Machine Learning Algorithms to Distinguish Myocardial Perfusion SPECT Polar Maps

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Myocardial perfusion imaging (MPI) is crucial for diagnosing coronary artery disease (CAD).
• Machine learning (ML) shows significant potential in medical image analysis.
• Automated classification of MPI scans can aid clinical decision-making.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in classifying normal versus abnormal gated Single Photon Emission Tom Tomography (SPECT) MPI polar maps.
• To compare the performance of different ML models for CAD detection using MPI data.

Main Methods:

• Analysis of 1007 gated SPECT MPI polar maps from patients with suspected or documented CAD.
• Feature extraction from polar map segmentation (horizontal and vertical slices).
• Evaluation of four ML models using cross-validation, data augmentation, and expert reader consensus as the gold standard.

Main Results:

• All evaluated ML models achieved accuracy >90% and Area Under the Curve (AUC) >0.80, except for Adaptive Boosting (AUC=0.77).
• Precision and sensitivity were consistently high, exceeding 96% and 92%, respectively.
• The Random Forest model exhibited the best performance with an AUC of 0.853, accuracy of 0.938, precision of 0.968, and sensitivity of 0.963.

Conclusions:

• ML algorithms demonstrate high performance in classifying MPI polar maps.
• These models are effective in distinguishing normal from abnormal scans, supporting CAD diagnosis.
• ML offers a promising tool for enhancing the interpretation of MPI in clinical practice.