Methods of artificial enlargement of the training set for statistical shape models

Juha Koikkalainen, T. Tölli, K. Lauerma, Kari Antila, Elina M. Mattila, M. Lilja, Lötjönen Jyrki

Research output: Contribution to journalArticleScientificpeer-review

35 Citations (Scopus)

Abstract

Due to the small size of training sets, statistical shape models often over-constrain the deformation in medical image segmentation. Hence, artificial enlargement of the training set has been proposed as a solution for the problem to increase the flexibility of the models. In this paper, different methods were evaluated to artificially enlarge a training set. Furthermore, the objectives were to study the effects of the size of the training set, to estimate the optimal number of deformation modes, to study the effects of different error sources, and to compare different deformation methods. The study was performed for a cardiac shape model consisting of ventricles, atria, and epicardium, and built from magnetic resonance (MR) volume images of 25 subjects. Both shape modeling and image segmentation accuracies were studied. The objectives were reached by utilizing different training sets and datasets, and two deformation methods. The evaluation proved that artificial enlargement of the training set improves both the modeling and segmentation accuracy. All but one enlargement techniques gave statistically significantly (p < 0.05) better segmentation results than the standard method without enlargement. The two best enlargement techniques were the nonrigid movement technique and the technique that combines principal component analysis (PCA) and finite element model (FEM). The optimal number of deformation modes was found to be near 100 modes in our application. The active shape model segmentation gave better segmentation accuracy than the one based on the simulated annealing optimization of the model weights.
Original languageEnglish
Pages (from-to)1643-1654
Number of pages12
JournalIEEE Transactions on Medical Imaging
Volume27
Issue number11
DOIs
Publication statusPublished - 2008
MoE publication typeA1 Journal article-refereed

Fingerprint

Statistical Models
Image segmentation
Pericardium
Principal Component Analysis
Research Design
Magnetic Resonance Spectroscopy
Magnetic resonance
Simulated annealing
Weights and Measures
Principal component analysis

Keywords

  • active shape model
  • cardiac magnetic resonance imaging
  • MRI
  • point distribution model
  • statistical shape model
  • training set

Cite this

Koikkalainen, Juha ; Tölli, T. ; Lauerma, K. ; Antila, Kari ; Mattila, Elina M. ; Lilja, M. ; Jyrki, Lötjönen. / Methods of artificial enlargement of the training set for statistical shape models. In: IEEE Transactions on Medical Imaging. 2008 ; Vol. 27, No. 11. pp. 1643-1654.
@article{89760e0867f043178a9109af25aa0166,
title = "Methods of artificial enlargement of the training set for statistical shape models",
abstract = "Due to the small size of training sets, statistical shape models often over-constrain the deformation in medical image segmentation. Hence, artificial enlargement of the training set has been proposed as a solution for the problem to increase the flexibility of the models. In this paper, different methods were evaluated to artificially enlarge a training set. Furthermore, the objectives were to study the effects of the size of the training set, to estimate the optimal number of deformation modes, to study the effects of different error sources, and to compare different deformation methods. The study was performed for a cardiac shape model consisting of ventricles, atria, and epicardium, and built from magnetic resonance (MR) volume images of 25 subjects. Both shape modeling and image segmentation accuracies were studied. The objectives were reached by utilizing different training sets and datasets, and two deformation methods. The evaluation proved that artificial enlargement of the training set improves both the modeling and segmentation accuracy. All but one enlargement techniques gave statistically significantly (p < 0.05) better segmentation results than the standard method without enlargement. The two best enlargement techniques were the nonrigid movement technique and the technique that combines principal component analysis (PCA) and finite element model (FEM). The optimal number of deformation modes was found to be near 100 modes in our application. The active shape model segmentation gave better segmentation accuracy than the one based on the simulated annealing optimization of the model weights.",
keywords = "active shape model, cardiac magnetic resonance imaging, MRI, point distribution model, statistical shape model, training set",
author = "Juha Koikkalainen and T. T{\"o}lli and K. Lauerma and Kari Antila and Mattila, {Elina M.} and M. Lilja and L{\"o}tj{\"o}nen Jyrki",
note = "Project code: 22601",
year = "2008",
doi = "10.1109/TMI.2008.929106",
language = "English",
volume = "27",
pages = "1643--1654",
journal = "IEEE Transactions on Medical Imaging",
issn = "0278-0062",
publisher = "Institute of Electrical and Electronic Engineers IEEE",
number = "11",

