Expanding interpretability through complexity reduction in machine learning-based modelling of cardiovascular disease: A myocardial perfusion imaging PET/CT prognostic study - open access original article

Aug 5, 2026

Upload by Rowan

Machine learning can enhance risk stratification in myocardial perfusion imaging by improving prediction of major adverse cardiovascular events in patients with suspected or established coronary artery disease. A new, robust method is presented to identify patients who develop MACE, designed to resist the harmful influence of outliers in training. The approach averages variable contributions across an ensemble of XGBoost models to form a sum-of-sigmoids predictor using rest and adenosine-stress 13N-ammonia PET/CT polar maps. Across 1185 studies with 2-year follow-up, it achieves 0.83 accuracy on the test set, matching more complex models while remaining interpretable for automated clinical risk stratification.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Problem in cardiovascular risk stratification and interpretability

Role of PET myocardial perfusion imaging

Prior machine learning approaches and limitations

Methods (from abstract)

Sum-of-sigmoids ensemble modeling approach

Imaging inputs and study design

Results and Conclusion (from abstract)

Prediction accuracy and comparators

Clinical suitability of interpretable risk stratification

Received: 7 November 2024 | Accepted: 9 January 2025 DOI: 10. 1111/eci.14391 ORIGINAL ARTICLE Expanding interpretability through complexity reduction in machine learning-based modelling of cardiovascular disease: A myocardial perfusion imaging PET/CT prognostic study Eero Lehtonen1 | Jarmo Teuho1 | Monire Vatandoust1 | Juhani Knuuti1 | Remco J. J. Knol2 | Friso M. van der Zant2 | Luis Eduardo Juárez-Orozco1,3 | Riku Klén1 1Turku PET Centre, Turku University Hospital and University of Turku, Turku, Finland 2Cardiac Imaging Division Alkmaar, Department of Nuclear Medicine, Northwest Clinics, Alkmaar, The Netherlands 3Department of Cardiology, Division Heart & Lungs, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands Correspondence Luis Eduardo Juárez-Orozco, Department of Cardiology, Division Heart & Lungs, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands. Email: [ljuarez2@umcutrecht.nl](ljuarez2@umcutrecht.nl) Funding information Suomen Kulttuurirahasto; State Research Funding of Turku University Hospital, Grant/Award Number: 30046; Research Council of Finland, Grant/ Award Number: 351482 Abstract Background: Machine learning-based analysis can be used in myocardial perfusion imaging data to improve risk stratification and the prediction of major adverse cardiovascular events for patients with suspected or established coronary artery disease. We present a new machine learning approach for the identification of patients who develop major adverse cardiovascular events. The new method is robust against the deleterious effect of outliers in the training set stratification and training process. Methods: The proposed sum-of-sigmoids model is obtained by averaging the contributions of various input variables in an ensemble of XGBoost models. To illustrate its performance, we have applied it to predict major adverse cardiovascular events from advanced imaging data extracted from rest and adenosine stress 13N-ammonia positron emission tomography myocardial perfusion imaging polar maps. There were 1185 individual studies performed, and the event occurrence was tracked over a follow-up period of 2 years. Results: The sum-of-sigmoids model achieved a prediction accuracy of .83 on the test set, matching the performance of significantly more complex and less interpretable models (whose accuracies were .83–.84) . Conclusion: The sum-of-sigmoids model is interpretable and simple, while achieving similar prediction accuracy to significantly more complex machine learning models in the considered prediction task. It should be suitable for applications such as automated clinical risk stratification, where clear and explicit justification of the classification procedure is highly pertinent. KEYWORDS data analytics, interpretability, machine learning, myocardial perfusion imaging, positron emission tomography This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). European Journal of Clinical Investigation published by John Wiley & Sons Ltd on behalf of Stichting European Society for Clinical Investigation Journal Foundation. 2 of 10 LEHTONEN et al. 1 | INTRODUCTION Chronic coronary syndromes represent the most common cause of death globally.1,2 Despite substantial developments in advanced cardiac imaging, accurate risk stratification and therefore prediction of acute events (major adverse cardiovascular events (MACE)) derived from myocardial ischemia, thromboembolism, heart decompensation and cardiac death remains a major challenge for clinicians and an unmet need in cardiology.1 Positron emission tomography (PET) myocardial perfusion imaging (MPI) enables quantitative assessment of myocardial blood flow (MBF) and flow reserve (MFR), thus allowing to detect functionally s

FAQs

What is the study’s main goal in cardiovascular risk prediction?

To develop a machine learning approach that identifies patients who develop major adverse cardiovascular events using PET myocardial perfusion imaging data, while improving interpretability and robustness.

How does the proposed sum-of-sigmoids model work?

It is built by averaging the contributions of multiple input variables across an ensemble of XGBoost models, producing a highly interpretable decision rule.

How well does the model perform compared with more complex methods?

It reaches a test-set prediction accuracy of about 0.83, matching the performance of significantly more complex and less interpretable models, with accuracies in the 0.83–0.84 range.

Aug 5, 2026

Upload by Rowan