@article{1e7be9e76f794363a5acbe101a0e44a3,
title = "Automated Interpretation of Myocardial Perfusion Images: Tell Me AI Where to Look",
keywords = "artificial intelligence, coronary artery disease, deep learning, myocardial perfusion imaging, single-photon emission computed tomography",
author = "Sengupta, \{Partho P.\} and M{\'a}rton Tokodi",
note = "Funding Information: The development of {\textquotedblleft}attention maps{\textquotedblright} for interpreting MPI is relevant, given that the average attention span for humans immersed in a digital world is rapidly narrowing.16 Moreover, in a multitasking mind, the visual cortex cannot process all the rich and diverse information that impinges on vision. Hence, items must compete for selection. This prioritization can lead to an inability to detect important visible objects and patterns, a well-researched phenomenon known as {\textquotedblleft}inattentional blindness{\textquotedblright}.17 This problem of sensory inattention is pronounced in physicians who juggle interpreting cardiac images, seeing patients, texting, answering phone calls, coordinating care, and completing electronic medical records. Such distractions can lead to errors in critical cognitive tasks that require intense focus and attention. Therefore, decision support and {\textquotedblleft}attention maps{\textquotedblright} for interpreting an MPI test seem logical steps. However, it needs further investigation to see whether physicians will adopt AI-generated {\textquotedblleft}attention maps{\textquotedblright} over their interpretation skills. Specifically, the imperfect correlation between the {\textquotedblleft}attention maps{\textquotedblright} and perfusion deficits seen in this investigation may be challenging for physicians to settle with. These differences between the area of noticeable perfusion defects and AI-driven {\textquotedblleft}attention maps{\textquotedblright} might reside in the information encoded at the pixel level that the naked eye may not readily distinguish. Further investigation of pixel-level data using a radiomics-based strategy could provide new insights into the model's ability to predict CAD.",
year = "2022",
month = jun,
doi = "10.1016/j.jcmg.2022.04.017",
language = "English (US)",
volume = "15",
pages = "1103--1106",
journal = "JACC: Cardiovascular Imaging",
issn = "1936-878X",
publisher = "Elsevier Inc.",
number = "6",
}