Magnetic Resonance Spectroscopy (MRS) enables noninvasive measurements of the metabolic concentrations in breast in vivo, which makes it an ideal diagnostic tool for breast cancer detection. Two-dimensional (2D) MRS has shown the potential to detect more metabolites of low concentrations and to resolve overlapping resonances that limit the accuracy of its one-dimensional (1D) counterpart. The proposed thesis work will focus on the applications of 2D MRS for breast cancer detection in vivo.

There are four pilot projects in this thesis. The first project was to combine DCE MRI with 2D L-COSY to increase breast cancer detection specificity. Using the most prevalent five diagonal and six cross peak metabolite ratios as input, we used two statistical tools (CART and Fisher's linear discriminate analysis) to determine a better statistical tools (CART and Fisher's linear discriminate analysis) to determine a better classification scheme to distinguish between malignant and benign tumors. In the second project, CART analysis was performed to classify four different breast tissues using the 2D COSY spectra. 18 metabolite peak ratios were selected as the potential markers for the statistical analysis. CART can reach an overall correct rate of 82.0% in distinguishing malignant tumor, benign tumor, healthy fatty and glandular tissues. In the third project, 2D MR spectra of breast acquired at 1.5T versus 3T were compared in terms of metabolite SNRs, diagonal and cross peak ratios and 2D spectral resolutions. Results showed that 2D MRS of breast tissues have better SNR and chemical shift dispersion at 3T than at 1.5T.

Single voxel 2D MRS has significant limitations in terms of breast coverage. Multi-dimensional magnetic spectroscopic imaging (MRSI) can solve the problem by generating high resolution metabolite images. In the fourth project, a four-echo based spatially resolved MRSI sequence was tested in breast tissue on a Siemens 3T scanner. From the four dimensional MRSI data, six metabolites (FAT, Water, UFD, UFL, UFR, TGFR) distribution images were generated. The images between healthy and benign subject are detectably different. The metabolite distributions may serve as the biomarkers for early diagnosis of breast disease.

In summary, all the proposed methods have the potential to be used clinically for noninvasive breast cancer detection in the future.