Cancerous tissue exhibits altered metabolite concentrations as compared to normal brain tissue. Magnetic resonance spectroscopy imaging (MRSI) reveals such abnormalities in altered spectral profiles. Although the relations between spectral profiles and histological findings are well established, the significant variability of in vivo spectra, which is due to the heterogeneity of tumor tissues, large voxel sizes, and the mixture of normal brain tissues with infiltrative tumors (partial volume effect), often limits their diagnostic potential. This variability complicates tumor diagnosis and grading, as well as the determination of tumor spatial extend. Different spectral analysis methods are being developed to address this problem.
Previously we proposed an algorithm called non-negative matrix factorization (NMF) that extracts constituent spectra associated with different tissue types by simultaneously analyzing all voxel spectra. In principle this method solves the partial volume effect as it determines also the proportion with which each constituent spectrum contributes to an individual voxel spectrum. The algorithm was shown to extract spectral profiles and their spatial distributions consistent with normal and cancerous tissue.
The goal of our research is to demonstrate the physiological relevance of this decomposition for routine clinical brain tumor scans. To do this we analyzed the extracted spectra and showed an improved correlation of choline (Cho) and N-acetyl aspartate (NAA) peak areas with tumor grade compared to conventional method. To validate the physical interpretation of abundances as volume fraction we compared in a phantom study the extracted abundance values with the expected values following the geometry of the phantom and the result showed a good match.
These results indicate that MRSI in combination with the proposed spectrum separation method can improve MRSI in the diagnosis of brain tumors, especially in defining tumor margins for treatment planning of radiation therapy or surgical resection.
1. Su Y, Thakur SB, Karimi S, Du S, Sajda P, Huang W, Parra LC. Spectrum Separation Resolves Partial Volume Effect of MRSI as Demonstrated on Brain Tumor Scans. NMR in Biomedicine, 2008, 21 (10): 1030-1042.PDF
2. Sajda P, Du S, Brown TR, Stoyanova R, Shungu DC, Mao X, Parra LC. Nonnegative matrix factorization for rapid recovery of constituent spectra in magnetic resonance chemical shift imaging of the brain. IEEE Trans Med Imaging, 2004,23(12):1453-1465. PDF
3. Lee JS, Lee DD, Choi S, Lee DS. Application of nonnegative matrix factorization to dynamic positron emission tomography. Third International Conference on Independent Component Analysis and Blind Signal Separation. San Diego, CA; 2001. p 556–562. PDF
4. Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature 1999;401(6755):788-791. PDF