Approved research

Software for Heart Chamber Segmentation and Characterization Using MRI

Lay summary

Cardiovascular disease (CVD) claims more lives each year than cancer, chronic respiratory diseases, accidents, and diabetes combined. Approximately 30% of all deaths are attributed to CVD. Cardiac MRI (CMR) imaging enables accurate quantification of cardiac function providing information for diagnosis and monitoring of CVDs. It is performed at high resolution, yet clinicians' analysis is remarkably variable. Current standard of analysis of CMR images relies on manual approaches, which is time consuming and prone to subjective errors. We propose to devise and deploy highly automated software to greatly speed CMR image analysis, improve reproducibility, in-crease accuracy, and harmonize analysis. It will reduce costs by decreasing manual labor, and with improved reproducibility, possibly enable the use of historical data, eliminating cost of a control arm. In the clinic, rather than manually reviewing 100s of images in a 4D scan, with fast software, it will be possible to present the heart chamber volumes in 3D and 4D, providing instant feedback to physicians on potential irregularities. Software will au-tomatically determine chamber volumes over time, allowing to determine ejection fraction, strains, strain rates, mass, and various other indices important for treatment decision making. Our platform will transform image analysis and diagnostics. Equipped with immersive patient-specific analysis, physicians will be able to make rapid and improved decisions on treatments, with higher confidence in their decision while reducing disagreement rate among experts. We aim to overcome critical barriers through the following aims: 1) Create prototype offline cardiac analysis soft-ware (MAAA-R), 2) Demonstrate robustness and accuracy of MAAA-R retrospectively on a large and diverse clinical dataset, and 3) Create speed optimized code demonstrating feasibility of a live time product (MAAA-C) for determination cardiac functions in clinical settings. To quantify impact, we will collaborate with Beth Israel Deaconess Medical Center (Boston) and Newark Beth Israel Medical Center (New Jersey). Both medical centers will conduct a comparison studies comparing our soft-ware performance against the current standard of work in these two institutions. To ensure success, we have assembled an expert interdisciplinary team spanning interventional cardiologist, deep learning experts, computer graphics, and software experts. Ultimately, the proposed platform will drive improved patient outcome which will result in greater patient satisfaction and retention by hospitals. A longer-term goal is to launch clinical studies demonstrating impact on patient outcomes, enabling Dyad Medical to provide personalized predictive analysis. Given our promising preliminary results, we estimate that this project will take 12-15 months to complete