Research
The Program for AI and Research in Cardiovascular Medicine aims to improve patient outcomes by improving the interpretation of cardiovascular diagnostic data. With the power of AI, we strive to provide doctors with more precise data to provide exceptional care for patients.
Our Approach
We have allocated resources and developed expertise in securely acquiring, managing, cross-linking and analyzing enterprise-wide patient health information (PHI) data for AI at our academic medical center campus.
Our robust digital cardiovascular database includes 2 million ECGs, 400,000 echocardiograms, 100,000 cardiac nuclear stress studies, 20,000 electrophysiology procedures and 32,000 cardiac catheterizations, and it can be linked to electronic medical records, including diagnosis codes, laboratories, medications, pharmacy data, radiology, and so on. Our database is ripe for developing and validating AI-powered diagnostic algorithms. Furthermore, we have generated meticulous datasets on thousands of procedures, including approximately 1,100 transesophageal echocardiograms, 600 cardiac resynchronization therapy procedures, 5,000 atrial fibrillation ablations, 1,000 left atrial appendage occlusions and 1,000 transcatheter aortic valve replacements.
Our team currently includes cardiovascular specialists, two computer scientist faculty with expertise in AI, one faculty in biomedical engineering, one full-time computer engineer, two doctoral students, one research fellow and multiple housestaff and medical students. Learn more about our team
Innovation
We are using deep learning to eliminate redundant information in the full 10-sec 12-lead ECG signal by encoding it into a limited number of latent variables, a process known as dimensionality reduction. Dimensionality reduction with deep learning can retain the feature extraction embedded in any direct deep-learning predictive model yet avoid overfitting by reducing the ECG data.
Further, training non-deep learning (non-neural network) algorithms using the deep-AI encodings extracted from the small-scale labeled ECGs allows this innovative framework to be feasible in smaller datasets.
Our framework makes us uniquely able to address the lack of large-scale labeling, overfitting and black box problems in deep-learning ECG interpretation by
- deep learning on unlabeled ECGs for dimensionality reduction, and
- using this deep model in smaller labeled datasets to reduce ECGs for training traditional non-deep algorithms.
Explore our recent publications and presentations to learn more.
Funding
American College of Cardiology
Presidential Career Development Award
Funding Period: 07/01/18 - 06/30/19
“Optimizing Cardiac Resynchronization Therapy with Electrocardiographic Imaging”
We used the novel non-invasive 4-dimensional electrical heart mapping to evaluate programming of cardiac resynchronization therapy for alleviating heart failure.
Role: PI (Amit Noheria)
Salary support $70,000
KU Medical Center Research Institute
Lied Clinical Research Pilot Grant
Funding Period: 07/01/20 – 06/30/21
"Quantitative Electrocardiography for Cardiac Resynchronization Therapy"
The goal is to use novel processing methods for processing ECG to evaluate cardiac resynchronization therapy for improving heart failure.
Role: PI (Amit Noheria)
Budget $35,000
Kansas NASA EPSCoR Program (KNEP)
Research Infrastructure Development effort
Funding Period: 5/1/23 – 4/30/24
“Cardiovascular Health Monitoring using Multiple Conformal Photoplethysmography Devices”
The goal is to extract valuable cardiovascular health information under lower body negative pressure to assess the impact of low gravity during space travel.
Role: Co-PI (Amit Noheria)
Total budget $168,265
Department of Cardiovascular Medicine
The department has protected Amit Noheria’s 20-25% effort for research since 2019 and supported salary for Chris Harvey since 2021.