Recent study employs machine learning for coronary artery disease assessment

TL;DR:

  • Recent study explores the potential of machine learning (ML) in diagnosing coronary artery disease (CAD) using myocardial perfusion imaging (MPI) SPECT scans.
  • ML models, featuring various algorithms, significantly enhance CAD risk assessment and diagnosis.
  • Stress features stand out as key contributors, outperforming other feature sets.
  • Boruta-gradient boosting model excels in CAD risk stratification.
  • ML-based automation reduces time-consuming analysis and enhances interpretability for clinicians.
  • Future studies may refine models for different stress techniques and broaden clinical factors for enhanced applicability.

Main AI News:

In the ever-evolving landscape of healthcare, the intersection of technology and medicine has opened new avenues for diagnosing and managing critical conditions. A recent study, as reported in Scientific Reports, delves into the promising realm of machine learning (ML) for assessing coronary artery disease (CAD) and its susceptibility. This research leverages the power of ML algorithms to evaluate radiomic features through myocardial perfusion imaging (MPI) single-photon emission computed tomography (SPECT) scans, shedding light on a potentially groundbreaking approach to combat one of the leading causes of global morbidity and mortality—cardiovascular diseases (CVD).

Elevating CAD Diagnosis with Radiomics and ML

CAD, notorious for its lethal impact, necessitates early detection and tailored interventions. Traditional methods for CAD assessment, such as observer-dependent optical evaluation of MPI SPECT scans, are not only error-prone but also time-consuming. Hence, the clamor for automated, objective approaches in assessing cardiac MPI SPECT has never been more urgent.

The research at hand focuses on harnessing ML to diagnose CAD, specifically through the analysis of MPI-SPECT images. The study meticulously evaluates the performance of various ML models, dissecting delta, stress, and rest MPI SPECT radiomics to ascertain CAD diagnosis and risk classification accuracy.

The Crucial Role of ML Models and Feature Selection

The researchers embark on a rigorous journey, testing the prowess of nine ML-based algorithms, including gradient boosting, extreme gradient boosting, K-nearest neighbor, decision tree, multi-layer perceptron, random forest, logistic regression, support vector machine, and Naive Bayes. Feature selection, a pivotal component of this analysis, is executed through three distinct methods: Maximum Relevance Minimum Redundancy (mRMR), Recursive Feature Elimination using the Random Forest classifier (RF-RFE), and Boruta.

The study’s cohort encompasses 395 individuals suspected of CAD, all subjected to a 48-hour rest-stress MPI SPECT. It is noteworthy that this population excludes individuals with a history of myocardial infarction. Among these participants, 78 exhibited normal conditions, while 317 individuals stood at varying risk levels for CAD. This latter group was further stratified into low-, intermediate-, and high-risk categories, with the left ventricular myocardium being manually delineated on the MPI-SPECT scans. Stress induction techniques included dobutamine, dipyridamole, and exercise.

Radiomics Unveiled: Transforming Scans into Insights

Beyond clinical variables such as family history, age, gender, smoking habits, ejection fraction, and diabetes mellitus status, the research extracted a trove of 118 radiomic features from the scans. These features, organized into sets that encompass stress, delta, rest, and a combination of them, were processed in accordance with the image biomarker initiative standardization (IBSI) and assessed using the Standardized Environment for Radiomics Analysis (SERA) protocol.

To train and validate their model, 80% of the data was allocated for training purposes, while the remaining 20% was reserved for validation. The model’s performance was meticulously scrutinized in two vital tasks: differentiating between normal (absence of CAD) and abnormal (presence of CAD) conditions and classifying low and high-risk scenarios. The evaluation employed key metrics, including the area under the receiver operating characteristic curve (AUC), specificity (SPE), accuracy (ACC), and sensitivity (SEN).

Precision and Consensus in Analysis

To ensure the integrity of the findings, the data underwent analysis by two nuclear medicine physicians. In instances of disagreement, a consensus was reached through collaboration or consultation with a senior physician. These medical experts also had access to conventional SPECT scores, such as the summed stress score (SSS), summed rest score (SRS), summed difference score (SDS), and data related to wall thickening and motion.

A Glimpse into the Results

The results illuminate the potential of stress features in CAD diagnosis, outshining other feature sets. Notably, the mRMR-KNN classifier within the Stress-feature set exhibited superior performance in the first task, boasting SPE, SEN, ACC, and AUC values of 0.6, 0.64, 0.63, and 0.61, respectively.

In the second task, where CAD risk stratification took center stage, the Boruta-gradient boosting model emerged as the frontrunner. With SPE, SEN, ACC, and AUC values of 0.76, 0.75, 0.76, and 0.79, respectively, this model showcased remarkable promise. Furthermore, the study identified key contributors to CAD risk classification, including non-uniformity-normalized dependence counts from the neighboring grey level dependence matrix (NGLDM) family and the status of diabetes mellitus from clinical parameters.

Unveiling the Implications

The implications of this study are profound. By harnessing the capabilities of machine learning models in analyzing MPI-SPECT images, the research opens new horizons in CAD diagnosis and risk assessment. These models have the potential to revolutionize the labor-intensive MPI SPECT analysis, providing clinicians with invaluable insights into diagnosis factors and augmenting trust in AI-powered automated models.

Future endeavors should explore refining the model for various stress-induction techniques and broaden the scope to include patients with myocardial infarction and other CAD-related clinical factors like body mass index (BMI) and hyperlipidemia. This pursuit of knowledge promises to enhance the applicability and reach of these groundbreaking findings, ultimately reshaping the landscape of CAD diagnosis and management.

Conclusion:

The integration of machine learning into CAD diagnosis through radiomics and MPI-SPECT scans holds immense potential to transform the market. With the ability to automate and enhance accuracy in CAD risk assessment, these models promise to streamline diagnostic processes, providing clinicians with valuable insights. Furthermore, expanding the scope to include various stress-induction techniques and a wider range of clinical factors will likely increase the market’s competitiveness, making AI-powered CAD diagnosis a crucial player in cardiovascular healthcare.

Source