An accurate and time-efficient deep learning-based system for automated segmentation and reporting of cardiac magnetic resonance-detected ischemic scar.

Autor: Papetti DM; Department of Informatics, Systems and Communication, University of Milano-Bicocca, Viale Sarca 336, Milano 20126, Italy., Van Abeelen K; Department of Medicine and Surgery, University of Milano-Bicocca, Milan 20126, Italy., Davies R; Institute of Cardiovascular Science, University College London, London WC1E 6DD, UK; Barts Heart Centre, St Bartholomew's Hospital, West Smithfield, London EC1A 7BE, UK; MRC Unit for Lifelong Health and Ageing, University College London, London WC1E 6DD, UK., Menè R; Department of Medicine and Surgery, University of Milano-Bicocca, Milan 20126, Italy; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy., Heilbron F; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy., Perelli FP; Department of Medicine and Surgery, University of Milano-Bicocca, Milan 20126, Italy; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy., Artico J; Institute of Cardiovascular Science, University College London, London WC1E 6DD, UK; Barts Heart Centre, St Bartholomew's Hospital, West Smithfield, London EC1A 7BE, UK., Seraphim A; Institute of Cardiovascular Science, University College London, London WC1E 6DD, UK; Department of Cardiac Electrophysiology, Barts Heart Centre, Barts Health NHS Trust, London EC1A 7BE, UK., Moon JC; Institute of Cardiovascular Science, University College London, London WC1E 6DD, UK; Barts Heart Centre, St Bartholomew's Hospital, West Smithfield, London EC1A 7BE, UK., Parati G; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy., Xue H; National Heart, Lung, and Blood Institute, National Institutes of Health, DHHS, Bethesda, MD, USA. Electronic address: hui.xue@nih.gov., Kellman P; Bicocca Bioinformatics Biostatistics and Bioimaging Centre (B4), University of Milano-Bicocca, Vedano al Lambro 20854, Italy., Badano LP; Department of Medicine and Surgery, University of Milano-Bicocca, Milan 20126, Italy; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy., Besozzi D; Department of Informatics, Systems and Communication, University of Milano-Bicocca, Viale Sarca 336, Milano 20126, Italy; Bicocca Bioinformatics Biostatistics and Bioimaging Centre (B4), University of Milano-Bicocca, Vedano al Lambro 20854, Italy. Electronic address: daniela.besozzi@unimib.it., Nobile MS; Bicocca Bioinformatics Biostatistics and Bioimaging Centre (B4), University of Milano-Bicocca, Vedano al Lambro 20854, Italy; Department of Environmental Sciences, Informatics and Statistics, Ca' Foscari University of Venice, Via Torino 155, Mestre, Venice 30172, Italy. Electronic address: marco.nobile@unive.it., Torlasco C; Department of Cardiology, IRCCS Istituto Auxologico Italiano, Via Magnasco 2, Milan 20145, Italy. Electronic address: c.torlasco@auxologico.it.
Jazyk: angličtina
Zdroj: Computer methods and programs in biomedicine [Comput Methods Programs Biomed] 2023 Feb; Vol. 229, pp. 107321. Date of Electronic Publication: 2022 Dec 20.
DOI: 10.1016/j.cmpb.2022.107321
Abstrakt: Background and Objectives: Myocardial infarction scar (MIS) assessment by cardiac magnetic resonance provides prognostic information and guides patients' clinical management. However, MIS segmentation is time-consuming and not performed routinely. This study presents a deep-learning-based computational workflow for the segmentation of left ventricular (LV) MIS, for the first time performed on state-of-the-art dark-blood late gadolinium enhancement (DB-LGE) images, and the computation of MIS transmurality and extent.
Methods: DB-LGE short-axis images of consecutive patients with myocardial infarction were acquired at 1.5T in two centres between Jan 1, 2019, and June 1, 2021. Two convolutional neural network (CNN) models based on the U-Net architecture were trained to sequentially segment the LV and MIS, by processing an incoming series of DB-LGE images. A 5-fold cross-validation was performed to assess the performance of the models. Model outputs were compared respectively with manual (LV endo- and epicardial border) and semi-automated (MIS, 4-Standard Deviation technique) ground truth to assess the accuracy of the segmentation. An automated post-processing and reporting tool was developed, computing MIS extent (expressed as relative infarcted mass) and transmurality.
Results: The dataset included 1355 DB-LGE short-axis images from 144 patients (MIS in 942 images). High performance (> 0.85) as measured by the Intersection over Union metric was obtained for both the LV and MIS segmentations on the training sets. The performance for both LV and MIS segmentations was 0.83 on the test sets. Compared to the 4-Standard Deviation segmentation technique, our system was five times quicker (<1 min versus 7 ± 3 min), and required minimal user interaction.
Conclusions: Our solution successfully addresses different issues related to automatic MIS segmentation, including accuracy, time-effectiveness, and the automatic generation of a clinical report.
Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2022. Published by Elsevier B.V.)
Databáze: MEDLINE
Externí odkaz: