Publications by authors named "Akhil Vaid"
Background: Acute kidney injury (AKI) is common in SARS-CoV-2 infection and COVID-19, often leading to long-term kidney dysfunction. However, the transcriptomic features of AKI severity and its long-term effects are underexplored.
Methods: We performed bulk RNA sequencing on peripheral blood mononuclear cells (PBMCs) from hospitalized SARS-CoV-2 patients and complemented these findings with proteomic data from the same cohort.
Background: Point-of-care ultrasonography (POCUS) enables cardiac imaging at the bedside and in communities but is limited by abbreviated protocols and variation in quality. We aimed to develop and test artificial intelligence (AI) models to screen for under-diagnosed cardiomyopathies from cardiac POCUS.
Methods: In a development set of 290 245 transthoracic echocardiographic videos across the Yale-New Haven Health System (YNHHS), we used augmentation approaches, and a customised loss function weighted for view quality to derive a POCUS-adapted, multi-label, video-based convolutional neural network that discriminates hypertrophic cardiomyopathy and transthyretin amyloid cardiomyopathy from controls without known disease.
Large language models (LLMs) can optimize clinical workflows; however, the economic and computational challenges of their utilization at the health system scale are underexplored. We evaluated how concatenating queries with multiple clinical notes and tasks simultaneously affects model performance under increasing computational loads. We assessed ten LLMs of different capacities and sizes utilizing real-world patient data.
View Article and Find Full Text PDFDrug repurposing-identifying new therapeutic uses for approved drugs-is often a serendipitous and opportunistic endeavour to expand the use of drugs for new diseases. The clinical utility of drug-repurposing artificial intelligence (AI) models remains limited because these models focus narrowly on diseases for which some drugs already exist. Here we introduce TxGNN, a graph foundation model for zero-shot drug repurposing, identifying therapeutic candidates even for diseases with limited treatment options or no existing drugs.
View Article and Find Full Text PDFArticle Synopsis
- Increased intracranial pressure (ICP) ≥15 mmHg can harm neurological health, but measuring it traditionally requires invasive methods; researchers developed a new AI-based biomarker (aICP) using non-invasive extracranial waveform data instead.
- The aICP was validated using an independent dataset and showed good performance metrics with an area under the receiver operating characteristic curve (AUROC) of 0.80 and an accuracy of 73.8%.
- Further analysis indicated that higher aICP predictions are linked to specific health conditions, such as brain tumors and intracerebral hemorrhages, suggesting its potential clinical relevance.
Article Synopsis
- - The diagnosis of congenital long QT syndrome (LQTS) is challenging due to a lack of scalable genetic testing, low prevalence, and normal QT intervals in patients with risky genotypes.
- - Researchers developed a deep learning model that combines ECG waveform data and electronic health records to identify patients with harmful genetic variants indicating LQTS.
- - After training on UK Biobank data and refining the model with diverse cohorts, the approach achieved good accuracy in distinguishing individuals with pathogenic mutations, showing potential for better patient prioritization in clinical settings.
Background: Artificial intelligence-enhanced electrocardiogram (AI-ECG) analysis shows promise to detect biventricular pathophysiology. However, AI-ECG analysis remains underexplored in congenital heart disease (CHD).
Objectives: The purpose of this study was to develop and externally validate an AI-ECG model to predict cardiovascular magnetic resonance (CMR)-defined biventricular dysfunction/dilation in patients with CHD.
Article Synopsis
- Drug repurposing means using old, already approved medicines for new diseases.
- The TXGNN model can help find new uses for drugs, even for diseases that have no treatments yet, and it's way better at predicting drug uses than other methods.
- TXGNN not only predicts where drugs can be used but also explains its predictions clearly, making it easier for doctors to understand and investigate its suggestions.
Background: Artificial intelligence (AI) and large language models (LLMs) can play a critical role in emergency room operations by augmenting decision-making about patient admission. However, there are no studies for LLMs using real-world data and scenarios, in comparison to and being informed by traditional supervised machine learning (ML) models. We evaluated the performance of GPT-4 for predicting patient admissions from emergency department (ED) visits.
View Article and Find Full Text PDFArticle Synopsis
- The study created a framework using an open-source Large Language Model (LLM) to allow clinicians to ask straightforward questions about patients' echocardiogram histories, aiming to improve patient care and research efficiency.
- Data from over a decade of echocardiogram reports at Mount Sinai was analyzed, with the LLaMA-2 70B model processing the information and generating answers that were then validated by cardiologists.
- The results showed the LLM answered 90% of questions accurately on various aspects of echocardiogram interpretations, indicating that this model can significantly improve access to relevant patient data compared to traditional search methods.
Article Synopsis
- Point-of-care ultrasonography (POCUS) is a bedside cardiac imaging technique, but its effectiveness is hampered by inconsistent protocols and image quality, prompting the development of AI models to enhance cardiomyopathy diagnosis.
- Researchers utilized a massive dataset of transthoracic echocardiographic videos to create an AI model that identifies hypertrophic cardiomyopathy (HCM) and transthyretin amyloid cardiomyopathy (ATTR-CM) from POCUS without prior disease knowledge.
- The AI model demonstrated high accuracy in screening for HCM and ATTR-CM, detecting cases about 2 years prior to clinical diagnoses and showing significant prognostic potential for individuals without known cardiomyopathy.
