Related Experiment Video
Updated: Sep 24, 2025
11:09
High-Resolution Cardiac Positron Emission Tomography/Computed Tomography for Small Animals
Published on: December 16, 2022
3.9K
Explainable Deep Learning Improves Physician Interpretation of Myocardial Perfusion Imaging
Robert J H Miller1,2, Keiichiro Kuronuma1,3, Ananya Singh1
1Division of Artificial Intelligence in Medicine, Department of Medicine, Imaging, and Biomedical Sciences, Cedars-Sinai Medical Center, Los Angeles, California.
Summary
Explainable deep learning (DL) significantly improved physician accuracy in interpreting myocardial perfusion imaging (MPI) for coronary artery disease (CAD). This AI tool enhances diagnostic capabilities when used as an aid to clinicians.
Area of Science:
- Cardiology
- Artificial Intelligence in Medicine
- Medical Imaging
Background:
- Myocardial perfusion imaging (MPI) is crucial for diagnosing coronary artery disease (CAD).
- Deep learning (DL) shows promise in improving diagnostic accuracy for CAD.
- The impact of explainable DL on physician interpretation of MPI is not well understood.
Purpose of the Study:
- To assess if explainable DL predictions enhance physician interpretation of MPI for obstructive CAD.
- To compare the diagnostic accuracy of physicians with and without access to DL insights.
Main Methods:
- A cohort of 240 patients undergoing MPI with invasive coronary angiography was studied.
- An explainable DL model (CAD-DL) was used to predict obstructive CAD.
- Physicians interpreted MPI studies with and without CAD-DL results, with accuracy measured by AUC.
Main Results:
- Physician interpretation aided by CAD-DL showed significantly higher diagnostic accuracy (AUC 0.779) compared to interpretation without DL (AUC 0.747).
- DL improved sensitivity and led to a 17.2% net reclassification improvement.
- Accuracy improvements varied among physicians, indicating differential acceptance of the technology.
Conclusions:
- Explainable DL predictions meaningfully improve physician interpretation of MPI.
- DL can be effectively implemented as an assistive tool to enhance diagnostic accuracy in MPI.
- Further research may explore physician adoption and integration of DL in clinical workflows.
