New deep learning method could revolutionize heart health monitoring

Coronary artery disease is the leading cause of death globally. One of the most common tools used to diagnose and monitor heart disease, myocardial perfusion imaging (MPI) by single photon emission computed tomography (SPECT), uses a radioactive tracer and special camera to provide detailed images of blood flow to the heart, helping doctors detect coronary artery disease and other cardiovascular abnormalities. However, traditional SPECT imaging requires an additional CT scan to ensure accurate results, exposing patients to more radiation and increasing costs.

A new deep learning technique developed by researchers at Washington University in St. Louis with collaborators from Cleveland Clinic and University of California Santa Barbara could transform the way heart health is monitored, making it safer and more accessible.

The method, known as CTLESS, leverages deep learning to remove the CT requirement without compromising diagnostic accuracy. The project, led by Abhinav Jha, associate professor of biomedical engineering in the McKelvey School of Engineering and of radiology at WashU Medicine Mallinckrodt institute of Radiology, was published online Nov. 25 in IEEE Transactions in Medical Imaging.

The next stage of research is for them to validate this method while working to make this tech more available to rural community hospitals. Their cost-saving technique is particularly significant for cases where access to such scans may be limited, such as in rural or otherwise resource-limited communities, said Jha.

SPECT imaging requires an additional CT scan for attenuation compensation (AC), which corrects for how the emitted signal weakens, or attenuates, as it moves through body tissue, potentially obscuring heart images and leading to diagnostic inaccuracies. Such CT scans are typically acquired on a SPECT/CT scanner, but many facilities do not have this CT component.

Due to cost, complexity, equipment availability, regulatory concerns and other local factors at hospitals and remote care centers, approximately 75% of all SPECT MPI scans are performed without AC, potentially compromising the diagnostic accuracy of these scans. By integrating concepts in physics and deep learning, the proposed CTLESS method estimates a synthetic attenuation map that is then used for AC. Thus, CTLESS may enable a mechanism where an additional scan may not be required."

Abhinav Jha, associate professor of biomedical engineering, McKelvey School of Engineering and of radiology at WashU Medicine Mallinckrodt institute of Radiology

CTLESS uses photons from the emission scan to estimate attenuation, which can then be used to enhance image quality and improve diagnostic interpretation. Jha and his collaborators evaluated the performance of CTLESS using real-world clinical data and found that their method showed comparable results to traditional attenuation compensation.

Notably, CTLESS demonstrated robust performance across different scanner models, degrees of heart damage and patient demographics. Jha noted that anatomical differences between men and women result in varying levels of attenuation in these groups and confirmed that the CTLESS method yields similar performance as traditional AC for both sexes. The performance of CTLESS was also relatively stable even as the size of the training data was reduced. All these observations make CTLESS a promising option for widespread clinical adoption following additional validation.

"Our results provide promise that in the future, a separate CT scan may not be required for performing attenuation correction in MPI SPECT. This is particularly significant for cases where access to such scans may be limited, such as in rural or otherwise resource-limited communities," Jha said. "By providing the ability to perform AC without requiring a CT, the proposed CTLESS method may help boost technological health equality across the U.S. and worldwide."

Yu Z, Rahman MA, Abbey CK, Laforest R, Siegel BA, Jha A. CTLESS: A scatter-window projection and deep learning-based transmission-less attenuation compensation method for myocardial perfusion SPECT. IEEE Transactions in Medical Imaging, Nov. 25, 2024, DOI: https://doi.org/10.1109/TMI.2024.3496870.

This work was supported in part by the National Institute of Biomedical Imaging and Bioengineering (R21-EB024647, R01-EB031051 and R01-EB031962), the Bradley-Alavi Student Fellowship from the Education and Research Foundation for Nuclear Medicine and Molecular Imaging, and the National Science Foundation CAREER Award (2239707).