Prediction of malignant breast lesions from MRI features: a comparison of artificial neural network and logistic
Christine E McLaren1, Wen-Pin Chen, Ke Nie
1Department of Epidemiology, University of California, Irvine, 224 Irvine Hall, Irvine, CA 92697-7550, USA. cmclaren@uci.edu
Rationale And Objectives:
Dynamic contrast-enhanced magnetic resonance imaging is a clinical imaging modality for the detection and diagnosis of breast lesions. Analytic methods were compared for diagnostic feature selection and the performance of lesion classification to differentiate between malignant and benign lesions in patients.
Materials And Methods:
The study included 43 malignant and 28 benign histologically proved lesions. Eight morphologic parameters, 10 gray-level co-occurrence matrix texture features, and 14 Laws texture features were obtained using automated lesion segmentation and quantitative feature extraction. Artificial neural network (ANN) and logistic regression analysis were compared for the selection of the best predictors of malignant lesions among the normalized features.
Results:
Using ANN, the final four selected features were compactness, energy, homogeneity, and Law_LS, with an area under the receiver-operating characteristic curve (AUC) of 0.82 and accuracy of 0.76. The diagnostic performance of these four features computed on the basis of logistic regression yielded an AUC of 0.80 (95% confidence interval [CI], 0.688-0.905), similar to that of ANN. The analysis also showed that the odds of a malignant lesion decreased by 48% (95% CI, 25%-92%) for every increase of 1 standard deviation in the Law_LS feature, adjusted for differences in compactness, energy, and homogeneity. Using logistic regression with z-score transformation, a model composed of compactness, normalized radial length entropy, and gray-level sum average was selected, and it had the highest overall accuracy, 0.75, among all models, with an AUC of 0.77 (95% CI, 0.660-0.880). When logistic modeling of transformations using the Box-Cox method was performed, the most parsimonious model with predictors compactness and Law_LS had an AUC of 0.79 (95% CI, 0.672-0.898).
Conclusion:
The diagnostic performance of models selected by ANN and logistic regression was similar. The analytic methods were found to be roughly equivalent in terms of predictive ability when a small number of variables were chosen. The robust ANN methodology uses a sophisticated nonlinear model, while logistic regression analysis provides insightful information to enhance the interpretation of the model features.