In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.

The free consultation period for this content is over.

It is now only available year-round to ESC Professional Members, Fellows of the ESC, and Young combined Members

The prognostic value of deep learning in PET myocardial perfusion for cardiovascular events

Session Nuclear perfusion imaging: "back to the future"

Speaker Luis Eduardo Juarez-Orozco

Event : ESC Congress 2018

  • Topic : imaging
  • Sub-topic : Positron Emission Tomography (PET)
  • Session type : Advances in Science

Authors : L E Juarez-Orozco (Turku,FI), RJJ Knol (Alkmaar,NL), O Martinez-Manzanera (Groningen,NL), FM Van Der Zant (Alkmaar,NL), J Knuuti (Turku,FI)

L.E. Juarez-Orozco1 , R.J.J. Knol2 , O. Martinez-Manzanera3 , F.M. Van Der Zant2 , J. Knuuti1 , 1Turku University Hospital, PET Center - Turku - Finland , 2Cardiac Imaging Division Alkmaar, Northwest Clinics, Department of Nuclear Medicine - Alkmaar - Netherlands , 3University Medical Center Groningen, Neurology - Groningen - Netherlands ,

European Heart Journal ( 2018 ) 39 ( Supplement ), 269

Background: Deep Learning (DL) (i.e. deep artificial neural networks) is an innovative approach to explore and learn complex patterns within data. Cardiovascular image recognition and classification represent well suited purposes for DL implementation. PET myocardial perfusion polar maps provide a topological summary of absolute myocardial perfusion measurements across the left ventricle in patients with known or suspected coronary artery disease (CAD). As we envision the generation of parallel machine learning-based systems that aid in the characterization of the myocardial ischemic burden and risk in individual patients, we aimed to utilize a pre-trained very-deep convolutional neural network (CNN) modified through transfer learning for the identification of patients who experience major adverse cardiovascular events (MACE) based on direct image processing of PET myocardial perfusion polar maps.

Methods: We analyzed data consisting of polar maps depicting myocardial perfusion reserve from 1,199 patients who underwent 13N-ammonia PET for suspected ischemia. DL was built through transfer learning by obtaining the architecture of the open-source pre-trained ResNet-50 CNN and replacing the last layer and associated weights with a new layer specialized for the classification of patients who experienced or not MACE (i.e. myocardial infarction [MI], PCI, cardiac death or heart failure) throughout follow-up using a 9:1 training to testing ratio and 5-fold cross validation. Performance was evaluated through accuracy, precision, recall, specificity and likelihood ratios (LR).

Results: There were 575 men and 625 women (mean age of 68±9 years) with a documented mean follow-up of 13.5±7 months. There were 27% of patients with positive family history for CAD, 16% with a previous MI, 14% were smokers, 16% had diabetes, 33% had dyslipidemia and 51% had arterial hypertension. Overall incidence of MACE was 13%. The modified ResNet-50 CNN processed the input polar maps and demonstrated an overall cross-validated validation accuracy of 77% with a corresponding precision and recall of 90.3% and 72% respectively, and a specificity of 87% in the identification of patients who experienced a MACE within follow-up. The positive LR was 5.51, while the negative LR was 0.32.

Conclusion: Implementation of DL in direct image processing of quantitative PET myocardial perfusion polar maps may convey prognostic value through the classification of patients that will or not present a MACE during follow-up. DL seems to perform better in the identification of patients who will be free of events at follow-up. Further research into the clinical value of considering DL estimations in the comprehensive evaluation of patients suspected with myocardial ischemia is warranted.

The free consultation period for this content is over.

It is now only available year-round to ESC Professional Members, Fellows of the ESC, and Young combined Members

Members get more

Join now
  • 1ESC Professional Members – access all resources from general ESC events 
  • 2ESC Association Members (Ivory, Silver, Gold) – access your Association’s resources
  • 3Under 40 or in training - with a Combined Membership, access all resources
Join now

Our sponsors

ESC 365 is supported by Bayer, Boehringer Ingelheim and Lilly Alliance, Bristol-Myers Squibb and Pfizer Alliance, Novartis Pharma AG and Vifor Pharma in the form of educational grants. The sponsors were not involved in the development of this platform and had no influence on its content.

logo esc

Our mission: To reduce the burden of cardiovascular disease

Who we are