Evaluation of myocardial perfusion imaging techniques and artificial intelligence (AI) tools in coronary artery...

Cardiovascular diseases represent the primary cause of global mortality, necessitating highly accurate diagnostic protocols for coronary artery disease. Prior research has shown that myocardial perfusion imaging (MPI) serves as a fundamental pillar for assessing cardiac function and identifying ischemic regions. It was already known that nuclear techniques like single photon emission computed tomography provide critical insights into blood flow through the heart muscle. Traditional diagnostic workflows often rely on subjective interpretation or single-metric evaluations, which may overlook the complex interplay between diagnostic accuracy and patient safety. The integration of machine learning and deep learning architectures has recently emerged to refine these imaging interpretations. However, a systematic comparison across diverse imaging modalities and algorithmic frameworks using multi-criteria frameworks remained largely absent. This absence of evidence motivated the current investigation into how specific performance metrics influence the selection of diagnostic tools.

This investigation employs a novel multi-criteria decision-making framework to determine how various parameters and performance metrics dictate the prioritization of imaging modalities. Researchers sought to identify the most effective combinations of artificial intelligence (AI) tools and myocardial perfusion imaging (MPI) techniques to enhance coronary artery disease (CAD) detection rates. The study specifically targets the comparative effectiveness of convolutional neural networks and classical machine learning models. Another primary objective involves weighing the trade-offs between nuclear and non-nuclear imaging approaches regarding radiation exposure and cost. By establishing a hierarchical ranking, the work provides a roadmap for clinical adoption of automated diagnostic aids. The analysis aims to bridge the gap between raw algorithmic performance and practical clinical utility in cardiology. This systematic approach ensures that the selection of diagnostic tools is based on a holistic evaluation of technical and economic factors.

The research team utilized the fuzzy-based preference ranking organization method for enrichment evaluation (PROMETHEE) to analyze complex diagnostic datasets. This multi-criteria decision-making (MCDM) method facilitated the comparison of nuclear techniques like positron emission tomography (PET) and single photon emission computed tomography (SPECT) against cardiovascular magnetic resonance (CMR) imaging. Evaluated computational models included deep learning architectures such as InceptionV3, VGG16, ResNet50, and DenseNet121. Parallel assessments were conducted on traditional classifiers including random forests (RF), K-nearest neighbor (KNN), support vector machine (SVM), and Naïve Bayes (NB). Performance was quantified using a comprehensive suite of metrics including F1-score, recall, specificity, precision, accuracy, and the area under the receiver operating characteristic curve (AUC-ROC). For the imaging modalities, criteria encompassed sensitivity, radiation dose, scan expenditure, and total study duration. The researchers applied these criteria to rank each combination of technology and algorithm to determine the most efficient diagnostic pathway.

Random forests (RF) emerged as the most efficient artificial intelligence (AI) tool for single photon emission computed tomography (SPECT) with a net flow value of 0.3778. Cardiovascular magnetic resonance (CMR) imaging proved to be the superior imaging modality for coronary artery disease (CAD) diagnosis, achieving a net flow of 0.3666. Comparative analysis indicated that cardiovascular magnetic resonance (CMR) imaging consistently outperformed both positron emission tomography (PET) and single photon emission computed tomography (SPECT) across most criteria. Single photon emission computed tomography (SPECT) was identified as the least advantageous technique, scoring below average on all metrics except for scan cost. The data suggests that integrating the random forest (RF) algorithm could significantly elevate the diagnostic utility of single photon emission computed tomography (SPECT). These findings highlight a distinct hierarchy in both algorithmic efficacy and imaging precision within the context of myocardial assessment. The results emphasize that non-nuclear methods offer significant advantages in terms of radiation safety and diagnostic accuracy.

The study demonstrates that non-nuclear imaging options provide a more balanced profile of safety and accuracy than traditional nuclear methods. Implementing multi-criteria decision-making (MCDM) methods allows for a more nuanced selection of diagnostic tools tailored to specific clinical constraints. Future research should focus on expanding these evaluations to include a broader range of clinical populations and emerging neural network architectures. The researchers suggest that the strategic pairing of specific machine learning models with existing imaging hardware can mitigate inherent modality limitations. Enhancing the diagnostic rate of coronary artery disease (CAD) depends on the adoption of these optimized, data-driven prioritization strategies. Ultimately, these results provide a quantitative foundation for updating clinical guidelines regarding myocardial perfusion imaging and artificial intelligence integration. This work paves the way for more personalized and efficient cardiovascular diagnostic workflows in modern healthcare settings.