Specificity improvement of a CAD system for multiparametric MR prostate cancer using texture features and artificial neural networks

Research output: Contribution to journalArticle

Abstract

Prostate cancer (PCa) is the most common cancer afflicting men in USA. Multiparametric Magnetic Resonance imaging is recently emerging as a powerful tool for PCa diagnosis, but its analysis and interpretation is time-consuming and affected by the radiologist experience. Computer aided detection (CAD) systems have been developed to overcome this limitation and to support radiologists in the PCa diagnosis. Although several studies proposed CAD systems with very high performances in terms of sensitivity, the analysis of false positive (FP) areas is usually not clearly presented. The aim of this study is to improve the performance of a CAD system in term of reduction of FPs findings, without affecting the sensitivity. To this scope, we developed a classifier composed by 3 Artificial Neural Networks (ANN) able to distinguish between malignant and healthy areas through a voting strategy. In this method, we exploit the role of the Gray Level Co-occurrence Matrix, the Gray Level Difference Method and Gray Level Run Length Method Matrix in differentiating tumoural from healthy tissues. We first extract 64 textural features from T2-weighted (T2w) images and the apparent diffusion coefficient (ADC) maps, then we discretized them to reduce the data variability. A features selection method, based on the correlation matrix, is finally applied to remove redundant variables, that are those highly correlated with others. The remaining set of features is fed into the three ANNs and a post-processing step is applied to remove very small areas. Results applied on a dataset of 58 patients showed a significant decrease of FPs (20 vs 12; p-value < 0.0001) and an increase of the precision of PCa segmentation (0.62 vs 0.71; p-value < 0.0001). Having less FPs is helpful to increase the performance of CAD systems in terms of specificity and to decrease the reporting time of radiologists. Moreover, having more precise PCa segmentation areas could be useful if a step of PCa characterization will be added to the CAD system.

Original languageEnglish
Pages (from-to)71-80
Number of pages10
JournalHealth and Technology
Volume7
Issue number1
DOIs
Publication statusPublished - Mar 1 2017

Fingerprint

Prostatic Neoplasms
Textures
Neural networks
Magnetic resonance
Politics
Feature extraction
Classifiers
Tissue
Imaging techniques
Magnetic Resonance Imaging
Processing
Radiologists
Neoplasms

Keywords

  • Artificial neural networks
  • CAD system
  • Multiparametric MRI
  • Prostate cancer
  • Texture features

ASJC Scopus subject areas

  • Biotechnology
  • Bioengineering
  • Applied Microbiology and Biotechnology
  • Biomedical Engineering

Cite this

@article{4ad40f0ee8564fa4b2d2b0ade2374ce5,
title = "Specificity improvement of a CAD system for multiparametric MR prostate cancer using texture features and artificial neural networks",
abstract = "Prostate cancer (PCa) is the most common cancer afflicting men in USA. Multiparametric Magnetic Resonance imaging is recently emerging as a powerful tool for PCa diagnosis, but its analysis and interpretation is time-consuming and affected by the radiologist experience. Computer aided detection (CAD) systems have been developed to overcome this limitation and to support radiologists in the PCa diagnosis. Although several studies proposed CAD systems with very high performances in terms of sensitivity, the analysis of false positive (FP) areas is usually not clearly presented. The aim of this study is to improve the performance of a CAD system in term of reduction of FPs findings, without affecting the sensitivity. To this scope, we developed a classifier composed by 3 Artificial Neural Networks (ANN) able to distinguish between malignant and healthy areas through a voting strategy. In this method, we exploit the role of the Gray Level Co-occurrence Matrix, the Gray Level Difference Method and Gray Level Run Length Method Matrix in differentiating tumoural from healthy tissues. We first extract 64 textural features from T2-weighted (T2w) images and the apparent diffusion coefficient (ADC) maps, then we discretized them to reduce the data variability. A features selection method, based on the correlation matrix, is finally applied to remove redundant variables, that are those highly correlated with others. The remaining set of features is fed into the three ANNs and a post-processing step is applied to remove very small areas. Results applied on a dataset of 58 patients showed a significant decrease of FPs (20 vs 12; p-value < 0.0001) and an increase of the precision of PCa segmentation (0.62 vs 0.71; p-value < 0.0001). Having less FPs is helpful to increase the performance of CAD systems in terms of specificity and to decrease the reporting time of radiologists. Moreover, having more precise PCa segmentation areas could be useful if a step of PCa characterization will be added to the CAD system.",
keywords = "Artificial neural networks, CAD system, Multiparametric MRI, Prostate cancer, Texture features",
author = "V. Giannini and S. Rosati and D. Regge and G. Balestra",
year = "2017",
month = "3",
day = "1",
doi = "10.1007/s12553-016-0150-6",
language = "English",
volume = "7",
pages = "71--80",
journal = "Health and Technology",
issn = "2190-7188",
publisher = "Springer Verlag",
number = "1",

}

TY - JOUR

T1 - Specificity improvement of a CAD system for multiparametric MR prostate cancer using texture features and artificial neural networks

AU - Giannini, V.

