Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR

Jorge Mariscal-Harana; Naomi Kifle; Reza Razavi; Andrew P. King; Bram Ruijsink; Esther Puyol-Antón

doi:10.1007/978-3-030-93722-5_10

Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR

Jorge Mariscal-Harana^*, Naomi Kifle, Reza Razavi, Andrew P. King, Bram Ruijsink, Esther Puyol-Antón

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

Abstract

Current artificial intelligence (AI) algorithms for short-axis cardiac magnetic resonance (CMR) segmentation achieve human performance for slices situated in the middle of the heart. However, an often-overlooked fact is that segmentation of the basal and apical slices is more difficult. During manual analysis, differences in the basal segmentations have been reported as one of the major sources of disagreement in human interobserver variability. In this work, we aim to investigate the performance of AI algorithms in segmenting basal and apical slices and design strategies to improve their segmentation. We trained all our models on a large dataset of clinical CMR studies obtained from two NHS hospitals (n = 4,228) and evaluated them against two external datasets: ACDC (n = 100) and M&Ms (n = 321). Using manual segmentations as a reference, CMR slices were assigned to one of four regions: non-cardiac, base, middle, and apex. Using the ‘nnU-Net’ framework as a baseline, we investigated two different approaches to reduce the segmentation performance gap between cardiac regions: (1) non-uniform batch sampling, which allows us to choose how often images from different regions are seen during training; and (2) a cardiac-region classification model followed by three (i.e. base, middle, and apex) region-specific segmentation models. We show that the classification and segmentation approach was best at reducing the performance gap across all datasets. We also show that improvements in the classification performance can subsequently lead to a significantly better performance in the segmentation task.

Original language	English
Title of host publication	Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers
Editors	Esther Puyol Antón, Alistair Young, Avan Suinesiaputra, Mihaela Pop, Carlos Martín-Isla, Maxime Sermesant, Oscar Camara, Karim Lekadir
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	84-92
Number of pages	9
ISBN (Print)	9783030937218
DOIs	https://doi.org/10.1007/978-3-030-93722-5_10
Publication status	Published - 2022
Event	12th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2021 held in conjunction with MICCAI 2021 - Strasbourg, France Duration: 27 Sept 2021 → 27 Sept 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13131 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	12th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2021 held in conjunction with MICCAI 2021
Country/Territory	France
City	Strasbourg
Period	27/09/2021 → 27/09/2021

Keywords

Cardiac magnetic resonance
Class imbalance
Deep learning
Segmentation

Access to Document

10.1007/978-3-030-93722-5_10

Cite this

Mariscal-Harana, J., Kifle, N., Razavi, R., King, A. P., Ruijsink, B., & Puyol-Antón, E. (2022). Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR. In E. Puyol Antón, A. Young, A. Suinesiaputra, M. Pop, C. Martín-Isla, M. Sermesant, O. Camara, & K. Lekadir (Eds.), Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers (pp. 84-92). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13131 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-93722-5_10

Mariscal-Harana, Jorge ; Kifle, Naomi ; Razavi, Reza et al. / Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR. Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers. editor / Esther Puyol Antón ; Alistair Young ; Avan Suinesiaputra ; Mihaela Pop ; Carlos Martín-Isla ; Maxime Sermesant ; Oscar Camara ; Karim Lekadir. Springer Science and Business Media Deutschland GmbH, 2022. pp. 84-92 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{729b86e7a93a43fbb7b1ee663a1300ed,

