Deep Learning for Echo Analysis, Tracking, and Evaluation Prospective Evaluation (DELINEATE-Prospective)
试验速览
- 阶段
- 不适用
- 状态
- 招募中
- 入组人数
- 50
- 试验地点
- 2
- 主要终点
- Proportion of Clinically Meaningful Reclassification by Panel Review
研究概览
简要总结
Heart disease is the leading cause of death in the United States, and echocardiography (or "echo") is the most common way doctors look at the heart. Echo is safe, painless, and can detect major heart problems, including weak heart pumping and valve disease.
Valve disease, especially aortic stenosis (narrowing) and mitral regurgitation (leakage), is common in older adults but often goes undiagnosed. While echo is the main tool for finding valve problems, it takes time, requires expert training, and results can vary between readers.
Recent advances in artificial intelligence (AI), especially deep learning (DL), have shown promise in automatically analyzing heart images. However, past research hasn't fully tackled key echo techniques-like color Doppler and spectral Doppler-that are crucial for measuring how blood moves through heart valves. AI tools also face challenges in being used in everyday medical practice because of workflow issues, lack of real-world testing, and concerns about how the algorithms make decisions.
At Columbia University Irving Medical Center, researchers have built a large database of heart tests over the last six years and developed AI programs to analyze echocardiograms. The current study will test whether providing AI analysis to cardiologists in real time during echo reading can make the process faster and more consistent.
详细描述
In a prior Columbia University study, a series of deep learning algorithms analyzing echocardiograms is in development. These algorithms include, but are not limited to, algorithms that enable view classification, structure identification, left ventricle (LV) dimension measurements, Left Ventricular Ejection Fraction (LVEF) determination, left atrium (LA) volume assessments, and valvular heart disease diagnosis. Briefly, these algorithms are based on architectures shown to be useful in image and video analysis, including ones specific to echocardiography interpretation. Algorithms based off these architectures can be generalized to interpretation of video-based echocardiogram data such as valvular regurgitation assessment. As part of this study protocol, these models will continue to be developed using patient echocardiogram data. This study aims to create an automated, end-to-end system that can deliver deep learning analyses of echocardiograms to the interpreting cardiologist in real-time. If successful, this program could enable improvements in echocardiography reading efficiency and reliability.
研究设计
- 研究类型
- Observational
- 观察模型
- Case Control
- 时间视角
- Prospective
入排标准
- 年龄范围
- 18 Years 至 —（Adult, Older Adult）
- 性别
- All
- 接受健康志愿者
- 否
入选标准
- •Attending cardiologist employed by Columbia University, ColumbiaDoctors, or NewYork Presbyterian Hospital who reads transthoracic echocardiograms in the Columbia echocardiography laboratory
- •Provided informed consent to take part in the questionnaires or pivotal study
排除标准
- •Physician in training (cardiology fellow or advanced imaging fellow)
研究组 & 干预措施
Intervention Group
Studies meeting the following criteria will undergo adjudication by an expert panel: Moderate, moderate-severe, or severe mitral, aortic, or tricuspid regurgitation by physician or AI model assessment.
Discrepancy between physician and AI interpretations, where AI-assessed severity is greater than the physician-assessed severity (i.e. indicates that more valvular regurgitation is present)
Control Group
A stratified random sample of cases will be selected to match the distribution of AI-flagged cases by physician-assessed valvular regurgitation severity and will undergo the same expert panel adjudication.
结局指标
主要结局
Proportion of Clinically Meaningful Reclassification by Panel Review
时间窗: 18 months
Proportion of cases where the expert panel reclassifies valvular regurgitation severity by at least one grade (upgrade or downgrade). The proportion will be calculated as the number of cases with reclassification ÷ total number of cases reviewed.
次要结局
- Proportion of Cases with AI-Based Reclassification Leading to a Change in Clinical Management(18 months)
- Proportion of Cases with AI-Based Reclassification Leading to Referral to a Valve Specialist or Surgeon(18 months)
- Proportion of Cases with AI-Based Reclassification Leading to a Change in Frequency of Follow-Up Echocardiography(18 months)
- Proportion of Cases with AI-Based Reclassification Leading to Referral for Further Testing (TEE or Cardiac MRI)(18 months)
研究者
Pierre Elias
Assistant Professor of Medicine in the Department of Biomedical Informatics
Columbia University