Original Article
Deep learning prediction of quantitative coronary angiography values using myocardial perfusion images with a CZT camera

https://doi.org/10.1007/s12350-022-02995-6Get rights and content

Abstract

Purpose

Evaluate the prediction of quantitative coronary angiography (QCA) values from MPI, by means of deep learning.

Methods

546 patients (67% men) undergoing stress 99mTc-tetrofosmin MPI in a CZT camera in the upright and supine position were included (1092 MPIs). Patients were divided into two groups: ICA group included 271 patients who performed an ICA within 6 months of MPI and a control group with 275 patients with low pre-test probability for CAD and a normal MPI. QCA analyses were performed using radiologic software and verified by an expert reader. Left ventricular myocardium was segmented using clinical nuclear cardiology software and verified by an expert reader. A deep learning model was trained using a double cross-validation scheme such that all data could be used as test data as well.

Results

Area under the receiver-operating characteristic curve for the prediction of QCA, with > 50% narrowing of the artery, by deep learning for the external test cohort: per patient 85% [95% confidence interval (CI) 84%-87%] and per vessel; LAD 74% (CI 72%-76%), RCA 85% (CI 83%-86%), LCx 81% (CI 78%-84%), and average 80% (CI 77%-83%).

Conclusion

Deep learning can predict the presence of different QCA percentages of coronary artery stenosis from MPIs.

Introduction

Myocardial perfusion imaging (MPI) is one of the most common cardiological examinations performed for the diagnosis of and risk assessment in patients with suspected coronary artery disease (CAD). It provides valuable information on ischemia, myocardial injuries, and left ventricular ejection fraction, among others. The technique has seen improvements in recent years with the introduction of cadmium-zinc-telluride (CZT) technology that enables one to perform the scan within a shorter duration,1 use low-dose radiotracer protocols,2,3 and achieve high diagnostic performance at the same time.4

Patients who are thought to have significant CAD in the MPI study are further examined and eventually treated by means of an invasive coronary angiography (ICA). The degree of coronary artery stenosis can be evaluated while performing ICA by means of a fractional flow reserve (FFR), instantaneous wave-free ratio (iFR), and/or quantitative coronary angiography (QCA), among others.5, 6, 7, 8 FFR is a hyperemic pressure-derived ratio that is considered the reference standard method for evaluating the functional severity of coronary stenosis based on substantial clinical outcome data. iFR is a nonhyperemic pressure-derived ratio, which correlates well with FFR and has similar coronary functional indexes. QCA obtains parameters from ICA that quantify objectively, measures the significance of a coronary stenosis in intervals, and expresses it in percentages (%), avoiding inter-observer variability and achieving reproducibility.

Since the pioneering work of Fujita et al.9 in the 1990s, artificial neural networks (ANNs) have been investigated for automatic evaluation of MPI images. ANNs and MPI technology have evolved, showing promising results for the diagnosis of CAD.10, 11, 12, 13 With deep learning, i.e., deep ANNs, and advanced neural network architectures more powerful algorithms can be trained. A shallow convolutional neural network (CNN) was used by Betancur et al. by utilizing MPI together with gender information for the automatic detection of CAD.11,12 In the recent study by Liu et al. the approach was extended by a larger dataset with 37,243 patients and a deeper neural network.10 While these efforts have been done for the diagnosis of CAD using ANNs and MPI, to the authors’ knowledge, there has been no attempt in the current scientific literature to predict QCA values from MPI. This is information which could give great aggregated value to the MPI study since only patients suspected to have a significant degree of stenosis will go further for ICA and re-vascularization.14 Inspired by the previous studies, a CNN has been designed and trained for the prediction of QCA values from MPI together with age, gender, and body mass index (BMI). Augmentation has been utilized to improve the generalization performance.

Using CNNs trained for regression, the aim of our study was to predict the degree of coronary artery stenosis (in percentage), similarly to that of QCA from ICA, from the main coronary artery regions – the left anterior artery (LAD), circumflex artery (LCx), and right coronary artery (RCA) – using stress MPIs in the upright and supine position with QCA from ICA as reference.