AU - Rosati, S.

AU - Regge, D.

AU - Balestra, G.

PY - 2017/3/1

Y1 - 2017/3/1

N2 - Prostate cancer (PCa) is the most common cancer afflicting men in USA. Multiparametric Magnetic Resonance imaging is recently emerging as a powerful tool for PCa diagnosis, but its analysis and interpretation is time-consuming and affected by the radiologist experience. Computer aided detection (CAD) systems have been developed to overcome this limitation and to support radiologists in the PCa diagnosis. Although several studies proposed CAD systems with very high performances in terms of sensitivity, the analysis of false positive (FP) areas is usually not clearly presented. The aim of this study is to improve the performance of a CAD system in term of reduction of FPs findings, without affecting the sensitivity. To this scope, we developed a classifier composed by 3 Artificial Neural Networks (ANN) able to distinguish between malignant and healthy areas through a voting strategy. In this method, we exploit the role of the Gray Level Co-occurrence Matrix, the Gray Level Difference Method and Gray Level Run Length Method Matrix in differentiating tumoural from healthy tissues. We first extract 64 textural features from T2-weighted (T2w) images and the apparent diffusion coefficient (ADC) maps, then we discretized them to reduce the data variability. A features selection method, based on the correlation matrix, is finally applied to remove redundant variables, that are those highly correlated with others. The remaining set of features is fed into the three ANNs and a post-processing step is applied to remove very small areas. Results applied on a dataset of 58 patients showed a significant decrease of FPs (20 vs 12; p-value < 0.0001) and an increase of the precision of PCa segmentation (0.62 vs 0.71; p-value < 0.0001). Having less FPs is helpful to increase the performance of CAD systems in terms of specificity and to decrease the reporting time of radiologists. Moreover, having more precise PCa segmentation areas could be useful if a step of PCa characterization will be added to the CAD system.

AB - Prostate cancer (PCa) is the most common cancer afflicting men in USA. Multiparametric Magnetic Resonance imaging is recently emerging as a powerful tool for PCa diagnosis, but its analysis and interpretation is time-consuming and affected by the radiologist experience. Computer aided detection (CAD) systems have been developed to overcome this limitation and to support radiologists in the PCa diagnosis. Although several studies proposed CAD systems with very high performances in terms of sensitivity, the analysis of false positive (FP) areas is usually not clearly presented. The aim of this study is to improve the performance of a CAD system in term of reduction of FPs findings, without affecting the sensitivity. To this scope, we developed a classifier composed by 3 Artificial Neural Networks (ANN) able to distinguish between malignant and healthy areas through a voting strategy. In this method, we exploit the role of the Gray Level Co-occurrence Matrix, the Gray Level Difference Method and Gray Level Run Length Method Matrix in differentiating tumoural from healthy tissues. We first extract 64 textural features from T2-weighted (T2w) images and the apparent diffusion coefficient (ADC) maps, then we discretized them to reduce the data variability. A features selection method, based on the correlation matrix, is finally applied to remove redundant variables, that are those highly correlated with others. The remaining set of features is fed into the three ANNs and a post-processing step is applied to remove very small areas. Results applied on a dataset of 58 patients showed a significant decrease of FPs (20 vs 12; p-value < 0.0001) and an increase of the precision of PCa segmentation (0.62 vs 0.71; p-value < 0.0001). Having less FPs is helpful to increase the performance of CAD systems in terms of specificity and to decrease the reporting time of radiologists. Moreover, having more precise PCa segmentation areas could be useful if a step of PCa characterization will be added to the CAD system.

KW - Artificial neural networks

KW - CAD system

KW - Multiparametric MRI

KW - Prostate cancer

KW - Texture features

UR - http://www.scopus.com/inward/record.url?scp=85015150746&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85015150746&partnerID=8YFLogxK

U2 - 10.1007/s12553-016-0150-6

DO - 10.1007/s12553-016-0150-6

M3 - Article

AN - SCOPUS:85015150746

VL - 7

SP - 71

EP - 80

JO - Health and Technology

JF - Health and Technology

SN - 2190-7188

IS - 1

ER -