Classification models for SPECT myocardial perfusion imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Myocardial ischemia and infarction detection is crucial for cardiovascular health.
• Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) is a key diagnostic tool.
• Automated analysis of SPECT MPI can improve diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate computer-aided classification models for SPECT MPI.
• To identify perfusion abnormalities, specifically myocardial ischemia and infarction.
• To compare the performance of deep learning and knowledge-based models.

Main Methods:

• Two classification models were developed: a deep learning (DL) model using transfer learning and a knowledge-based model.
• The DL model employed pre-trained deep neural networks and a support vector machine classifier.
• The knowledge-based model translated expert reader knowledge into image processing techniques.
• A dataset of 192 patients' summed stress and rest SPECT MPI images was utilized, with expert reader assessments serving as the reference standard.

Main Results:

• The DL-based model achieved maximum accuracy of 94%, sensitivity of 88%, and specificity of 100%.
• The knowledge-based model achieved maximum accuracy of 93%, sensitivity of 100%, and specificity of 86%.
• Both models demonstrated high diagnostic performance.

Conclusions:

• The developed computer-aided classification models exhibit diagnostic performance comparable to expert analysis.
• These models can assist clinicians in decision-making for SPECT MPI interpretation.
• The findings support the use of AI in identifying myocardial ischemia and infarction from SPECT MPI data.