title = "Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR",

abstract = "Current artificial intelligence (AI) algorithms for short-axis cardiac magnetic resonance (CMR) segmentation achieve human performance for slices situated in the middle of the heart. However, an often-overlooked fact is that segmentation of the basal and apical slices is more difficult. During manual analysis, differences in the basal segmentations have been reported as one of the major sources of disagreement in human interobserver variability. In this work, we aim to investigate the performance of AI algorithms in segmenting basal and apical slices and design strategies to improve their segmentation. We trained all our models on a large dataset of clinical CMR studies obtained from two NHS hospitals (n = 4,228) and evaluated them against two external datasets: ACDC (n = 100) and M&Ms (n = 321). Using manual segmentations as a reference, CMR slices were assigned to one of four regions: non-cardiac, base, middle, and apex. Using the {\textquoteleft}nnU-Net{\textquoteright} framework as a baseline, we investigated two different approaches to reduce the segmentation performance gap between cardiac regions: (1) non-uniform batch sampling, which allows us to choose how often images from different regions are seen during training; and (2) a cardiac-region classification model followed by three (i.e. base, middle, and apex) region-specific segmentation models. We show that the classification and segmentation approach was best at reducing the performance gap across all datasets. We also show that improvements in the classification performance can subsequently lead to a significantly better performance in the segmentation task.",

keywords = "Cardiac magnetic resonance, Class imbalance, Deep learning, Segmentation",

author = "Jorge Mariscal-Harana and Naomi Kifle and Reza Razavi and King, {Andrew P.} and Bram Ruijsink and Esther Puyol-Ant{\'o}n",

note = "Funding Information: Acknowledgements. This work was supported by the EPSRC (EP/P001009/1 and the Advancing Impact Award scheme of the Impact Acceleration Account at King{\textquoteright}s College London) and the Wellcome EPSRC Centre for Medical Engineering at the School of Biomedical Engineering and Imaging Sciences, King{\textquoteright}s College London (WT 203148/Z/16/Z). Publisher Copyright: {\textcopyright} 2022, Springer Nature Switzerland AG.; 12th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2021 held in conjunction with MICCAI 2021 ; Conference date: 27-09-2021 Through 27-09-2021",

year = "2022",

doi = "10.1007/978-3-030-93722-5_10",

language = "English",

isbn = "9783030937218",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "84--92",

editor = "{Puyol Ant{\'o}n}, Esther and Alistair Young and Avan Suinesiaputra and Mihaela Pop and Carlos Mart{\'i}n-Isla and Maxime Sermesant and Oscar Camara and Karim Lekadir",

booktitle = "Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers",

address = "Germany",

}

Mariscal-Harana, J, Kifle, N, Razavi, R, King, AP, Ruijsink, B & Puyol-Antón, E 2022, Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR. in E Puyol Antón, A Young, A Suinesiaputra, M Pop, C Martín-Isla, M Sermesant, O Camara & K Lekadir (eds), Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13131 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 84-92, 12th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2021 held in conjunction with MICCAI 2021, Strasbourg, France, 27/09/2021. https://doi.org/10.1007/978-3-030-93722-5_10

Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR. / Mariscal-Harana, Jorge; Kifle, Naomi; Razavi, Reza et al.
Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers. ed. / Esther Puyol Antón; Alistair Young; Avan Suinesiaputra; Mihaela Pop; Carlos Martín-Isla; Maxime Sermesant; Oscar Camara; Karim Lekadir. Springer Science and Business Media Deutschland GmbH, 2022. p. 84-92 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13131 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

TY - CHAP

T1 - Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR

AU - Mariscal-Harana, Jorge

AU - Kifle, Naomi

AU - Razavi, Reza

AU - King, Andrew P.

AU - Ruijsink, Bram

AU - Puyol-Antón, Esther

N1 - Funding Information: Acknowledgements. This work was supported by the EPSRC (EP/P001009/1 and the Advancing Impact Award scheme of the Impact Acceleration Account at King’s College London) and the Wellcome EPSRC Centre for Medical Engineering at the School of Biomedical Engineering and Imaging Sciences, King’s College London (WT 203148/Z/16/Z). Publisher Copyright: © 2022, Springer Nature Switzerland AG.

