PROJECT SUMMARY
Coronary revascularization (CR) is a standard interventional treatment for patients with symptomatic stable
coronary artery disease (CAD). However, the performance of such interventions without appropriate patient
selection is not superior to medical therapy alone. Consequently, to achieve enhanced patient selection, many
have advocated for physiological/anatomical information integration of myocardial functions and coronary
arteries before the CR. As a result, we aim to develop enabling technologies allowing for comprehensive and
quantitative assessments of myocardial functions and coronary arteries. Leveraging the proposed technologies,
an interventional cardiologist can identify the most appropriate lesions to treat using CR, impacting the
management of patients with CAD.
This project first combines gated SPECT myocardial perfusion imaging (MPI) with invasive coronary angiography
(ICA). Then, machine-learning-based methods can reliably interpret the fusion results to support clinical decision-
making. Studies have already demonstrated that CR decisions based on MPI improve outcomes over anatomical
assessment or medical therapy alone. Once ICA is combined, the SPECT-ICA fusion map provides
complementary information about both myocardial functions (perfusion and wall motion) and coronary arteries
(anatomy and assessments of stenosis). Technological developments in the PI’s lab demonstrate it is feasible
to fuse MPI with ICA data. Our recent clinical validation showed that: the number of coronary stenosis segments
with uncertainty was significantly reduced by our 3D SPECT-ICA fusion compared with side-by-side analysis;
patients who received CR congruent with guidance by 3D fusion had superior outcomes when compared with
those who did not.
In this proposal, using state-of-the-art machine learning techniques, further technological developments improve
the clinical utility of our software system. More specifically, our primary objectives are twofold. First, we use
machine-learning-based ICA image processing to reduce our technologies' processing time and minimize user
variabilities, making clinical translation feasible. Second, we improve the result interpretation of the SPECT-ICA
fusion map using machine-learning methods. More specifically, a 3D SPECT-ICA fusion map can be generated
for each patient after multimodality fusion. A new model using the latest reinforcement learning techniques will
enhance the interpretation of the SPECT-ICA fusion map and further improve CR outcomes. It is important to
note that all the new algorithms and techniques will receive rigorous validations. A retrospective single-center
study in a cath lab will be conducted to identify further clinical values of our fusion approach in guiding the CR.
Completion of this proposal will not only advance the clinical translation of technologies developed in the PI’s
lab, improving the clinical decision-making for patients under consideration for CR, but also enhance Michigan
Tech’s research profile by introducing translational cardiology to our students.
