Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound

F. van den Noort (Corresponding Author), A.T.M. Grob, C.H. Slump, C. H. van der Vaart , M. van Stralen

Research output: Contribution to journalArticleAcademicpeer-review

1 Citation (Scopus)
22 Downloads (Pure)

Abstract

Objectives: The introduction of 3D analysis of the puborectalis muscle, for diagnostic purposes, into daily practice is hindered by the need for appropriate training of the observers. Automatic 3D segmentation of the puborectalis muscle in 3D transperineal ultrasound may aid to its adaption in clinical practice.

Method: A manual 3D segmentation protocol was developed to segment the puborectalis muscle. The data of 20 women, in their first trimester of pregnancy, was used to validate the reproducibility of this protocol. For automatic segmentation, active appearance models of the puborectalis muscle were developed. Those models were trained using manual segmentation data of 50 women. The performance of both manual and automatic segmentation was analyzed by measuring the overlap and distance between the segmentations. Also, the interclass correlation coefficients and their 95% confidence intervals were determined for mean echogenicity and volume of the puborectalis muscle.

Results: The ICC values of mean echogenicity (0.968-0.991) and volume (0.626-0.910) are good to very good for both automatic and manual segmentation. The results of overlap and distance for manual segmentation are as expected, showing only few pixels (2–3) mismatch on average and a reasonable overlap. Based on overlap and distance 5 mismatches in automatic segmentation were detected, resulting in an automatic segmentation a success rate of 90%.

Conclusions: In conclusion, this study presents a reliable manual and automatic 3D segmentation of the puborectalis muscle. This will facilitate future investigation of the puborectalis muscle. It also allows for reliable measurements of clinically potentially valuable parameters like mean echogenicity.
Original languageEnglish
Pages (from-to)97-102
Number of pages6
JournalUltrasound in obstetrics & gynecology
Volume52
Issue number1
DOIs
Publication statusPublished - 1 Jul 2018

Fingerprint

Muscles
First Pregnancy Trimester
Confidence Intervals

Keywords

  • UT-Hybrid-D
  • Active appearance model
  • Puborectalis muscle
  • Ultrasound
  • 3D segmentation

Cite this

@article{38c13b3a8f45424aad71c9f445a5b012,
title = "Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound",
abstract = "Objectives: The introduction of 3D analysis of the puborectalis muscle, for diagnostic purposes, into daily practice is hindered by the need for appropriate training of the observers. Automatic 3D segmentation of the puborectalis muscle in 3D transperineal ultrasound may aid to its adaption in clinical practice.Method: A manual 3D segmentation protocol was developed to segment the puborectalis muscle. The data of 20 women, in their first trimester of pregnancy, was used to validate the reproducibility of this protocol. For automatic segmentation, active appearance models of the puborectalis muscle were developed. Those models were trained using manual segmentation data of 50 women. The performance of both manual and automatic segmentation was analyzed by measuring the overlap and distance between the segmentations. Also, the interclass correlation coefficients and their 95{\%} confidence intervals were determined for mean echogenicity and volume of the puborectalis muscle.Results: The ICC values of mean echogenicity (0.968-0.991) and volume (0.626-0.910) are good to very good for both automatic and manual segmentation. The results of overlap and distance for manual segmentation are as expected, showing only few pixels (2–3) mismatch on average and a reasonable overlap. Based on overlap and distance 5 mismatches in automatic segmentation were detected, resulting in an automatic segmentation a success rate of 90{\%}.Conclusions: In conclusion, this study presents a reliable manual and automatic 3D segmentation of the puborectalis muscle. This will facilitate future investigation of the puborectalis muscle. It also allows for reliable measurements of clinically potentially valuable parameters like mean echogenicity.",
keywords = "UT-Hybrid-D, Active appearance model, Puborectalis muscle, Ultrasound, 3D segmentation",
author = "{van den Noort}, F. and A.T.M. Grob and C.H. Slump and {van der Vaart}, {C. H.} and {van Stralen}, M.",
note = "Wiley deal",
year = "2018",
month = "7",
day = "1",
doi = "10.1002/uog.18927",
language = "English",
volume = "52",
pages = "97--102",
journal = "Ultrasound in obstetrics & gynecology",
issn = "0960-7692",
publisher = "Wiley",
number = "1",

}

Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound. / van den Noort, F. (Corresponding Author); Grob, A.T.M.; Slump, C.H.; van der Vaart , C. H. ; van Stralen , M. .

