Annotated normal CT data of the abdomen for deep learning

Challenges and strategies for implementation

S. Park, Linda Chi Hang Chu, E. K. Fishman, A. L. Yuille, Bert Vogelstein, Kenneth W Kinzler, K. M. Horton, R. H. Hruban, Eva S. Zinreich, D. Fadaei Fouladi, S. Shayesteh, J. Graves, Satomi Kawamoto

Research output: Contribution to journalArticle

Abstract

Purpose: The purpose of this study was to report procedures developed to annotate abdominal computed tomography (CT) images from subjects without pancreatic disease that will be used as the input for deep convolutional neural networks (DNN) for development of deep learning algorithms for automatic recognition of a normal pancreas. Materials and methods: Dual-phase contrast-enhanced volumetric CT acquired from 2005 to 2009 from potential kidney donors were retrospectively assessed. Four trained human annotators manually and sequentially annotated 22 structures in each datasets, then expert radiologists confirmed the annotation. For efficient annotation and data management, a commercial software package that supports three-dimensional segmentation was used. Results: A total of 1150 dual-phase CT datasets from 575 subjects were annotated. There were 229 men and 346 women (mean age: 45 ± 12 years; range: 18–79 years). The mean intra-observer intra-subject dual-phase CT volume difference of all annotated structures was 4.27 mL (7.65%). The deep network prediction for multi-organ segmentation showed high fidelity with 89.4% and 1.29 mm in terms of mean Dice similarity coefficients and mean surface distances, respectively. Conclusions: A reliable data collection/annotation process for abdominal structures was developed. This process can be used to generate large datasets appropriate for deep learning.

Original languageEnglish (US)
JournalDiagnostic and Interventional Imaging
DOIs
StatePublished - Jan 1 2019

Fingerprint

Abdomen
Cone-Beam Computed Tomography
Tomography
Learning
Pancreatic Diseases
Pancreas
Software
Tissue Donors
Kidney
Datasets
Data Curation
Recognition (Psychology)
Radiologists

Keywords

  • Abdominal computed tomography (CT)
  • Artificial intelligence (AI)
  • Image segmentation
  • Machine learning
  • Normal structures

ASJC Scopus subject areas

  • Radiological and Ultrasound Technology
  • Radiology Nuclear Medicine and imaging

Cite this

Annotated normal CT data of the abdomen for deep learning : Challenges and strategies for implementation. / Park, S.; Chu, Linda Chi Hang; Fishman, E. K.; Yuille, A. L.; Vogelstein, Bert; Kinzler, Kenneth W; Horton, K. M.; Hruban, R. H.; Zinreich, Eva S.; Fadaei Fouladi, D.; Shayesteh, S.; Graves, J.; Kawamoto, Satomi.

In: Diagnostic and Interventional Imaging, 01.01.2019.

Research output: Contribution to journalArticle

@article{99546ad3236c465284daf12f478293c8,
title = "Annotated normal CT data of the abdomen for deep learning: Challenges and strategies for implementation",
abstract = "Purpose: The purpose of this study was to report procedures developed to annotate abdominal computed tomography (CT) images from subjects without pancreatic disease that will be used as the input for deep convolutional neural networks (DNN) for development of deep learning algorithms for automatic recognition of a normal pancreas. Materials and methods: Dual-phase contrast-enhanced volumetric CT acquired from 2005 to 2009 from potential kidney donors were retrospectively assessed. Four trained human annotators manually and sequentially annotated 22 structures in each datasets, then expert radiologists confirmed the annotation. For efficient annotation and data management, a commercial software package that supports three-dimensional segmentation was used. Results: A total of 1150 dual-phase CT datasets from 575 subjects were annotated. There were 229 men and 346 women (mean age: 45 ± 12 years; range: 18–79 years). The mean intra-observer intra-subject dual-phase CT volume difference of all annotated structures was 4.27 mL (7.65{\%}). The deep network prediction for multi-organ segmentation showed high fidelity with 89.4{\%} and 1.29 mm in terms of mean Dice similarity coefficients and mean surface distances, respectively. Conclusions: A reliable data collection/annotation process for abdominal structures was developed. This process can be used to generate large datasets appropriate for deep learning.",
keywords = "Abdominal computed tomography (CT), Artificial intelligence (AI), Image segmentation, Machine learning, Normal structures",
author = "S. Park and Chu, {Linda Chi Hang} and Fishman, {E. K.} and Yuille, {A. L.} and Bert Vogelstein and Kinzler, {Kenneth W} and Horton, {K. M.} and Hruban, {R. H.} and Zinreich, {Eva S.} and {Fadaei Fouladi}, D. and S. Shayesteh and J. Graves and Satomi Kawamoto",
year = "2019",
month = "1",
day = "1",
doi = "10.1016/j.diii.2019.05.008",
language = "English (US)",
journal = "Diagnostic and interventional imaging",
issn = "2211-5684",
publisher = "Elsevier Masson",

