Fully automated, quality-controlled cardiac analysis from CMR: validation and large-scale application to characterize cardiac function

Jacobus Bernardus Ruijsink; Esther Puyol Anton; Ilkay Oksuz; Matthew Sinclair; Wenja Bai; Julia Anne Schnabel; Reza Razavi; Andrew Peter King

doi:10.1016/j.jcmg.2019.05.030

Fully automated, quality-controlled cardiac analysis from CMR: validation and large-scale application to characterize cardiac function

Jacobus Bernardus Ruijsink, Esther Puyol Anton, Ilkay Oksuz, Matthew Sinclair, Wenja Bai, Julia Anne Schnabel, Reza Razavi, Andrew Peter King

Imperial College London

Research output: Contribution to journal › Article › peer-review

142 Citations (Scopus)

306 Downloads (Pure)

Abstract

Objectives:Develop a fully automated framework for cardiac function analysis from cardiacmagnetic resonance (CMR), including comprehensive quality control (QC) algorithmsto detect erroneous output.Background:Analysis of cine CMR imaging using deep learning (DL) algorithms could automateventricular function assessment. However, variable image quality, variability inphenotypes of disease and unavoidable weaknesses in training of DL algorithmscurrently prevent their use in clinical practice.Methods:The framework consists of a pre-analysis DL image QC, followed by a DL algorithmfor biventricular segmentation in long- and short-axis, myocardial feature-tracking(FT) and a post-analysis QC to detect erroneous results. We validated the frameworkin healthy subjects and cardiac patients by comparison against manual analysis(n=100) and evaluation of the QC steps’ ability to detect erroneous results (n=700).Next, we utilized our method to obtain reference values for cardiac function metricsfrom the UK Biobank.Results:Automated analysis correlated highly with manual analysis for left and rightventricular volumes (all r>0.95), strain (circumferential: r=0.89, longitudinal: r>0.89)and filling and ejection rates (all r≥0.93). There was no significant bias for cardiacvolumes and filling and ejection rates, except for RVESV (bias +1.80 mL, p=.01).The bias for FT-strain was <1.3%. The sensitivity of detection of erroneous outputwas 95% for volume-derived parameters and 93% for FT strain. Finally, referencevalues were automatically derived from 2,029 CMR exams in healthy subjects.Conclusions:We demonstrate a DL-based framework for automated, quality-controlledcharacterization of cardiac function from cine CMR, without the need for directclinician oversight.

Original language	English
Pages (from-to)	684-695
Journal	JACC Cardiovascular Imaging
Volume	13
Issue number	3
Early online date	17 Jul 2019
DOIs	https://doi.org/10.1016/j.jcmg.2019.05.030
Publication status	Published - 2019

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1016/j.jcmg.2019.05.030Licence: CC BY

Fully Automated, Quality-Controlled_RUJISINK_Acc8Apr2019Epub17Jul2019_GOLD VoR (CC BY)Proof, 996 KBLicence: CC BY

Embargoed Document

Fully automated, quality-controlled_RUIJSINK_Accepted8April2019_GREEN AAM
Accepted author manuscript, 2.25 MB
Licence: CC BY-NC-ND
Embargo ends: 1/01/2099

Cite this

@article{104b135bbf3f41dd94fdfb835c5e1b13,

title = "Fully automated, quality-controlled cardiac analysis from CMR: validation and large-scale application to characterize cardiac function",

abstract = "Objectives:Develop a fully automated framework for cardiac function analysis from cardiacmagnetic resonance (CMR), including comprehensive quality control (QC) algorithmsto detect erroneous output.Background:Analysis of cine CMR imaging using deep learning (DL) algorithms could automateventricular function assessment. However, variable image quality, variability inphenotypes of disease and unavoidable weaknesses in training of DL algorithmscurrently prevent their use in clinical practice.Methods:The framework consists of a pre-analysis DL image QC, followed by a DL algorithmfor biventricular segmentation in long- and short-axis, myocardial feature-tracking(FT) and a post-analysis QC to detect erroneous results. We validated the frameworkin healthy subjects and cardiac patients by comparison against manual analysis(n=100) and evaluation of the QC steps{\textquoteright} ability to detect erroneous results (n=700).Next, we utilized our method to obtain reference values for cardiac function metricsfrom the UK Biobank.Results:Automated analysis correlated highly with manual analysis for left and rightventricular volumes (all r>0.95), strain (circumferential: r=0.89, longitudinal: r>0.89)and filling and ejection rates (all r≥0.93). There was no significant bias for cardiacvolumes and filling and ejection rates, except for RVESV (bias +1.80 mL, p=.01).The bias for FT-strain was <1.3%. The sensitivity of detection of erroneous outputwas 95% for volume-derived parameters and 93% for FT strain. Finally, referencevalues were automatically derived from 2,029 CMR exams in healthy subjects.Conclusions:We demonstrate a DL-based framework for automated, quality-controlledcharacterization of cardiac function from cine CMR, without the need for directclinician oversight.",