In: Ultrasound in obstetrics & gynecology, Vol. 52, No. 1, 01.07.2018, p. 97-102.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound

AU - van den Noort, F.

AU - Grob, A.T.M.

AU - Slump, C.H.

AU - van der Vaart , C. H.

AU - van Stralen , M.

N1 - Wiley deal

PY - 2018/7/1

Y1 - 2018/7/1

N2 - Objectives: The introduction of 3D analysis of the puborectalis muscle, for diagnostic purposes, into daily practice is hindered by the need for appropriate training of the observers. Automatic 3D segmentation of the puborectalis muscle in 3D transperineal ultrasound may aid to its adaption in clinical practice.Method: A manual 3D segmentation protocol was developed to segment the puborectalis muscle. The data of 20 women, in their first trimester of pregnancy, was used to validate the reproducibility of this protocol. For automatic segmentation, active appearance models of the puborectalis muscle were developed. Those models were trained using manual segmentation data of 50 women. The performance of both manual and automatic segmentation was analyzed by measuring the overlap and distance between the segmentations. Also, the interclass correlation coefficients and their 95% confidence intervals were determined for mean echogenicity and volume of the puborectalis muscle.Results: The ICC values of mean echogenicity (0.968-0.991) and volume (0.626-0.910) are good to very good for both automatic and manual segmentation. The results of overlap and distance for manual segmentation are as expected, showing only few pixels (2–3) mismatch on average and a reasonable overlap. Based on overlap and distance 5 mismatches in automatic segmentation were detected, resulting in an automatic segmentation a success rate of 90%.Conclusions: In conclusion, this study presents a reliable manual and automatic 3D segmentation of the puborectalis muscle. This will facilitate future investigation of the puborectalis muscle. It also allows for reliable measurements of clinically potentially valuable parameters like mean echogenicity.

AB - Objectives: The introduction of 3D analysis of the puborectalis muscle, for diagnostic purposes, into daily practice is hindered by the need for appropriate training of the observers. Automatic 3D segmentation of the puborectalis muscle in 3D transperineal ultrasound may aid to its adaption in clinical practice.Method: A manual 3D segmentation protocol was developed to segment the puborectalis muscle. The data of 20 women, in their first trimester of pregnancy, was used to validate the reproducibility of this protocol. For automatic segmentation, active appearance models of the puborectalis muscle were developed. Those models were trained using manual segmentation data of 50 women. The performance of both manual and automatic segmentation was analyzed by measuring the overlap and distance between the segmentations. Also, the interclass correlation coefficients and their 95% confidence intervals were determined for mean echogenicity and volume of the puborectalis muscle.Results: The ICC values of mean echogenicity (0.968-0.991) and volume (0.626-0.910) are good to very good for both automatic and manual segmentation. The results of overlap and distance for manual segmentation are as expected, showing only few pixels (2–3) mismatch on average and a reasonable overlap. Based on overlap and distance 5 mismatches in automatic segmentation were detected, resulting in an automatic segmentation a success rate of 90%.Conclusions: In conclusion, this study presents a reliable manual and automatic 3D segmentation of the puborectalis muscle. This will facilitate future investigation of the puborectalis muscle. It also allows for reliable measurements of clinically potentially valuable parameters like mean echogenicity.

KW - UT-Hybrid-D

KW - Active appearance model

KW - Puborectalis muscle

KW - Ultrasound

KW - 3D segmentation

U2 - 10.1002/uog.18927

DO - 10.1002/uog.18927

M3 - Article

VL - 52

SP - 97

EP - 102

JO - Ultrasound in obstetrics & gynecology

JF - Ultrasound in obstetrics & gynecology

SN - 0960-7692

IS - 1

ER -