Dissimilarity Representations in Lung Parenchyma Classification

Lauge Emil Borch Laurs Sørensen; Marleen de Bruijne

doi:10.1117/12.812505

Dissimilarity Representations in Lung Parenchyma Classification

Lauge Emil Borch Laurs Sørensen, Marleen de Bruijne

Department of Computer Science

5 Citations (Scopus)

Abstract

Agood problem representation is important for a pattern recognition systemto be successful. The traditional approach to statistical pattern recognitionis feature representation. More specifically, objects are represented by anumber of features in a feature vector space, and classifiersare built in this representation. This is also the generaltrend in lung parenchyma classification in computed tomography (CT) images,where the features often are measures on feature histograms. Instead,we propose to build normal density based classifiers in dissimilarityrepresentations for lung parenchyma classification. This allows for the classifiersto work on dissimilarities between objects, which might be amore natural way of representing lung parenchyma. In this context,dissimilarity is defined between CT regions of interest (ROI)s. ROIsare represented by their CT attenuation histogram and ROI dissimilarityis defined as a histogram dissimilarity measure between the attenuationhistograms. In this setting, the full histograms are utilized accordingto the chosen histogram dissimilarity measure. We apply this ideato classification of different emphysema patterns as well as normal,healthy tissue. Two dissimilarity representation approaches as well as differenthistogram dissimilarity measures are considered. The approaches are evaluated ona set of 168 CT ROIs using normal density basedclassifiers all showing good performance. Compared to using histogram dissimilaritydirectly as distance in a emph{k} nearest neighbor classifier, whichachieves a classification accuracy of $92.9%$, the best dissimilarity representationbased classifier is significantly better with a classification accuracy of97.0% ($text{emph{p" border="0" class="imgtopleft"> = 0.046$).

Original language	English
Title of host publication	Medical Imaging 2009 : Computer-Aided Diagnosis
Editors	Nico Karssemeijer, Maryellen L. Giger
Number of pages	12
Publisher	SPIE - International Society for Optical Engineering
Publication date	2009
DOIs	https://doi.org/10.1117/12.812505
Publication status	Published - 2009
Event	SPIE Medical Imaging - Lake Buena Vista, United States Duration: 7 Feb 2009 → 12 Feb 2009

Conference

Conference	SPIE Medical Imaging
Country/Territory	United States
City	Lake Buena Vista
Period	07/02/2009 → 12/02/2009

Series	Proceedings of SPIE
Number	7260

Access to Document

10.1117/12.812505

Cite this

Medical Imaging 2009: Computer-Aided Diagnosis. ed. / Nico Karssemeijer; Maryellen L. Giger. SPIE - International Society for Optical Engineering, 2009. (Proceedings of SPIE; No. 7260).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Sørensen, LEBL & de Bruijne, M 2009, Dissimilarity Representations in Lung Parenchyma Classification. in N Karssemeijer & ML Giger (eds), Medical Imaging 2009: Computer-Aided Diagnosis. SPIE - International Society for Optical Engineering, Proceedings of SPIE, no. 7260, SPIE Medical Imaging, Lake Buena Vista, United States, 07/02/2009. https://doi.org/10.1117/12.812505

@inproceedings{9d5f7310a9a111ddb5e9000ea68e967b,

title = "Dissimilarity Representations in Lung Parenchyma Classification",

abstract = "A good problem representation is important for a pattern recognition system to be successful. The traditional approach to statistical pattern recognition is feature representation. More specifically, objects are represented by a number of features in a feature vector space, and classifiers are built in this representation. This is also the general trend in lung parenchyma classification in computed tomography (CT) images, where the features often are measures on feature histograms. Instead, we propose to build normal density based classifiers in dissimilarity representations for lung parenchyma classification. This allows for the classifiers to work on dissimilarities between objects, which might be a more natural way of representing lung parenchyma. In this context, dissimilarity is defined between CT regions of interest (ROI)s. ROIs are represented by their CT attenuation histogram and ROI dissimilarity is defined as a histogram dissimilarity measure between the attenuation histograms. In this setting, the full histograms are utilized according to the chosen histogram dissimilarity measure. We apply this idea to classification of different emphysema patterns as well as normal, healthy tissue. Two dissimilarity representation approaches as well as different histogram dissimilarity measures are considered. The approaches are evaluated on a set of 168 CT ROIs using normal density based classifiers all showing good performance. Compared to using histogram dissimilarity directly as distance in a emph{k} nearest neighbor classifier, which achieves a classification accuracy of $92.9%$, the best dissimilarity representation based classifier is significantly better with a classification accuracy of 97.0% ($text{emph{p{"} border={"}0{"} class={"}imgtopleft{"}> = 0.046$). ",

