Comparison of artificial neural networks and regression models for the classification of breast and ovarian tumors by color Doppler sonography

Dr. W. Sauerbrei, FDM, Institute of Medical Biometry and Medical Informatics

Dipl. Math. R. Roßner, Institute of Medical Biometry and Medical Informatics

Prof. Dr. M. Schumacher, FDM, Institute of Medical Biometry and Medical Informatics

Prof. Dr. W. Vach, Department of Statistics and Demography, Odense University

PD Dr. H. Madjar, Department of Gynecology, University Hospital Freiburg

PD Dr. H. Prömpeler, Department of Gynecology, University Hospital Freiburg

• Actual and future projects
  

In the time period between 1992 and 1994, about 450 patients with breast tumors have been examined in the Department of Gynecology with color Doppler sonography in order to differentiate between benign and malignant lesions. Histological verification was used as "gold standard" (1). Based on a logistic regression model and on a classification and regression tree (CART) approach, simple classification rules were developed having high sensitivity as well as high specificity (3). In addition, a comparison of these statistical methods with the application of artificial neural networks has been performed in order to evaluate whether the latter can provide classification rules with greater accuracy (4,5). In subsequent studies, similar comparisons shall be undertaken for the classification of ovarian tumors (2). In addition, statistical methods based on cross-validation and resampling techniques are developed that allow an unbiased estimation of misclassification rates.

• Publications

(1)

Madjar, H., Prömpeler, H.J., Sauerbrei, W., Wolfarth, R., Pfleiderer, A.
Doppler flow criteria of breast lesions,
Ultrasound Med Biol. 20, 849-858, 1994.

(2)

Prömpeler, H.J., Madjar, H., Sauerbrei, W., Lattermann, U., du Bois, A., Pfleiderer, A.
Quantitative flow measurements for classification of ovarian tumors by transvaginal color Doppler sonography in postmenopausal patients,
Ultrasound Obstet. Gynecol. 4, 406-413, 1994.

(3)

Sauerbrei, W., Madjar, H., Prömpeler, H.J.
Differentiation of benign and malignant breast tumors by logistic regression and a classification tree using doppler flow signals,
Methods of Information in Medicine 37, 226-234, 1998.

(4)

Schumacher, M., Roßner, R., Vach, W.
Neural networks and logistic regression. Part I,
Computational Statistics and Data Analysis 21, 661-682 [FDM-Preprint Nr. 2], 1996.

(5)

Vach, W., Roßner, R., Schumacher, M.
Neural networks and logistic regression. Part II,
Computational Statistics and Data Analysis 21, 683-701 [FDM-Preprint Nr. 2], 1996.