Article Synopsis
- Accurate assessment of ECGs is vital for patient diagnosis and care, but current automated systems lack flexibility and reliability, especially in low-resource areas where specialists review each ECG manually.
- AI systems show promise for improved accuracy but often have limitations in the variety of conditions they can assess and require raw data not typically available to doctors.
- The ECG-GPT model was developed to generate expert-level diagnosis directly from ECG images, demonstrating strong performance across diverse healthcare settings and providing an accessible web application for accurate triage of patients.
Article Synopsis
- Increased intracranial pressure (ICP) can lead to serious neurological problems, but requires invasive methods for monitoring, prompting the need for a non-invasive alternative.
- The study focused on creating and validating an AI model that detects increased ICP using non-invasive physiological data from patients, rather than requiring direct ICP measurements.
- Developed using data from an ICU database, the AI model demonstrated high accuracy and sensitivity in detecting elevated ICP, with promising results in external validation from a separate hospital dataset.
Background: Artificial intelligence-enhanced ECG analysis shows promise to detect ventricular dysfunction and remodeling in adult populations. However, its application to pediatric populations remains underexplored.
Methods: A convolutional neural network was trained on paired ECG-echocardiograms (≤2 days apart) from patients ≤18 years of age without major congenital heart disease to detect human expert-classified greater than mild left ventricular (LV) dysfunction, hypertrophy, and dilation (individually and as a composite outcome).
Quantitative Prediction of Right Ventricular Size and Function From the ECG.
J Am Heart Assoc
January 2024
Background: Right ventricular ejection fraction (RVEF) and end-diastolic volume (RVEDV) are not readily assessed through traditional modalities. Deep learning-enabled ECG analysis for estimation of right ventricular (RV) size or function is unexplored.
Methods And Results: We trained a deep learning-ECG model to predict RV dilation (RVEDV >120 mL/m), RV dysfunction (RVEF ≤40%), and numerical RVEDV and RVEF from a 12-lead ECG paired with reference-standard cardiac magnetic resonance imaging volumetric measurements in UK Biobank (UKBB; n=42 938).
Background: Acute kidney injury (AKI) is common in hospitalized patients with SARS-CoV2 infection despite vaccination and leads to long-term kidney dysfunction. However, peripheral blood molecular signatures in AKI from COVID-19 and their association with long-term kidney dysfunction are yet unexplored.
Methods: In patients hospitalized with SARS-CoV2, we performed bulk RNA sequencing using peripheral blood mononuclear cells(PBMCs).
Background: Substantial effort has been directed toward demonstrating uses of predictive models in health care. However, implementation of these models into clinical practice may influence patient outcomes, which in turn are captured in electronic health record data. As a result, deployed models may affect the predictive ability of current and future models.
View Article and Find Full Text PDFArticle Synopsis
- The study evaluated the performance of AI models, specifically ChatGPT and GPT-4, on USMLE questions related to soft skills like communication, ethics, empathy, and professionalism, which hadn't been assessed before.
- Out of 80 questions from reputable sources, GPT-4 achieved a success rate of 90%, while ChatGPT scored 62.5%, with GPT-4 demonstrating more consistent answers.
- The results suggest that GPT-4 not only outperformed previous AMBOSS users in soft skills but also showed a capacity for empathy, highlighting AI's potential role in addressing the interpersonal demands of medical practice.
Background: Premature ventricular complexes (PVCs) are prevalent and, although often benign, they may lead to PVC-induced cardiomyopathy. We created a deep-learning algorithm to predict left ventricular ejection fraction (LVEF) reduction in patients with PVCs from a 12-lead electrocardiogram (ECG).
Objectives: This study aims to assess a deep-learning model to predict cardiomyopathy among patients with PVCs.
Intradialytic hypotension is common in patients who are on hemodialysis. We applied deep learning techniques to ECGs to predict patients at risk of IDH. The performance of the model was good with an AUC of 0.
View Article and Find Full Text PDFArticle Synopsis
- Interatrial block (IAB) is linked to higher risks of stroke, mortality, and heart failure, particularly in patients without any history of atrial fibrillation (AF) or atrial flutter (AFL).
- A large study analyzed nearly 5 million ECGs from over 1 million patients to explore the association between IAB and adverse outcomes.
- The findings indicate that IAB significantly increases the risk of stroke and other health issues, regardless of the presence of AF/AFL, emphasizing the need for monitoring even in patients without previous arrhythmias.
Article Synopsis
- Acute kidney injury (AKI) is a serious complication of COVID-19, leading to higher in-hospital death rates; researchers used proteomics to find markers for COVID-AKI and long-term kidney issues.
- In a study with two groups of COVID-19 hospitalized patients, they identified 413 proteins with elevated levels and 30 with decreased levels tied to AKI, validating 62 of these in a second group.
- The findings reveal that proteins indicating kidney and heart injury correlate with acute and long-term kidney dysfunction, suggesting that AKI is influenced by various factors, including blood flow issues and heart damage.
The electrocardiogram (ECG) is a ubiquitous diagnostic modality. Convolutional neural networks (CNNs) applied towards ECG analysis require large sample sizes, and transfer learning approaches for biomedical problems may result in suboptimal performance when pre-training is done on natural images. We leveraged masked image modeling to create a vision-based transformer model, HeartBEiT, for electrocardiogram waveform analysis.
View Article and Find Full Text PDF