Computer Aided Classification System for Breast Ultrasound Based on Breast Imaging Reporting and Data System (BI-RADS)

Wei Chih Shen, Ruey Feng Chang, Woo Kyung Moon

Research output: Contribution to journalArticle

42 Citations (Scopus)

Abstract

Clinically, the ultrasound findings are evaluated by its sonographic characteristics and then assigned to assessment categories according to the definitions of Breast Imaging Reporting and Data System (BI-RADS) developed by the American College of Radiology. In this study, a computer-aided classification (CAC) system was proposed to classify the masses into assessment categories 3, 4 and 5, which simulated the clinical diagnosis of radiologists. Compared with current computer-aided diagnosis systems, the proposed CAC system classifies the indeterminate cases into BI-RADS category 4 for further diagnosis. Six hundred twenty-six cases were collected from three ultrasound systems and confirmed by pathology and retrospectively classified into categories 3, 4 and 5 by radiologists. The multinomial logistic regression model was trained as the CAC system for predicting the assessment category from the computerized BI-RADS features and from a set of machine-dependent factors. By using the machine-dependent factors to indicate the adopted ultrasound systems, the same regression model could be applied for the cases acquired from different ultrasound systems. A basic CAC system was trained by using the classification result of radiologists. A weighted CAC system, to improve the capacity of the basic CAC system in differentiating benign from malignant lesions, was trained by adding the pathologic result. Between the radiologists and the basic CAC system, a substantial agreement was indicated by Cohen's kappa statistic and the differences in either the performance indices or the AZ of receiver operating characteristic (ROC) analysis were not statistically significant. For the weighted CAC system, the performance indices accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 73.00% (457 of 626), 98.17% (215 of 219), 59.46% (242 of 407), 56.58% (215 of 380) and 98.37% (242 of 246), respectively; the AZ was 0.94; and the correlation with the radiologists was also substantial agreement. The indices accuracy and specificity of weighted CAC system, compared with those of the radiologists, were improved by 5.91% and 8.85%, respectively and the indices of sensitivity and NPV, compared with those of a conventional CAD system, were improved by 10.5% and 5.21%, respectively; all improvements were statistically significant. To classify the mass into BI-RADS assessment categories by the CAC system is feasible. Moreover, the proposed CAC system is flexible because it can be used to diagnose the cases acquired from different ultrasound systems. (E-mail: rfchang@csie.ntu.edu.tw).

Original languageEnglish
Pages (from-to)1688-1698
Number of pages11
JournalUltrasound in Medicine and Biology
Volume33
Issue number11
DOIs
StatePublished - 1 Nov 2007

Fingerprint

Information Systems
Breast
Logistic Models
Computer Systems
Radiology
ROC Curve
Radiologists
Pathology
Sensitivity and Specificity

Keywords

  • BI-RADS
  • Breast cancer
  • Computer-aided classification (CAC) system
  • Computer-aided diagnosis (CAD) system
  • Logistic regression
  • Ultrasound image

Cite this

@article{d23f26d8b80046638f05d1ed66c76cb6,
title = "Computer Aided Classification System for Breast Ultrasound Based on Breast Imaging Reporting and Data System (BI-RADS)",
abstract = "Clinically, the ultrasound findings are evaluated by its sonographic characteristics and then assigned to assessment categories according to the definitions of Breast Imaging Reporting and Data System (BI-RADS) developed by the American College of Radiology. In this study, a computer-aided classification (CAC) system was proposed to classify the masses into assessment categories 3, 4 and 5, which simulated the clinical diagnosis of radiologists. Compared with current computer-aided diagnosis systems, the proposed CAC system classifies the indeterminate cases into BI-RADS category 4 for further diagnosis. Six hundred twenty-six cases were collected from three ultrasound systems and confirmed by pathology and retrospectively classified into categories 3, 4 and 5 by radiologists. The multinomial logistic regression model was trained as the CAC system for predicting the assessment category from the computerized BI-RADS features and from a set of machine-dependent factors. By using the machine-dependent factors to indicate the adopted ultrasound systems, the same regression model could be applied for the cases acquired from different ultrasound systems. A basic CAC system was trained by using the classification result of radiologists. A weighted CAC system, to improve the capacity of the basic CAC system in differentiating benign from malignant lesions, was trained by adding the pathologic result. Between the radiologists and the basic CAC system, a substantial agreement was indicated by Cohen's kappa statistic and the differences in either the performance indices or the AZ of receiver operating characteristic (ROC) analysis were not statistically significant. For the weighted CAC system, the performance indices accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 73.00{\%} (457 of 626), 98.17{\%} (215 of 219), 59.46{\%} (242 of 407), 56.58{\%} (215 of 380) and 98.37{\%} (242 of 246), respectively; the AZ was 0.94; and the correlation with the radiologists was also substantial agreement. The indices accuracy and specificity of weighted CAC system, compared with those of the radiologists, were improved by 5.91{\%} and 8.85{\%}, respectively and the indices of sensitivity and NPV, compared with those of a conventional CAD system, were improved by 10.5{\%} and 5.21{\%}, respectively; all improvements were statistically significant. To classify the mass into BI-RADS assessment categories by the CAC system is feasible. Moreover, the proposed CAC system is flexible because it can be used to diagnose the cases acquired from different ultrasound systems. (E-mail: rfchang@csie.ntu.edu.tw).",
keywords = "BI-RADS, Breast cancer, Computer-aided classification (CAC) system, Computer-aided diagnosis (CAD) system, Logistic regression, Ultrasound image",
author = "Shen, {Wei Chih} and Chang, {Ruey Feng} and Moon, {Woo Kyung}",
year = "2007",
month = "11",
day = "1",
doi = "10.1016/j.ultrasmedbio.2007.05.016",
language = "English",
volume = "33",
pages = "1688--1698",
journal = "Ultrasound in Medicine and Biology",
issn = "0301-5629",
publisher = "Elsevier USA",
number = "11",

}

Computer Aided Classification System for Breast Ultrasound Based on Breast Imaging Reporting and Data System (BI-RADS). / Shen, Wei Chih; Chang, Ruey Feng; Moon, Woo Kyung.

