PROJECT SUMMARY
Cardiac resynchronization therapy (CRT) is a standard treatment for heart failure (HF) by coordinating the
function of the left and right ventricles. However, 30-40% of CRT recipients do not have improved clinical
symptoms and cardiac functions. The main reasons for CRT non-response include: (1) Selection of patients
based on electrical dyssynchrony measured by ECG under current guidelines is not optimal. (2) Mechanical
dyssynchrony is proven effective but is not fully explored. (3) The CRT left ventricular (LV) lead may not be
placed in an appropriate position in a significant number of patients. Due to the complexity of HF and the
mechanism of CRT pacing, the advancement of image-guided approaches for CRT is still limited: existing
predictors that measure electrical dyssynchrony and mechanical dyssynchrony are insufficient to characterize
the severity of electrical/mechanical dyssynchrony in all ventricular segments; on the other hand, numerous
complicated inter-correlated predictors entangle multi-stage clinical decision making for CRT delivery.
The objective of this research is to improve CRT patient selection and LV lead pacing by integrative
analysis of electrical dyssynchrony on ECG and mechanical dyssynchrony on gated SPECT myocardial
perfusion imaging (MPI). Different from existing studies, which use supervised machine learning (ML) to
combine all clinical factors to predict CRT response, this translational approach is dedicated to knowledge
discovery and clinical decision-making support; we will use unsupervised machine learning to find patient
subgroups that have a higher likelihood to respond to CRT and use reinforcement learning (RL) to both optimize
and explain the multi-stage clinical decision-making process of CRT patient selection, and design a new method
incorporating electrical dyssynchrony, myocardial viability, and mechanical dyssynchrony to recommend the LV
pacing sites. Completion of this proposed project will result in the discovery of new clinically interpretable
knowledge and computer techniques to improve CRT response in clinical practice. It is important to note that all
the new algorithms and knowledge will receive rigorous validations.
The proposed research shows our continuous effort and innovative methods to investigate this long-lasting and
significant cardiovascular problem. It utilizes state-of-the-art computer algorithms and techniques to analyze
cardiovascular images for improved medical treatments, and will greatly benefit our students, offering
opportunities for them to engage in cutting-edge cardiovascular research. It will thus diversify university research
by introducing clinical cardiology practice to our well-established computing programs and promoting integrative
education and discovery-based learning for undergraduate students. The preliminary data, the PI’s experience
in developing innovative computer algorithms on medical image analysis and machine learning and supervising
undergraduate research students, and our interdisciplinary collaboration, have fully prepared the team for the
execution of this project.
