Surfacing insights into left and right ventricular dysfunction through deep learning
Inventors
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Assignees
Member
Icahn School of Medicine at Mount SinaiThe Icahn School of Medicine at Mount Sinai, located in New York City, is an international leader in biomedical education, research, and patient care. As the academic partner of the Mount Sinai Health System, the school is renowned for its innovative education, groundbreaking research, and commitment to health equity. With over 7,000 faculty, 1,200 students, and 2,500 residents and fellows, the institution fosters a culture of bold thinking, multidisciplinary teamwork, and a willingness to challenge conventional wisdom. Its mission is to radically advance the art and science of medical care through collaborative learning, scholarly inquiry, and a deep respect for diversity, preparing the next generation of healthcare leaders to revolutionize medicine and biomedical science.
The Icahn School of Medicine at Mount Sinai, located in New York City, is an international leader in biomedical education, research, and patient care. As the academic partner of the Mount Sinai Health System, the school is renowned for its innovative education, groundbreaking research, and commitment to health equity. With over 7,000 faculty, 1,200 students, and 2,500 residents and fellows, the institution fosters a culture of bold thinking, multidisciplinary teamwork, and a willingness to challenge conventional wisdom. Its mission is to radically advance the art and science of medical care through collaborative learning, scholarly inquiry, and a deep respect for diversity, preparing the next generation of healthcare leaders to revolutionize medicine and biomedical science.
Publication Number
US-12102485-B2
Publication Date
2024-10-01
Expiration Date
2042-04-05
Abstract
Introduced here approaches to developing, training, and implementing algorithms to cardiac dysfunction through automated analysis of physiological data. As an example, a model may be developed and then trained to quantify left and right ventricular dysfunction using electrocardiogram waveform data that is associated with a population of individuals who are diverse in terms of age, gender, ethnicity, socioeconomic status, and the like. This approach to training allows the model to predict the presence of left and right ventricular dysfunction in a diverse population. Also introduced here is a regression framework for predicting numeric values of left ventricular ejection fraction.
Core Innovation
The invention introduces approaches for developing, training, and implementing algorithms to detect cardiac dysfunction through automated analysis of physiological data, specifically electrocardiogram (ECG) waveform data. A model is developed and trained using ECG data associated with a diverse population characterized by age, gender, ethnicity, and socioeconomic status to quantify left and right ventricular dysfunction, enabling prediction of such dysfunctions in a diverse population.
This approach extends to a regression framework for predicting numeric values of left ventricular ejection fraction (LVEF), allowing quantitative assessment of ventricular function. The methods encompass processing diverse datasets, including electronic health records and echocardiogram reports, using neural networks that combine ECG waveform data and other clinical parameters to provide diagnostic predictions regarding ventricular health.
The problem addressed is the significant health burden of heart failure, particularly difficulties in early detection and quantification of left and right ventricular dysfunction. Existing methods such as echocardiography to measure LVEF face barriers including the need for skilled personnel, variability in measurements, limited availability in resource-poor settings, and challenges in assessing right ventricular function. Electrocardiogram data, although widely available and inexpensive, has limited utility due to subjective interpretation and lack of guidelines, particularly for right ventricular dysfunction.
The invention solves these problems by leveraging deep learning to analyze ECG data for detecting and quantifying ventricular dysfunction, including right ventricular systolic dysfunction (RVSD) and right ventricular dilation (RVD). By training models on paired ECG and echocardiogram data processed via natural language processing (NLP) methods, this approach enables automated, objective, and wide-scale screening and diagnosis of cardiac dysfunction with high accuracy, generalizability, and applicability across diverse populations.
Claims Coverage
The patent includes a set of independent claims that cover computer-readable media with instructions, and methods for training and using computer-based models for cardiac dysfunction detection.
Training computer-based model using paired ECG and echocardiogram data
Training a model by accessing left ventricular information from ECGs and right ventricular information from transthoracic echocardiograms, processing pairs within a predetermined time threshold, associating them with recorded medical observations, and using them to train the model to detect cardiac dysfunction.
Filtering ECG data based on LVEF thresholds
Filtering first cardiac information by discarding ECG data where left ventricular ejection fraction values exceed upper or lower thresholds to remove outliers and improve model training.
Predicting ventricular dysfunction using trained model
Generating prompts with patient ECG data outside the training set, transmitting to the trained model, and receiving predictions regarding right ventricular dysfunction, left ventricular dysfunction, or both.
Using regression framework for LVEF prediction
Employing a regression framework within the trained model to predict numeric values of left ventricular ejection fraction.
Applying natural language processing to medical observations
Applying an NLP algorithm to unstructured medical data (e.g., echocardiogram reports) to determine medical observations that serve as labels or inputs for training the model.
Deploying graphical user interface for patient data interaction
Providing a graphical user interface with controls to receive information related to patients and healthcare facilities to acquire appropriate cardiac data for model training or application.
Computer-based model comprising an artificial neural network
Implementing the computer-based model as an artificial neural network to perform the analysis and predictions.
The independent claims cover training computer-based models using paired ECG and echocardiogram data, filtering and processing medical data, applying NLP, deploying user interfaces, and using neural networks to automatically detect left and right ventricular dysfunction, predict LVEF values, and provide clinical predictions from ECG data.
Stated Advantages
Automated analysis of physiological data provides earlier and more objective detection of ventricular dysfunction, including subclinical pathology, compared to manual ECG interpretation.
The approach is applicable to a diverse patient population, improving generalizability across age, gender, ethnicity, and socioeconomic status.
The regression framework allows quantifying numeric LVEF values, which supports monitoring disease progression and treatment response.
Models reduce reliance on skilled echocardiographers and costly imaging modalities, enabling affordable and accessible screening especially in resource-limited settings.
The method improves diagnostic accuracy and consistency over traditional expert interpretation by using deep learning neural networks.
Documented Applications
Detecting left and right ventricular dysfunction using ECG waveform data in a diverse patient population.
Quantifying left ventricular ejection fraction as a numeric value from ECG data.
Screening and diagnosing right ventricular systolic dysfunction and right ventricular dilation through analysis of ECG data paired with echocardiogram-derived labels.
Guided diagnosis of valvular diseases, including subtypes of severe aortic stenosis, using ECG data combined with NLP analysis of echo reports.
Deploying the models within clinical diagnostic platforms and interfaces for assisting healthcare professionals in cardiac disease diagnosis and patient stratification.
Interested in licensing this patent?