}

Methods of artificial enlargement of the training set for statistical shape models. / Koikkalainen, Juha; Tölli, T.; Lauerma, K.; Antila, Kari; Mattila, Elina M.; Lilja, M.; Jyrki, Lötjönen.

In: IEEE Transactions on Medical Imaging, Vol. 27, No. 11, 2008, p. 1643-1654.

Research output: Contribution to journalArticleScientificpeer-review

TY - JOUR

T1 - Methods of artificial enlargement of the training set for statistical shape models

AU - Koikkalainen, Juha

AU - Tölli, T.

AU - Lauerma, K.

AU - Antila, Kari

AU - Mattila, Elina M.

AU - Lilja, M.

AU - Jyrki, Lötjönen

N1 - Project code: 22601

PY - 2008

Y1 - 2008

N2 - Due to the small size of training sets, statistical shape models often over-constrain the deformation in medical image segmentation. Hence, artificial enlargement of the training set has been proposed as a solution for the problem to increase the flexibility of the models. In this paper, different methods were evaluated to artificially enlarge a training set. Furthermore, the objectives were to study the effects of the size of the training set, to estimate the optimal number of deformation modes, to study the effects of different error sources, and to compare different deformation methods. The study was performed for a cardiac shape model consisting of ventricles, atria, and epicardium, and built from magnetic resonance (MR) volume images of 25 subjects. Both shape modeling and image segmentation accuracies were studied. The objectives were reached by utilizing different training sets and datasets, and two deformation methods. The evaluation proved that artificial enlargement of the training set improves both the modeling and segmentation accuracy. All but one enlargement techniques gave statistically significantly (p < 0.05) better segmentation results than the standard method without enlargement. The two best enlargement techniques were the nonrigid movement technique and the technique that combines principal component analysis (PCA) and finite element model (FEM). The optimal number of deformation modes was found to be near 100 modes in our application. The active shape model segmentation gave better segmentation accuracy than the one based on the simulated annealing optimization of the model weights.

AB - Due to the small size of training sets, statistical shape models often over-constrain the deformation in medical image segmentation. Hence, artificial enlargement of the training set has been proposed as a solution for the problem to increase the flexibility of the models. In this paper, different methods were evaluated to artificially enlarge a training set. Furthermore, the objectives were to study the effects of the size of the training set, to estimate the optimal number of deformation modes, to study the effects of different error sources, and to compare different deformation methods. The study was performed for a cardiac shape model consisting of ventricles, atria, and epicardium, and built from magnetic resonance (MR) volume images of 25 subjects. Both shape modeling and image segmentation accuracies were studied. The objectives were reached by utilizing different training sets and datasets, and two deformation methods. The evaluation proved that artificial enlargement of the training set improves both the modeling and segmentation accuracy. All but one enlargement techniques gave statistically significantly (p < 0.05) better segmentation results than the standard method without enlargement. The two best enlargement techniques were the nonrigid movement technique and the technique that combines principal component analysis (PCA) and finite element model (FEM). The optimal number of deformation modes was found to be near 100 modes in our application. The active shape model segmentation gave better segmentation accuracy than the one based on the simulated annealing optimization of the model weights.

KW - active shape model

KW - cardiac magnetic resonance imaging

KW - MRI

KW - point distribution model

KW - statistical shape model

KW - training set

U2 - 10.1109/TMI.2008.929106

DO - 10.1109/TMI.2008.929106

M3 - Article

VL - 27

SP - 1643

EP - 1654

JO - IEEE Transactions on Medical Imaging

JF - IEEE Transactions on Medical Imaging

SN - 0278-0062

IS - 11

ER -