Section snippets

Study population

All the adult subjects referred for MPI from our center during the 1st of June, 2014 to the 30th of October, 2019 (N = 3658) were retrospectively reviewed and considered to be included in the study. The referral for stress testing was at the clinical discretion of the referring cardiologist. Subjects with a left-bundle branch block, congenital heart disease, and cardiac transplantation were excluded. Due to the risk of bronchospasm from regadenoson,15 all patients with known asthma or chronic

Data analysis

The characteristics of the included control group and the ICA group (consisting of patients who based on the MPI and/or clinical symptoms were thought to have CAD and therefore underwent ICA) are similar in terms of age, gender, BMI, and type of physical stress test (Table 1). Note that 20.3% of the ICA group did not have any vessel with QCA ≥ 50%. A majority of the vessels with QCA ≥ 50% did also have QCA ≥ 80% (Table 2).

Validation sets

The AUC for the prediction of QCA values for the two different QCA

Discussion

In 1992, Fujita et al.9 applied an ANN to the myocardial SPECT bull’s eye images and showed that the approach was useful for the computer-aided diagnosis of CAD, opening the path for further research in the field. In our study, we have applied ANNs to create an algorithm for the automatic prediction of QCA values from stress MPI polar maps and compared it with QCA values from ICA, demonstrating that deep learning achieves high diagnostic efficacy in predicting the QCA values. While other

New knowledge gained

We have applied deep learning and successfully created an algorithm for the automatic prediction of QCA values from stress MPI and demonstrated that deep learning achieves high diagnostic efficacy in predicting the percentage of coronary artery stenosis compared to QCA values from ICA. This work is, to our knowledge, the first study with deep learning that works with the prediction of QCA values from stress MPI.

Conclusion

Using CNNs, our algorithm can estimate the QCA values from MPI images and achieve very satisfactory results, compared to QCA values from ICA. More research is needed in order to establish the hemodynamic significance of the scintigraphic QCA, develop, and clinically validate this promising novel technology.

Disclosures

The authors have indicated that they have no financial conflict of interest.

Data availability

The de-identified data underlying the current manuscript can be shared upon reasonable request.

Funding

Grant support was obtained by Analytic Imaging Diagnostics Arena, Vinnova Grant 2017-02447, Department of Clinical Physiology and Department of Radiology, Region Östergötland.

Code availability

Not applicable.

Consent to participate

Written consent was waived for this retrospective analysis.

Ethical approval

This research study was conducted retrospectively from data obtained for clinical purposes. The retrospective study has approval by the Regional Ethics Review Board in Region Östergötland, Sweden (approval number 2019-00097).

Informed consent

Informed consent for this retrospective assessment was waived in accordance with the Review Board’s decision. All images and patient data used for this retrospective study were fully anonymized and coded in order to comply with the General Data Protection Regulation in the European Union 2016/679. Additionally, all patients from our institution are informed that their medical data can be rendered anonymous and used for scientific purposes.

Consent to participate

Written consent was waived for this retrospective analysis.

References (31)

  • GarroneP et al.

    Quantitative coronary angiography in the current era: Principles and applications

    J Interv Cardiol

    (2009)
  • EbiharaS et al.

    Rate pressure products affect the relationship between the fractional flow reserve and instantaneous wave-free ratio

    J Interv Cardiol

    (2020)
  • FujitaH et al.

    Application of artificial neural network to computer-aided diagnosis of coronary artery disease in myocardial SPECT bull’s-eye images

    J Nucl Med

    (1992)
  • LiuH et al.

    Diagnostic accuracy of stress-only myocardial perfusion SPECT improved by deep learning

    Eur J Nucl Med Mol Imaging

    (2021)
  • BetancurJ et al.

    Deep learning analysis of upright-supine high-efficiency SPECT myocardial perfusion imaging for prediction of obstructive coronary disease: A multicenter study

    J Nucl Med

    (2019)

    • Cited by (1)

    • Diagnostic performance of a novel deep learning attenuation correction software for MPI using a cardio dedicated CZT camera: Experience in the clinical practice

      2023, Revista Espanola de Medicina Nuclear e Imagen Molecular

    The authors of this article have provided a PowerPoint file, available for download at SpringerLink, which summarizes the contents of the paper and is free for re-use at meetings and presentations. Search for the article DOI on SpringerLink.com.

    The authors have also provided an audio summary of the article, which is available to download as ESM, or to listen to via the JNC/ASNC Podcast.

    View full text
    Copyright © 2023 American Society of Nuclear Cardiology. Published by ELSEVIER INC. All rights reserved. Published by Mosby, Inc. All rights reserved.