Probability Maps for Tumour Localisation, Using 3D 1H-MRSI and LCModel Fitting of Model "Benign" and "Tumour" Basis Spectra.

Alan Wright¹, Thiele Kobus¹, Arend Heerschap¹

Probability maps were generated that determine the likelihood of tumour at each spatial location within the prostate based on 3D ¹H-MRSI data sets. Each spectrum was fitted with LCModel using a basis set of two model spectra, one for benign and one for tumour tissue. These basis spectra were generated by independent component analysis as the first-two stable components of a data set of 2360 spectra from 42 patients. The probability score separated tumour from benign spectra better than conventional metabolite maps (AUC: 0.782, 0.721 respectively) in this data set and successfully localised significant tumour foci in five new test patients.