Breast cancer and non-malignant breast data: Quantification by FPT - Signal Processing in MRS with Biomedical Applications - page 307

mean choline SNR among the patients with breast cancer was 6.2±2.1 versus
2.4±0.7 for those with benign breast lesions. The authors from Ref. [388]
defined the distinguishing cutpoint for choline SNR to be 4.1, with the highest
benign value being 4, and lowest malignant value being 4.1. They remarked
that unequivocal detection of the choline resonance was sometimes di cult,
due to overlap with sidebands of residual water or lipid resonances. Improved
data processing methods were cited as a necessary precondition for reliable
quantification of choline in breast lesions [388].
Bolan et al. [389] employed a high static magnetic field (4T) and optimized
surface coils with single voxel in vivo MRS, and attempted to quantify total
choline by an algorithm which fitted the peaks one at a time over a narrow
frequency band. The average of spectra at various TE values was employed
to try to separate noncoupled metabolite resonances from the lipidinduced
sidebands [390]. This method was applied to 500 spectra from patients with
breast cancer, with benign breast pathology as well as with normal breast
tissue. Total choline was significantly higher in the malignant compared to
benign lesions, but there was also overlap in the ranges. Importantly, total
choline was undetectable in several breast cancers, and high in some benign le
sions, as well as in a few normal breasts. The total choline amplitude appeared
to be influenced by the lipid content of the voxel, either through baseline arte
facts not suppressed by TE averaging or by a true resonance at 3.25 ppm. In
smaller voxels, the fitting error was also reportedly higher [389].
In the recent investigation of 124 patients by Sardanelli et al. [380] em
ploying a 1.5T MR scanner, the cutpoint for the integral under the total
choline peak was 1.9 arbitrary units to classify a breast lesion as malignant.
Using this threshold, sensitivity of 90% and specificity of 89% were obtained,
with 3 patients excluded due to poor SNR. The authors [380] note that the
diagnostic accuracy was better for the larger lesions (> 1 cm).
Other authors have also noted that, as currently applied, in vivo MRS is
limited regarding the diagnosis of smaller tumors [101, 378, 381], [391]-[393].
We refer once again to the observation of surgical oncologist Gluch [345]:
“di culties arise not so much in regards to the large lesion with suspicious
imaging characteristics, but rather the lesion < 1 cm in size, or ductal carci
noma in situ. In vivo studies have not satisfactorily addressed these entities.
A chance of missing a breast cancer of the order of only 1% would translate
into a significant medicolegal concern”(p. 467). Moreover, as pointed out by
Bartella and Huang [378], it is usually necessary to evaluate several lesions on
an MR image, so that in fact, this becomes a multiply important limitation.
10.2.3 In vivo MRS to assess response of breast cancer to
therapy
In the earlierdescribed study by Jagannathan et al. [386] some (but not all)
of the patients with breast cancer followed by in vivo MRS showed a drop
in the unsuppressed water to lipid ratio after chemotherapy; this response