}

TY - JOUR

T1 - Annotated normal CT data of the abdomen for deep learning

T2 - Challenges and strategies for implementation

AU - Park, S.

AU - Chu, Linda Chi Hang

AU - Fishman, E. K.

AU - Yuille, A. L.

AU - Vogelstein, Bert

AU - Kinzler, Kenneth W

AU - Horton, K. M.

AU - Hruban, R. H.

AU - Zinreich, Eva S.

AU - Fadaei Fouladi, D.

AU - Shayesteh, S.

AU - Graves, J.

AU - Kawamoto, Satomi

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Purpose: The purpose of this study was to report procedures developed to annotate abdominal computed tomography (CT) images from subjects without pancreatic disease that will be used as the input for deep convolutional neural networks (DNN) for development of deep learning algorithms for automatic recognition of a normal pancreas. Materials and methods: Dual-phase contrast-enhanced volumetric CT acquired from 2005 to 2009 from potential kidney donors were retrospectively assessed. Four trained human annotators manually and sequentially annotated 22 structures in each datasets, then expert radiologists confirmed the annotation. For efficient annotation and data management, a commercial software package that supports three-dimensional segmentation was used. Results: A total of 1150 dual-phase CT datasets from 575 subjects were annotated. There were 229 men and 346 women (mean age: 45 ± 12 years; range: 18–79 years). The mean intra-observer intra-subject dual-phase CT volume difference of all annotated structures was 4.27 mL (7.65%). The deep network prediction for multi-organ segmentation showed high fidelity with 89.4% and 1.29 mm in terms of mean Dice similarity coefficients and mean surface distances, respectively. Conclusions: A reliable data collection/annotation process for abdominal structures was developed. This process can be used to generate large datasets appropriate for deep learning.

AB - Purpose: The purpose of this study was to report procedures developed to annotate abdominal computed tomography (CT) images from subjects without pancreatic disease that will be used as the input for deep convolutional neural networks (DNN) for development of deep learning algorithms for automatic recognition of a normal pancreas. Materials and methods: Dual-phase contrast-enhanced volumetric CT acquired from 2005 to 2009 from potential kidney donors were retrospectively assessed. Four trained human annotators manually and sequentially annotated 22 structures in each datasets, then expert radiologists confirmed the annotation. For efficient annotation and data management, a commercial software package that supports three-dimensional segmentation was used. Results: A total of 1150 dual-phase CT datasets from 575 subjects were annotated. There were 229 men and 346 women (mean age: 45 ± 12 years; range: 18–79 years). The mean intra-observer intra-subject dual-phase CT volume difference of all annotated structures was 4.27 mL (7.65%). The deep network prediction for multi-organ segmentation showed high fidelity with 89.4% and 1.29 mm in terms of mean Dice similarity coefficients and mean surface distances, respectively. Conclusions: A reliable data collection/annotation process for abdominal structures was developed. This process can be used to generate large datasets appropriate for deep learning.

KW - Abdominal computed tomography (CT)

KW - Artificial intelligence (AI)

KW - Image segmentation

KW - Machine learning

KW - Normal structures

UR - http://www.scopus.com/inward/record.url?scp=85069671560&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85069671560&partnerID=8YFLogxK

U2 - 10.1016/j.diii.2019.05.008

DO - 10.1016/j.diii.2019.05.008

M3 - Article

JO - Diagnostic and interventional imaging

JF - Diagnostic and interventional imaging

SN - 2211-5684

ER -