Validation of fuzzy logic method for automated mass spectral classification for mineral imaging

Imaging mass spectrometry requires the acquisition and interpretation of hundreds to thousands of individual spectra in order to map the mineral phases within heterogeneous geomatrices. A fuzzy logic inference engine (FLIE) was developed to automate data interpretation. To evaluate the strengths and limitations of FLIE, the chemical images obtained using FLIE were compared with those developed using two chemometric methods: principle component analysis (PCA) and cluster analysis (K-Means). Two heterogeneous geomatrices, a low-grade chalcopyrite ore and basalt, were imaged using a laser-desorption Fourier transform mass spectrometer. Similar mineral distribution patterns in the chalcopyrite ore sample were obtained by the three data analysis methods with most of the differences occurring at the interfaces between mineral phases. PCA missed one minor mineral phase in the chalcopyrite ore sample and did not clearly differentiate among the mineral classes of the basalt. K-Means cluster analysis differentiated among the various mineral phases in both samples, but improperly grouped some spectra in the chalcopyrite sample that only contained unanticipated high mass peaks. Unlike the chemometric methods, FLIE was able to classify spectra as unknowns for those spectra that fell below the confidence level threshold. A nearest neighbor approach, included in FLIE, was used to classify the unknowns to form a visually complete image; however, the unknowns identified by FLIE can be informative because they highlight potential problems or overlooked results. In conclusion, this study validated the fuzzy logic-based approach used in our laboratory and reveald some limitations in the three techniques that were evaluated.