In: Ultrasound in Medicine and Biology, Vol. 33, No. 11, 01.11.2007, p. 1688-1698.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Computer Aided Classification System for Breast Ultrasound Based on Breast Imaging Reporting and Data System (BI-RADS)

AU - Shen, Wei Chih

AU - Chang, Ruey Feng

AU - Moon, Woo Kyung

PY - 2007/11/1

Y1 - 2007/11/1

N2 - Clinically, the ultrasound findings are evaluated by its sonographic characteristics and then assigned to assessment categories according to the definitions of Breast Imaging Reporting and Data System (BI-RADS) developed by the American College of Radiology. In this study, a computer-aided classification (CAC) system was proposed to classify the masses into assessment categories 3, 4 and 5, which simulated the clinical diagnosis of radiologists. Compared with current computer-aided diagnosis systems, the proposed CAC system classifies the indeterminate cases into BI-RADS category 4 for further diagnosis. Six hundred twenty-six cases were collected from three ultrasound systems and confirmed by pathology and retrospectively classified into categories 3, 4 and 5 by radiologists. The multinomial logistic regression model was trained as the CAC system for predicting the assessment category from the computerized BI-RADS features and from a set of machine-dependent factors. By using the machine-dependent factors to indicate the adopted ultrasound systems, the same regression model could be applied for the cases acquired from different ultrasound systems. A basic CAC system was trained by using the classification result of radiologists. A weighted CAC system, to improve the capacity of the basic CAC system in differentiating benign from malignant lesions, was trained by adding the pathologic result. Between the radiologists and the basic CAC system, a substantial agreement was indicated by Cohen's kappa statistic and the differences in either the performance indices or the AZ of receiver operating characteristic (ROC) analysis were not statistically significant. For the weighted CAC system, the performance indices accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 73.00% (457 of 626), 98.17% (215 of 219), 59.46% (242 of 407), 56.58% (215 of 380) and 98.37% (242 of 246), respectively; the AZ was 0.94; and the correlation with the radiologists was also substantial agreement. The indices accuracy and specificity of weighted CAC system, compared with those of the radiologists, were improved by 5.91% and 8.85%, respectively and the indices of sensitivity and NPV, compared with those of a conventional CAD system, were improved by 10.5% and 5.21%, respectively; all improvements were statistically significant. To classify the mass into BI-RADS assessment categories by the CAC system is feasible. Moreover, the proposed CAC system is flexible because it can be used to diagnose the cases acquired from different ultrasound systems. (E-mail: rfchang@csie.ntu.edu.tw).

AB - Clinically, the ultrasound findings are evaluated by its sonographic characteristics and then assigned to assessment categories according to the definitions of Breast Imaging Reporting and Data System (BI-RADS) developed by the American College of Radiology. In this study, a computer-aided classification (CAC) system was proposed to classify the masses into assessment categories 3, 4 and 5, which simulated the clinical diagnosis of radiologists. Compared with current computer-aided diagnosis systems, the proposed CAC system classifies the indeterminate cases into BI-RADS category 4 for further diagnosis. Six hundred twenty-six cases were collected from three ultrasound systems and confirmed by pathology and retrospectively classified into categories 3, 4 and 5 by radiologists. The multinomial logistic regression model was trained as the CAC system for predicting the assessment category from the computerized BI-RADS features and from a set of machine-dependent factors. By using the machine-dependent factors to indicate the adopted ultrasound systems, the same regression model could be applied for the cases acquired from different ultrasound systems. A basic CAC system was trained by using the classification result of radiologists. A weighted CAC system, to improve the capacity of the basic CAC system in differentiating benign from malignant lesions, was trained by adding the pathologic result. Between the radiologists and the basic CAC system, a substantial agreement was indicated by Cohen's kappa statistic and the differences in either the performance indices or the AZ of receiver operating characteristic (ROC) analysis were not statistically significant. For the weighted CAC system, the performance indices accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 73.00% (457 of 626), 98.17% (215 of 219), 59.46% (242 of 407), 56.58% (215 of 380) and 98.37% (242 of 246), respectively; the AZ was 0.94; and the correlation with the radiologists was also substantial agreement. The indices accuracy and specificity of weighted CAC system, compared with those of the radiologists, were improved by 5.91% and 8.85%, respectively and the indices of sensitivity and NPV, compared with those of a conventional CAD system, were improved by 10.5% and 5.21%, respectively; all improvements were statistically significant. To classify the mass into BI-RADS assessment categories by the CAC system is feasible. Moreover, the proposed CAC system is flexible because it can be used to diagnose the cases acquired from different ultrasound systems. (E-mail: rfchang@csie.ntu.edu.tw).

KW - BI-RADS

KW - Breast cancer

KW - Computer-aided classification (CAC) system

KW - Computer-aided diagnosis (CAD) system

KW - Logistic regression

KW - Ultrasound image

UR - http://www.scopus.com/inward/record.url?scp=35449000163&partnerID=8YFLogxK

U2 - 10.1016/j.ultrasmedbio.2007.05.016

DO - 10.1016/j.ultrasmedbio.2007.05.016

M3 - Article

C2 - 17681678

AN - SCOPUS:35449000163

VL - 33

SP - 1688

EP - 1698

JO - Ultrasound in Medicine and Biology

JF - Ultrasound in Medicine and Biology

SN - 0301-5629

IS - 11

ER -