author = "S{\o}rensen, {Lauge Emil Borch Laurs} and {de Bruijne}, Marleen",

year = "2009",

doi = "10.1117/12.812505",

language = "English",

series = "Proceedings of SPIE",

publisher = "SPIE - International Society for Optical Engineering",

number = "7260",

editor = "Nico Karssemeijer and Giger, {Maryellen L.}",

booktitle = "Medical Imaging 2009",

note = "SPIE Medical Imaging ; Conference date: 07-02-2009 Through 12-02-2009",

}

TY - GEN

T1 - Dissimilarity Representations in Lung Parenchyma Classification

AU - Sørensen, Lauge Emil Borch Laurs

AU - de Bruijne, Marleen

PY - 2009

Y1 - 2009

N2 - A good problem representation is important for a pattern recognition system to be successful. The traditional approach to statistical pattern recognition is feature representation. More specifically, objects are represented by a number of features in a feature vector space, and classifiers are built in this representation. This is also the general trend in lung parenchyma classification in computed tomography (CT) images, where the features often are measures on feature histograms. Instead, we propose to build normal density based classifiers in dissimilarity representations for lung parenchyma classification. This allows for the classifiers to work on dissimilarities between objects, which might be a more natural way of representing lung parenchyma. In this context, dissimilarity is defined between CT regions of interest (ROI)s. ROIs are represented by their CT attenuation histogram and ROI dissimilarity is defined as a histogram dissimilarity measure between the attenuation histograms. In this setting, the full histograms are utilized according to the chosen histogram dissimilarity measure. We apply this idea to classification of different emphysema patterns as well as normal, healthy tissue. Two dissimilarity representation approaches as well as different histogram dissimilarity measures are considered. The approaches are evaluated on a set of 168 CT ROIs using normal density based classifiers all showing good performance. Compared to using histogram dissimilarity directly as distance in a emph{k} nearest neighbor classifier, which achieves a classification accuracy of $92.9%$, the best dissimilarity representation based classifier is significantly better with a classification accuracy of 97.0% ($text{emph{p" border="0" class="imgtopleft"> = 0.046$).

AB - A good problem representation is important for a pattern recognition system to be successful. The traditional approach to statistical pattern recognition is feature representation. More specifically, objects are represented by a number of features in a feature vector space, and classifiers are built in this representation. This is also the general trend in lung parenchyma classification in computed tomography (CT) images, where the features often are measures on feature histograms. Instead, we propose to build normal density based classifiers in dissimilarity representations for lung parenchyma classification. This allows for the classifiers to work on dissimilarities between objects, which might be a more natural way of representing lung parenchyma. In this context, dissimilarity is defined between CT regions of interest (ROI)s. ROIs are represented by their CT attenuation histogram and ROI dissimilarity is defined as a histogram dissimilarity measure between the attenuation histograms. In this setting, the full histograms are utilized according to the chosen histogram dissimilarity measure. We apply this idea to classification of different emphysema patterns as well as normal, healthy tissue. Two dissimilarity representation approaches as well as different histogram dissimilarity measures are considered. The approaches are evaluated on a set of 168 CT ROIs using normal density based classifiers all showing good performance. Compared to using histogram dissimilarity directly as distance in a emph{k} nearest neighbor classifier, which achieves a classification accuracy of $92.9%$, the best dissimilarity representation based classifier is significantly better with a classification accuracy of 97.0% ($text{emph{p" border="0" class="imgtopleft"> = 0.046$).

U2 - 10.1117/12.812505

DO - 10.1117/12.812505

M3 - Article in proceedings

T3 - Proceedings of SPIE

BT - Medical Imaging 2009

A2 - Karssemeijer, Nico

A2 - Giger, Maryellen L.

PB - SPIE - International Society for Optical Engineering

T2 - SPIE Medical Imaging

Y2 - 7 February 2009 through 12 February 2009

ER -

Dissimilarity Representations in Lung Parenchyma Classification

Abstract

Conference

Access to Document

Fingerprint

Cite this