Skip to Main content Skip to Navigation
Master Thesis

Caractérisation de l’infarctus du myocarde chronique par analyse automatique d’images de scanner cardiaque à la phase artérielle : corrélations cliniques et validation par comparaison à l’IRM

Abstract : Background: myocardial wall thinning and fat as mapped from CT was shown to correlate with scar location and sites responsible for arrhythmias in patients undergoing catheter ablation for ventricular tachycardia. However, image processing remains both time consuming and operator-dependent. Objectives: to develop an automated method for left ventricular (LV) wall segmentation allowing for thickness and fat mapping from CT images, and to report on its correlations with clinical characteristics and scar distribution as assessed by CMR. Methods: a deep learning algorithm was trained on a database of 500 CT scans with available ground truth LV wall segmentations. From wall segmentations, wall thickness and intra-mural fat were automatically quantified. This method was applied on an additional population of 78 patients with history of myocardial infarction and with both cardiac CT and late gadolinium-enhanced CMR available. CMR images were manually segmented to assess total scar volume and the distribution of scar transmurality. The accuracy of automated CT segmentation was reported, and its inter-scan reproducibility was analyzed in a small subset of patients who underwent 2 serial CT acquisitions. The clinical correlates of wall thinning and fat were analyzed. CT and CMR were registered to study the relationship between thickness and scar transmurality at the segmental level. Results: the deep learning method for automated segmentation showed millimetric accuracy and sub-millimetric inter-scan reproducibility for the measurement of LV wall thickness. When applied to our population of 78 ischemic patients with both CT and CMR available (age 63±14, 85% men, LVEF 47±13%), automated quantifications of wall thinning and intramural fat closely related to scar volume on CMR, as well as to ventricular arrhythmias. Unlike scar volume on CMR, wall thinning and fat burden on CT were markers of infarct ageing. The area of severe thinning (<2mm) was the imaging characteristic the most closely associated with arrhythmias, independently from LVEF or scar volume on CMR. The regional distribution of wall thinning matched the distribution of scar transmurality in scars older than 5 years, while this association was weak or absent in more recent infarcts. Conclusions: the proposed method allows for accurate and extremely reproducible LV thickness measurements. Such CT characterization brings additional value over CMR for the assessment of post-infarction scar. By grasping structural features associated with infarct ageing, CT documents the maturation of scar towards arrhythmogenicity. This opens promising perspectives for the stratification of risk after myocardial infarction, and for the guidance of ablation therapy in patients with post-infarction tachycardia.
Complete list of metadatas

Cited literature [168 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02427157
Contributor : Médecine Ub <>
Submitted on : Friday, January 3, 2020 - 12:07:09 PM
Last modification on : Saturday, January 18, 2020 - 1:30:24 AM
Long-term archiving on: : Monday, April 6, 2020 - 6:17:30 PM

File

 Restricted access
To satisfy the distribution rights of the publisher, the document is embargoed until : 2021-06-14

Please log in to resquest access to the document

Identifiers

  • HAL Id : dumas-02427157, version 1

Collections

Citation

Clémence Scappaticci. Caractérisation de l’infarctus du myocarde chronique par analyse automatique d’images de scanner cardiaque à la phase artérielle : corrélations cliniques et validation par comparaison à l’IRM. Sciences du Vivant [q-bio]. 2019. ⟨dumas-02427157⟩

Share

Metrics

Record views

29