Pseudo-color visualizations of DCE-MR image series for MR mammograpy

T. Twellmann, T.W. Nattkemper, M. Heisen, B.M. Haar Romenij, ter

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademic

Abstract

In recent years, dynamic contrast-enhanced magnetic resonance (DCE-MR) imaging has become a valuable tool for detection, diagnosis and management of breast cancer. Several criteria for describing morphologic and dynamic characteristics of suspiciously enhancing tissue regions have been collected in the ACR BIRADS MRI lexicon. However, evaluation of these criteria is nonetheless a challenging task for human observers due to the huge amount and the mutlitemporal nature of the image data. Therefore, computer aided diagnosis (CAD) tools based on artificial neural networks (ANN) or pharamcokinetic models receive growing attention from the radiologic community. In DCE-MR imaging, each voxel is associated with a vector s = (s1, . . . , st) reflecting the temporal variation of the local signal intensity after intravenous administration of a contrast agent (Gd-DTPA). Due to changes in their vascular structure, benign and malignant tissue expose characteristic intensity-time curves (ITC). These curves enable radiologists to infer information about the tissue state from the image data, a time-consuming task owing to the heterogeneity of cancerous tissue. To aid evaluation of DCE-MR image series, we propose a pseudo-color visualization of the temporal information based on ANNs. An ANN is trained with labeled ITCs sampled from a number of histologically verified training cases to classify each temporal signal sx,y,z as being indicative for malignant (m), normal (n) or benign (b) tissue according to the returned posteriori probabilities p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z). Pseudo-color visualizations of unseen image series are computed by displaying suspiciously enhancing voxels with RGB colors reflecting the ANN based signal assessment: bright red, green and blue voxels indicate high p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z) values, respectively. Therewith, temporal characteristics of tissue regions are revealed, enabling radiologists to assess the architecture of lesions by means of a single 3D color image.
Original languageEnglish
Title of host publicationBook of Abstracts, First Dutch Conference on Bio-Medical Engineering
Place of PublicationNetherlands, Egmond aan Zee
Pages81-
Publication statusPublished - 2007

Fingerprint

Magnetic Resonance Spectroscopy
Color
Magnetic Resonance Imaging
Gadolinium DTPA
Intravenous Administration
Contrast Media
Blood Vessels
Breast Neoplasms
Radiologists

Cite this

Twellmann, T., Nattkemper, T. W., Heisen, M., & Haar Romenij, ter, B. M. (2007). Pseudo-color visualizations of DCE-MR image series for MR mammograpy. In Book of Abstracts, First Dutch Conference on Bio-Medical Engineering (pp. 81-). Netherlands, Egmond aan Zee.
Twellmann, T. ; Nattkemper, T.W. ; Heisen, M. ; Haar Romenij, ter, B.M. / Pseudo-color visualizations of DCE-MR image series for MR mammograpy. Book of Abstracts, First Dutch Conference on Bio-Medical Engineering. Netherlands, Egmond aan Zee, 2007. pp. 81-
@inproceedings{71affa6f426a4ae4b207e9f8e28a746e,
title = "Pseudo-color visualizations of DCE-MR image series for MR mammograpy",
abstract = "In recent years, dynamic contrast-enhanced magnetic resonance (DCE-MR) imaging has become a valuable tool for detection, diagnosis and management of breast cancer. Several criteria for describing morphologic and dynamic characteristics of suspiciously enhancing tissue regions have been collected in the ACR BIRADS MRI lexicon. However, evaluation of these criteria is nonetheless a challenging task for human observers due to the huge amount and the mutlitemporal nature of the image data. Therefore, computer aided diagnosis (CAD) tools based on artificial neural networks (ANN) or pharamcokinetic models receive growing attention from the radiologic community. In DCE-MR imaging, each voxel is associated with a vector s = (s1, . . . , st) reflecting the temporal variation of the local signal intensity after intravenous administration of a contrast agent (Gd-DTPA). Due to changes in their vascular structure, benign and malignant tissue expose characteristic intensity-time curves (ITC). These curves enable radiologists to infer information about the tissue state from the image data, a time-consuming task owing to the heterogeneity of cancerous tissue. To aid evaluation of DCE-MR image series, we propose a pseudo-color visualization of the temporal information based on ANNs. An ANN is trained with labeled ITCs sampled from a number of histologically verified training cases to classify each temporal signal sx,y,z as being indicative for malignant (m), normal (n) or benign (b) tissue according to the returned posteriori probabilities p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z). Pseudo-color visualizations of unseen image series are computed by displaying suspiciously enhancing voxels with RGB colors reflecting the ANN based signal assessment: bright red, green and blue voxels indicate high p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z) values, respectively. Therewith, temporal characteristics of tissue regions are revealed, enabling radiologists to assess the architecture of lesions by means of a single 3D color image.",
author = "T. Twellmann and T.W. Nattkemper and M. Heisen and {Haar Romenij, ter}, B.M.",
year = "2007",
language = "English",
pages = "81--",
booktitle = "Book of Abstracts, First Dutch Conference on Bio-Medical Engineering",

}

Twellmann, T, Nattkemper, TW, Heisen, M & Haar Romenij, ter, BM 2007, Pseudo-color visualizations of DCE-MR image series for MR mammograpy. in Book of Abstracts, First Dutch Conference on Bio-Medical Engineering. Netherlands, Egmond aan Zee, pp. 81-.

