Myocardial Iron Overload Assessment with Automatic Segmentation of Cardiac MR Images based on Deep Neural Networks

Articles in Press

Document Type : Original Paper

Authors

1 Department of Medical Informatics, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran

2 Institute of Bioinformatics, University Medicine Greifswald, Greifswald, Germany.

3 Department of Radiology, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran

10.22038/ijmp.2024.77319.2362

Abstract

Introduction: Heart failure due to myocardial iron overload is one of the main causes of death in thalassemia major (TM) patients. Therefore, cardiac magnetic resonance (CMR) imaging method with a multi-echo sequence can be used to assess the iron overload of TM patients.

Materials and Methods: This study aimed to evaluate the myocardial iron overload in TM patients with automatic left ventricular (LV) segmentation of CMR images. Thirty-six TM patients were selected to acquire CMR images and clinical data. Automatic LV segmentation was implemented with U-Net, an automatically adapted deep convolutional neural network based on U-Net. With the signal intensity of the LV segmented area, T2* value can be calculated at different echo times, a widely used and approved method to assess myocardial iron overload.

Results: The accuracy of LV segmentation as measured by intersection-over-union (0.95) was substantially higher than non-deep learning based methods and at par with other deep learning based methods like . In addition, our results indicate that the proposed method outperformed in assessing LV iron overload over other deep learning based methods in terms of negative predictive value, positive predictive value, and Jaccard.

Conclusion: Relying on these outcomes, the proposed method as a deep learning based model yields better LV segmentation and notably impacts assessing myocardial iron overload.

Keywords

Main Subjects

Iranian Journal of Medical Physics

Articles in Press, Accepted Manuscript
Available Online from 16 March 2024

Files

History

  • Receive Date: 06 January 2024
  • Revise Date: 08 March 2024
  • Accept Date: 16 March 2024

Share

How to cite

Statistics