PY - 2022

Y1 - 2022

N2 - Current artificial intelligence (AI) algorithms for short-axis cardiac magnetic resonance (CMR) segmentation achieve human performance for slices situated in the middle of the heart. However, an often-overlooked fact is that segmentation of the basal and apical slices is more difficult. During manual analysis, differences in the basal segmentations have been reported as one of the major sources of disagreement in human interobserver variability. In this work, we aim to investigate the performance of AI algorithms in segmenting basal and apical slices and design strategies to improve their segmentation. We trained all our models on a large dataset of clinical CMR studies obtained from two NHS hospitals (n = 4,228) and evaluated them against two external datasets: ACDC (n = 100) and M&Ms (n = 321). Using manual segmentations as a reference, CMR slices were assigned to one of four regions: non-cardiac, base, middle, and apex. Using the ‘nnU-Net’ framework as a baseline, we investigated two different approaches to reduce the segmentation performance gap between cardiac regions: (1) non-uniform batch sampling, which allows us to choose how often images from different regions are seen during training; and (2) a cardiac-region classification model followed by three (i.e. base, middle, and apex) region-specific segmentation models. We show that the classification and segmentation approach was best at reducing the performance gap across all datasets. We also show that improvements in the classification performance can subsequently lead to a significantly better performance in the segmentation task.

AB - Current artificial intelligence (AI) algorithms for short-axis cardiac magnetic resonance (CMR) segmentation achieve human performance for slices situated in the middle of the heart. However, an often-overlooked fact is that segmentation of the basal and apical slices is more difficult. During manual analysis, differences in the basal segmentations have been reported as one of the major sources of disagreement in human interobserver variability. In this work, we aim to investigate the performance of AI algorithms in segmenting basal and apical slices and design strategies to improve their segmentation. We trained all our models on a large dataset of clinical CMR studies obtained from two NHS hospitals (n = 4,228) and evaluated them against two external datasets: ACDC (n = 100) and M&Ms (n = 321). Using manual segmentations as a reference, CMR slices were assigned to one of four regions: non-cardiac, base, middle, and apex. Using the ‘nnU-Net’ framework as a baseline, we investigated two different approaches to reduce the segmentation performance gap between cardiac regions: (1) non-uniform batch sampling, which allows us to choose how often images from different regions are seen during training; and (2) a cardiac-region classification model followed by three (i.e. base, middle, and apex) region-specific segmentation models. We show that the classification and segmentation approach was best at reducing the performance gap across all datasets. We also show that improvements in the classification performance can subsequently lead to a significantly better performance in the segmentation task.

KW - Cardiac magnetic resonance

KW - Class imbalance

KW - Deep learning

KW - Segmentation

UR - http://www.scopus.com/inward/record.url?scp=85124030996&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-93722-5_10

DO - 10.1007/978-3-030-93722-5_10

M3 - Conference paper

AN - SCOPUS:85124030996

SN - 9783030937218

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 84

EP - 92

BT - Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers

A2 - Puyol Antón, Esther

A2 - Young, Alistair

A2 - Suinesiaputra, Avan

A2 - Pop, Mihaela

A2 - Martín-Isla, Carlos

A2 - Sermesant, Maxime

A2 - Camara, Oscar

A2 - Lekadir, Karim

PB - Springer Science and Business Media Deutschland GmbH

T2 - 12th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2021 held in conjunction with MICCAI 2021

Y2 - 27 September 2021 through 27 September 2021

ER -

Mariscal-Harana J, Kifle N, Razavi R, King AP, Ruijsink B, Puyol-Antón E. Improved AI-Based Segmentation of Apical and Basal Slices from Clinical Cine CMR. In Puyol Antón E, Young A, Suinesiaputra A, Pop M, Martín-Isla C, Sermesant M, Camara O, Lekadir K, editors, Statistical Atlases and Computational Models of the Heart. Multi-Disease, Multi-View, and Multi-Center Right Ventricular Segmentation in Cardiac MRI Challenge - 12th International Workshop, STACOM 2021, Held in Conjunction with MICCAI 2021, Revised Selected Papers. Springer Science and Business Media Deutschland GmbH. 2022. p. 84-92. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-93722-5_10