Detection of choline signal in human breast lesions with chemical-shift imaging

Abstract

Purpose: To investigate the application of MR spectroscopy using chemical-shift imaging (CSI) for characterizing human breast lesions at 1.5T, and to evaluate the diagnostic performance using ROC (receiver operating characteristics) analysis. Materials and Methods: Thirty-six patients (35-73 years old, mean 52), with 27 malignant and 9 benign lesions, underwent anatomical imaging, dynamic contrast-enhanced MR imaging, and CSI. The ROC analysis was performed and the cutoff point yielding the highest accuracy was found to be a choline (Cho) signal-to-noise ratio (SNR) >3.2. Results: The mean Cho SNR was 2.8 ± 0.8 (range, 1.8-4.3) for the benign group and 5.9 ± 3.4 (2.1-17.5) for the malignant group (P = 0.01). Based on the criterion of Cho SNR >3.2 as malignant, CSI correctly diagnosed 22 of 27 malignant lesions and 7 of 9 benign lesions, resulting in a sensitivity of 81%, specificity of 78%, and overall accuracy of 81%. If the criterion was set higher at Cho SNR >4.0 the specificity improved to 89% but sensitivity was lowered to 67%. Conclusion: The ROC analysis presented in this work could be used to set an objective diagnostic criterion depending on preferred emphasis on sensitivity or specificity. © 2008 Wiley-Liss, Inc.