Classification of in vivo H-1 MR spectra from breast tissue using artificial neural networks

Background: The study was designed in or der to investigate whether ar arti ficial neural networks could be used for analysis of in vivo magnetic reson ance (MR) spectra from breast cancer patients. Materials and Methods: In vi vo H MR spectra with three different echo times (TE 135, 350 and 450 msec) were acquired from patients with benign and malignant breast lesions and fr om healthy volunteers, of whom some were breast-feeding. A spectral region (4.0 - 2.5 ppm) was used as input far artificial neural network analysis, f or the attempted classification of the data into different groups. Results: Data recorded at all three echo times were necessary to obtain the best re sults. Furthermore, malignant tissue was differentiated from benign tumours using; this approach whereas benign tumours were poorly separated from hea lthy tissue. Conclusion: The results presented here indicate that in vivo M R spectroscopy in conjunction with neural network analysis might be useful for the evaluation of breast lesions.