Biomedical engineer Abhinav Jha, an assistant professor in the McKelvey School of Engineering and School of Medicine in the Department of Radiology at Washington University in St. Louis, has long argued that artificial intelligence (AI) tools used in medical imaging applications need to be evaluated based on the clinical task, not on their visual appeal.
In a study published in IEEE Transactions on Radiation and Plasma Medical Sciences, Jha and colleagues developed a tool that shows potential for improving clinical performance by denoising myocardial perfusion imaging (MPI) single-photon emission computed tomography (SPECT) images.
To obtain these images, which help doctors evaluate blood flow to the heart muscle, patients must first be administered a radioactive tracer and then remain motionless for up to 15 minutes while the scan is performed. Reducing the amount of tracer administered, the time required, or both would benefit patients, streamline the process, and reduce imaging costs. However, it would also reduce the ability to visualize heart defects in the images.
Inspired by their understanding of how the human visual system works, Jha's team developed a detection task-specific deep learning-based approach to denoise and improve the quality of these low-count MPI SPECT images. This new tool, called DEMIST, leverages a deep learning framework to selectively clean up such images while preserving features that impact the detection task.
For more information, visit the McKelvey School of Engineering website.