author = "Ruijsink, {Jacobus Bernardus} and {Puyol Anton}, Esther and Ilkay Oksuz and Matthew Sinclair and Wenja Bai and Schnabel, {Julia Anne} and Reza Razavi and King, {Andrew Peter}",

year = "2019",

doi = "10.1016/j.jcmg.2019.05.030",

language = "English",

volume = "13",

pages = "684--695",

journal = "JACC Cardiovascular Imaging",

issn = "1936-878X",

number = "3",

}

TY - JOUR

T1 - Fully automated, quality-controlled cardiac analysis from CMR

T2 - validation and large-scale application to characterize cardiac function

AU - Ruijsink, Jacobus Bernardus

AU - Puyol Anton, Esther

AU - Oksuz, Ilkay

AU - Sinclair, Matthew

AU - Bai, Wenja

AU - Schnabel, Julia Anne

AU - Razavi, Reza

AU - King, Andrew Peter

PY - 2019

Y1 - 2019

N2 - Objectives:Develop a fully automated framework for cardiac function analysis from cardiacmagnetic resonance (CMR), including comprehensive quality control (QC) algorithmsto detect erroneous output.Background:Analysis of cine CMR imaging using deep learning (DL) algorithms could automateventricular function assessment. However, variable image quality, variability inphenotypes of disease and unavoidable weaknesses in training of DL algorithmscurrently prevent their use in clinical practice.Methods:The framework consists of a pre-analysis DL image QC, followed by a DL algorithmfor biventricular segmentation in long- and short-axis, myocardial feature-tracking(FT) and a post-analysis QC to detect erroneous results. We validated the frameworkin healthy subjects and cardiac patients by comparison against manual analysis(n=100) and evaluation of the QC steps’ ability to detect erroneous results (n=700).Next, we utilized our method to obtain reference values for cardiac function metricsfrom the UK Biobank.Results:Automated analysis correlated highly with manual analysis for left and rightventricular volumes (all r>0.95), strain (circumferential: r=0.89, longitudinal: r>0.89)and filling and ejection rates (all r≥0.93). There was no significant bias for cardiacvolumes and filling and ejection rates, except for RVESV (bias +1.80 mL, p=.01).The bias for FT-strain was <1.3%. The sensitivity of detection of erroneous outputwas 95% for volume-derived parameters and 93% for FT strain. Finally, referencevalues were automatically derived from 2,029 CMR exams in healthy subjects.Conclusions:We demonstrate a DL-based framework for automated, quality-controlledcharacterization of cardiac function from cine CMR, without the need for directclinician oversight.

AB - Objectives:Develop a fully automated framework for cardiac function analysis from cardiacmagnetic resonance (CMR), including comprehensive quality control (QC) algorithmsto detect erroneous output.Background:Analysis of cine CMR imaging using deep learning (DL) algorithms could automateventricular function assessment. However, variable image quality, variability inphenotypes of disease and unavoidable weaknesses in training of DL algorithmscurrently prevent their use in clinical practice.Methods:The framework consists of a pre-analysis DL image QC, followed by a DL algorithmfor biventricular segmentation in long- and short-axis, myocardial feature-tracking(FT) and a post-analysis QC to detect erroneous results. We validated the frameworkin healthy subjects and cardiac patients by comparison against manual analysis(n=100) and evaluation of the QC steps’ ability to detect erroneous results (n=700).Next, we utilized our method to obtain reference values for cardiac function metricsfrom the UK Biobank.Results:Automated analysis correlated highly with manual analysis for left and rightventricular volumes (all r>0.95), strain (circumferential: r=0.89, longitudinal: r>0.89)and filling and ejection rates (all r≥0.93). There was no significant bias for cardiacvolumes and filling and ejection rates, except for RVESV (bias +1.80 mL, p=.01).The bias for FT-strain was <1.3%. The sensitivity of detection of erroneous outputwas 95% for volume-derived parameters and 93% for FT strain. Finally, referencevalues were automatically derived from 2,029 CMR exams in healthy subjects.Conclusions:We demonstrate a DL-based framework for automated, quality-controlledcharacterization of cardiac function from cine CMR, without the need for directclinician oversight.

U2 - 10.1016/j.jcmg.2019.05.030

DO - 10.1016/j.jcmg.2019.05.030

M3 - Article

SN - 1936-878X

VL - 13

SP - 684

EP - 695

JO - JACC Cardiovascular Imaging

JF - JACC Cardiovascular Imaging

IS - 3

ER -

Fully automated, quality-controlled cardiac analysis from CMR: validation and large-scale application to characterize cardiac function

Abstract

UN SDGs

Access to Document

Embargoed Document

Fingerprint

Cite this