Interpretation of SPECT wall motion with deep learning.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Single-photon emission computed tomography (SPECT) wall motion assessment faces limitations in temporal and spatial resolution.
• Visual interpretation of SPECT wall motion is subjective and prone to errors.
• Artificial intelligence (AI) presents a potential solution to enhance the accuracy of wall motion assessment.

Purpose of the Study:

• To develop a novel deep learning (DL) workflow for interpreting SPECT wall motion.
• To compare the diagnostic performance of DL with human readers and quantitative parameters.

Main Methods:

• A total of 1038 patients undergoing rest electrocardiogram (ECG)-gated SPECT and echocardiography were included.
• A DL model was trained using echocardiography as ground truth to predict abnormal wall motion and assess regional wall motion.
• A 10-fold cross-validation was employed, and DL performance was compared against human readers and quantitative parameters.

Main Results:

• The DL model demonstrated superior diagnostic performance compared to human readers and quantitative parameters, with higher area under the receiver operating characteristic curve (AUC) and accuracy (ACC).
• DL model achieved an AUC of .82 (95% CI: .79-.85) and ACC of .88, outperforming human readers (AUC: .77; ACC: .82) and quantitative parameters (AUC: .74; ACC: .78).
• The DL model provided results rapidly (within 30 seconds) via a user interface, enabling preliminary interpretations.

Conclusions:

• Deep learning (DL) effectively enhances the interpretation of rest SPECT wall motion.
• DL-based analysis surpasses the diagnostic capabilities of current human readers and quantitative parameter diagnoses for SPECT.
• The developed DL workflow offers a more accurate and efficient method for SPECT wall motion interpretation.