Pseudo-color visualizations of DCE-MR image series for MR mammograpy. / Twellmann, T.; Nattkemper, T.W.; Heisen, M.; Haar Romenij, ter, B.M.

Book of Abstracts, First Dutch Conference on Bio-Medical Engineering. Netherlands, Egmond aan Zee, 2007. p. 81-.

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademic

TY - GEN

T1 - Pseudo-color visualizations of DCE-MR image series for MR mammograpy

AU - Twellmann, T.

AU - Nattkemper, T.W.

AU - Heisen, M.

AU - Haar Romenij, ter, B.M.

PY - 2007

Y1 - 2007

N2 - In recent years, dynamic contrast-enhanced magnetic resonance (DCE-MR) imaging has become a valuable tool for detection, diagnosis and management of breast cancer. Several criteria for describing morphologic and dynamic characteristics of suspiciously enhancing tissue regions have been collected in the ACR BIRADS MRI lexicon. However, evaluation of these criteria is nonetheless a challenging task for human observers due to the huge amount and the mutlitemporal nature of the image data. Therefore, computer aided diagnosis (CAD) tools based on artificial neural networks (ANN) or pharamcokinetic models receive growing attention from the radiologic community. In DCE-MR imaging, each voxel is associated with a vector s = (s1, . . . , st) reflecting the temporal variation of the local signal intensity after intravenous administration of a contrast agent (Gd-DTPA). Due to changes in their vascular structure, benign and malignant tissue expose characteristic intensity-time curves (ITC). These curves enable radiologists to infer information about the tissue state from the image data, a time-consuming task owing to the heterogeneity of cancerous tissue. To aid evaluation of DCE-MR image series, we propose a pseudo-color visualization of the temporal information based on ANNs. An ANN is trained with labeled ITCs sampled from a number of histologically verified training cases to classify each temporal signal sx,y,z as being indicative for malignant (m), normal (n) or benign (b) tissue according to the returned posteriori probabilities p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z). Pseudo-color visualizations of unseen image series are computed by displaying suspiciously enhancing voxels with RGB colors reflecting the ANN based signal assessment: bright red, green and blue voxels indicate high p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z) values, respectively. Therewith, temporal characteristics of tissue regions are revealed, enabling radiologists to assess the architecture of lesions by means of a single 3D color image.

AB - In recent years, dynamic contrast-enhanced magnetic resonance (DCE-MR) imaging has become a valuable tool for detection, diagnosis and management of breast cancer. Several criteria for describing morphologic and dynamic characteristics of suspiciously enhancing tissue regions have been collected in the ACR BIRADS MRI lexicon. However, evaluation of these criteria is nonetheless a challenging task for human observers due to the huge amount and the mutlitemporal nature of the image data. Therefore, computer aided diagnosis (CAD) tools based on artificial neural networks (ANN) or pharamcokinetic models receive growing attention from the radiologic community. In DCE-MR imaging, each voxel is associated with a vector s = (s1, . . . , st) reflecting the temporal variation of the local signal intensity after intravenous administration of a contrast agent (Gd-DTPA). Due to changes in their vascular structure, benign and malignant tissue expose characteristic intensity-time curves (ITC). These curves enable radiologists to infer information about the tissue state from the image data, a time-consuming task owing to the heterogeneity of cancerous tissue. To aid evaluation of DCE-MR image series, we propose a pseudo-color visualization of the temporal information based on ANNs. An ANN is trained with labeled ITCs sampled from a number of histologically verified training cases to classify each temporal signal sx,y,z as being indicative for malignant (m), normal (n) or benign (b) tissue according to the returned posteriori probabilities p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z). Pseudo-color visualizations of unseen image series are computed by displaying suspiciously enhancing voxels with RGB colors reflecting the ANN based signal assessment: bright red, green and blue voxels indicate high p(m|s_x,y,z), p(n|s_x,y,z) and p(b|s_x,y,z) values, respectively. Therewith, temporal characteristics of tissue regions are revealed, enabling radiologists to assess the architecture of lesions by means of a single 3D color image.

M3 - Conference contribution

SP - 81-

BT - Book of Abstracts, First Dutch Conference on Bio-Medical Engineering

CY - Netherlands, Egmond aan Zee

ER -

Twellmann T, Nattkemper TW, Heisen M, Haar Romenij, ter BM. Pseudo-color visualizations of DCE-MR image series for MR mammograpy. In Book of Abstracts, First Dutch Conference on Bio-Medical Engineering. Netherlands, Egmond aan Zee. 2007. p. 81-