Although readily acquired in MRI and MRS exams, fat information is used sparingly-if at all- for diagnostic purposes. In fact, many publications exist, aiming to improve fat suppression and produce fat-free images^1-2 or MRS spectra^3. Accumulating evidence, however, points to the dynamic nature of lipid profiles, which can change during the menstrual cycle⁴, carcinogenesis⁵ and gene therapy⁶. Disruptions in the expression of genes related to fat metabolism were also found in the contralateral, unaffected breast of cancer patients⁷. Growing numbers of publications (Figure 1) highlight our evolving understanding of the correlation between carcinogenesis and lipid profile changes. The local or global nature of these changes is yet to be determined; their existence, however, hints to the fact that non-invasive lipid profile measurements may offer better tumor detection/characterization. It is the goal of our study to understand measurement capability and spatial variability of single voxel MRS in normal volunteers and breast cancer patients undergoing neo-adjuvant chemotherapy (NAC).

Single voxel MRS data were acquired on 3T, MR750 scanners, using 8-channel breast coils. A STEAM sequence (TE/TR=14/1500ms, 64 averages) was used to acquire data from 1-3 single voxels per subject. Small voxels (1 to 1.9cc), chosen to allow for coverage of small tumors, were placed over glandular tissue. Since cancer cells produce their own supply of fatty acids and simultaneously co-opt fat reserves from the local micro-environment for energy and cell division needs⁸, we hypothesize that changes in lipid metabolism may be detectable in and around the tumor areas. Data were acquired in six normal volunteers and two patients undergoing NAC. Apart from subject 3, for whom data from a single voxel was acquired, two voxels/volunteer were queried. Each voxel was scanned four times, to determine measurement repeatability. In the cancer patients, the three voxels were placed over over the tumor/adjacent to the tumor/in the contralateral breast, respectively. As before, efforts were made to avoid adipose tissue. Figure 2 exemplifies voxel placement in the study subjects.

Data was quantified using LCModel and analyzed statistically in Minitab. Fat peaks at 1.3+1.6ppm (L13+L16), 2.1+2.3ppm (L21+L23), 2.8ppm (L28), 4.1+4.3ppm (L41+L43) and 5.2+5.3ppm (L52+L53) were expressed as ratios to the CH3 peak (L09); this peak was found to be stable throughout the menstrual cycle⁴ and carcinogenesis⁵. The polyunsaturated fatty acids (PUFA), known to be associated with tumorigenesis, were also computed [as (L28/L23)⁹].

Figure 3 displays an exemplary spectrum/fit from a normal volunteer. Figure 4 presents a summary of our results; here peak ratios are displayed for all the subjects, voxels and repeat measurements. Subjects 1-6 are normal volunteers; subjects 7-8 are cancer patients. For normal volunteers, the filled triangles/open circles represent repeat scans of the first/second voxel, respectively. For the cancer patients, the red/blue/green filled circles represent data from the tumor/adjacent to the tumor/contralateral breast tissue, respectively. Note a few very interesting features of these graphs:

- While measurement variability is non-negligible (and was likely exacerbated by our choice of small voxels and intentional avoidance of adipose tissue), there is evidence that between voxel differences can exceed measurement variability (note, e.g. the frequent separation between the triangle/circle voxel clusters in all the lipid ratios). This suggests that lipid metabolism may be spatially varying even in normal volunteers.

- ANOVA indicates that volunteer is a significant factor (p<0.05) in determining all lipid ratios, except for (L21+L23)/L09.

- Most importantly, some of the lipid ratios appear to provide very good separation between tumor and normal tissue. For example, the (L13+L16)/L09 ratio is significantly lower than normal in tumors, and higher than normal in tissues adjacent to the tumor (p=0.04). Contralateral breast exhibits values comparable to normal tissues (Figure 5).

In our data choline was only present in a single tumor voxel. Although few similar studies exist in the literature, our tumor (L13+L16)/L09 ratio is comparable to one previously reported in tumors⁵. While our study shows decreased such ratios in tumors compared to normal tissue, the study of⁵ demonstrates increased ratios in cancerous lesions compared to benign lesions. The broader picture can only understood by acquiring spatially dependent data in a wider pool of normal volunteers/cancer patients, a study which is currently ongoing.

This work presents preliminary evidence, indicating that lipid metabolism may be spatially varying in normal volunteers and cancer patients. Moreover, MRS-based measurements of lipid peak ratios may provide useful markers for cancer detection and characterization. Further expansion of the current study and translation from a spectroscopy to an imaging approach will provide further insight into the ultimate usefulness of lipid profiling, as measured by MRI/MRS, as non-invasive cancer markers.