was developed to enhance the interpretation of transthoracic echocardiography (TTE) for assessing the severity of mitral regurgitation (MR) by integrating multiple video assessments.
- The system was tested with a large dataset (over 61,000 TTEs) and showed high accuracy in classifying MR severity, achieving exact accuracy rates of 82% for internal and 79% for external test sets.
- Most misclassifications occurred between none/trace and mild MR categories, and the use of multiple TTE views improved classification accuracy.
Article Synopsis
- * AI technologies are being applied across various areas such as wearables, electrocardiograms, and genetics, achieving unprecedented detection accuracy for diseases like valvular heart disease and cardiomyopathies.
- * While the number of studies is increasing, rigorous validation is needed to ensure effectiveness and equity, with ongoing trials focused on demonstrating real-world improvements in patient outcomes.
Article Synopsis
- Recent advancements in AI for cardiovascular care are promising for improving diagnosis, treatment, and patient outcomes, with 10% of FDA-approved clinical AI algorithms dedicated to this area.
- The review highlights the use of multimodal inputs and generative AI in cardiology, indicating a shift toward more complex and effective healthcare solutions.
- It emphasizes the importance of careful implementation, ethical considerations, and rigorous evaluation to ensure AI enhances patient care and supports healthcare providers effectively.
Article Synopsis
- The study focused on developing a deep learning model to identify severe left ventricular hypertrophy (SLVH) and dilated left ventricle (DLV) from chest X-rays (CXRs), which could help in early heart failure detection.
- It analyzed 71,589 CXRs from nearly 25,000 patients, validated the model's performance against external data, and compared its accuracy to that of board-certified radiologists.
- The model achieved strong results, outperforming radiologists in detecting abnormalities and providing a publicly available dataset to support further research in this area.
Article Synopsis
- The study investigates how the size of the heart affects the accuracy of SPECT myocardial perfusion imaging (MPI) in identifying obstructive coronary artery disease (CAD).
- Among 2066 patients, it was found that those with a low left ventricular volume had lower diagnostic performance compared to those with larger volumes, particularly affecting older and male patients.
- The results indicate that smaller heart sizes lead to a significant decrease in the effectiveness of SPECT MPI, highlighting the need for tailored diagnostic approaches based on cardiac size, age, and sex.
Electrocardiographic deep learning for predicting post-procedural mortality: a model development and validation study.
Lancet Digit Health
January 2024
Article Synopsis
- Existing preoperative risk assessments are inadequate at predicting postoperative mortality, but deep-learning analysis of ECGs can highlight hidden risk factors.
- A deep-learning algorithm was developed using ECG data from nearly 46,000 patients to more accurately forecast postoperative mortality, and its performance was compared to the Revised Cardiac Risk Index (RCRI).
- In testing, the algorithm achieved an AUC of 0.83, significantly outperforming the RCRI score (AUC of 0.67), indicating its effectiveness across multiple health-care systems.
Artificial Intelligence, Computational Simulations, and Extended Reality in Cardiovascular Interventions.
JACC Cardiovasc Interv
October 2023
Article Synopsis
- Advances in health care are being driven by 21st-century technologies like artificial intelligence, computational simulations, and extended reality, collectively referred to as AISER.
- AISER is being applied in cardiovascular therapies for preprocedural planning, virtual clinical trials, and training health care professionals.
- The review also addresses challenges related to AISER's implementation and highlights the collaboration needed among various experts to enhance its use in cardiovascular medicine.
Article Synopsis
- The electrocardiogram (ECG) is a common test to assess heart health, but its long-term predictive value for cardiovascular risks has been uncertain.
- Researchers developed SEER, a deep learning model that analyzes resting ECGs to estimate long-term cardiovascular mortality risk, achieving high accuracy in predictions.
- SEER not only identifies patients at higher risk for cardiovascular diseases but also enhances existing risk assessment methods, revealing previously unrecognized patients who may need treatment like statin therapy.
Article Synopsis
- - Recent advancements in large language models (LLMs), like GPT-3.5 and ChatGPT, have shown promise in performing well on tasks without needing extensive training, especially in medical evidence summarization across various clinical areas.
- - This study evaluates these models through both automatic and human assessments and highlights that automatic metrics may not reliably reflect the actual quality of the summaries produced.
- - Results indicate that LLMs can generate summaries that contain factual inaccuracies, make dubious or vague statements, and are particularly challenged when summarizing longer texts, raising concerns about the risk of misinformation in high-stakes medical settings.
Article Synopsis
- Researchers studied the link between left cardiac 3D echocardiographic measurements and brain injury in newborns with neonatal encephalopathy in a single hospital.
- They found that on the second day of life, babies with brain injury had larger left ventricle end-diastolic volume and stroke volume compared to those without brain injury.
- Additionally, these infants showed a higher peak global circumferential strain on the 3D echocardiogram, suggesting possible cardiac changes associated with brain injury.
Article Synopsis
- Recent advancements in large language models (LLMs) like GPT-3.5 and ChatGPT show promise in performing zero- and few-shot tasks, including medical evidence summarization in various clinical areas.
- *Our research includes both automatic and human evaluations to assess summary quality, revealing that automated metrics don’t always reflect the true quality of the summaries.
- *We identified specific errors in the models' outputs, such as generating factually inconsistent information and struggling with longer texts, which raises concerns about the potential for misinformation in high-stakes medical contexts.*