Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2015 Oct;2(4):044006.
doi: 10.1117/1.JMI.2.4.044006. Epub 2015 Dec 30.

Automated segmentation of the thyroid gland on thoracic CT scans by multiatlas label fusion and random forest classification

Divya Narayanan  1 , Jiamin Liu  1 , Lauren Kim  1 , Kevin W Chang  1 , Le Lu  1 , Jianhua Yao  1 , Evrim B Turkbey  1 , Ronald M Summers  1
Affiliations

Automated segmentation of the thyroid gland on thoracic CT scans by multiatlas label fusion and random forest classification

Divya Narayanan et al. J Med Imaging (Bellingham). 2015 Oct.

Abstract

The thyroid is an endocrine gland that regulates metabolism. Thyroid image analysis plays an important role in both diagnostic radiology and radiation oncology treatment planning. Low tissue contrast of the thyroid relative to surrounding anatomic structures makes manual segmentation of this organ challenging. This work proposes a fully automated system for thyroid segmentation on CT imaging. Following initial thyroid segmentation with multiatlas joint label fusion, a random forest (RF) algorithm was applied. Multiatlas label fusion transfers labels from labeled atlases and warps them to target images using deformable registration. A consensus atlas solution was formed based on optimal weighting of atlases and similarity to a given target image. Following the initial segmentation, a trained RF classifier employed voxel scanning to assign class-conditional probabilities to the voxels in the target image. Thyroid voxels were categorized with positive labels and nonthyroid voxels were categorized with negative labels. Our method was evaluated on CT scans from 66 patients, 6 of which served as atlases for multiatlas label fusion. The system with independent multiatlas label fusion method and RF classifier achieved average dice similarity coefficients of [Formula: see text] and [Formula: see text], respectively. The system with sequential multiatlas label fusion followed by RF correction increased the dice similarity coefficient to [Formula: see text] and improved the segmentation accuracy.

Keywords: multiatlas label fusion; random forest; thyroid segmentation.

PubMed Disclaimer

Figures

Fig. 1
Fig. 1
(a) 3-D model of the thyroid (white), (b) thyroid in two-dimensional axial CT slice with (c) ground truth segmentation (pink).
Fig. 2
Fig. 2
Thyroid ROI. (a) x- and y-dimensions (red arrows) are determined by reference segmentations of the spinal column (ivory), spinal canal (teal), and trachea (dark blue), giving rise to the ROI (dashed yellow box). (b) Height of the ROI set to 6 cm (yellow arrow) around thyroid (pink).
Fig. 3
Fig. 3
Illustration of MALF.
Fig. 4
Fig. 4
(a) Thyroid ROI in target image. (b) Thyroid probability map L^T from multiatlas joint label fusion. (c) Initial estimated thyroid with surrounding voxels (all voxels with L^T≥0.10). (d) Thyroid probability map after RF correction.
Fig. 5
Fig. 5
(a) The RF is trained with positive samples (+) from all voxels of manually labeled thyroids and (b) negative samples (−) randomly selected from a surrounding shell.
Fig. 6
Fig. 6
MALF performance as a function of the number of atlases.
Fig. 7
Fig. 7
Box and whisker plot comparing DSC for three methods.
Fig. 8
Fig. 8
(a) Thyroid in ROI of CT scan. (b) Manually labeled thyroid (pink). (c) Segmentation from MALF (light blue). (d) Segmentation from MALF with RF correction (green). The improvements (yellow boxes) by the RF correction are highlighted in (d).
Fig. 9
Fig. 9
(a) Example of CT slice where thyroid is not visible. (b) Thyroid probability map from MALF. (c) Thyroid probability map from MALF with RF correction.
Fig. 10
Fig. 10
Three cases (a)–(c). (a) Thyroid in ROI, (b) manually labeled GT segmentations, and (c) comparable automated thyroid segmentations.
Fig. 11
Fig. 11
Three cases (a)–(c). (a) Thyroid in ROI, (b) manually labeled GT segmentations, and (c) poor automated thyroid segmentations.
Fig. 12
Fig. 12
Thyroid volume Bland–Altman plot comparing the automated method (MALF with RF correction) and ground truth (GT). The mean difference in volume (red line) and a 95% confidence interval (dotted red lines).
Fig. 13
Fig. 13
Thyroid segmentation in 3-D using MALF with RF correction (green) and GT (pink), with overlay between them (blue).
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Shabana W., Peeters E., De Maeseneer M., “Measuring thyroid gland volume: should we change the correction factor?” Am. J. Roentgenol. 186, 234–236 (2006).10.2214/AJR.04.0816 - DOI - PubMed
    1. Chen A., et al. , “Evaluation of multiple-atlas-based strategies for the segmentation of the thyroid gland in head and neck CT images for IMRT,” Phys. Med. Biol. 57, 93–111 (2012).10.1088/0031-9155/57/1/93 - DOI - PMC - PubMed
    1. Sethian J. A., “Level set methods and fast marching methods: evolving interfaces in computational geometry, fluid mechanics, computer vision, and materials science,” in Cambridge Monographs on Applied and Computational Mathematics, 2nd ed., p. 378, Cambridge University Press, Cambridge, UK: (1999).
    1. Boykov Y., Veksler O., Zabih R., “Fast approximate energy minimization via graph cuts,” IEEE Trans. Pattern Anal. Mach. Intell. 23, 1222–1239 (2001).10.1109/34.969114 - DOI
    1. Asman A. J., Landman B. A., “Non-local statistical label fusion for multi-atlas segmentation,” Med. Image Anal. 17, 194–208 (2013).10.1016/j.media.2012.10.002 - DOI - PMC - PubMed