|
Chemical Factor Analysis (CFA) algorithms were applied to transform complex Fourier transform infrared fiberoptical evanescent wave (FTIR-FEW) normal and malignant skin tissue spectra into factor spaces for analysis and classification. The factor space approach classified melanoma beyond prior pathological classifications related to specific biochemical alterations to health states in cluster diagrams allowing diagnosis with more biochemical specificity, resolving biochemical component spectra and employing health state eigenvector angular configurations as disease state sensors. This study demonstrated a wealth of new information from in vivo FTIR-FEW spectral tissue data, without extensive a priori information or clinically invasive procedures. In particular, we employed a variety of methods used in CFA to select the rank of spectroscopic data sets of normal benign and cancerous skin tissue. We used the Malinowski indicator function (IND), significance level and F-Tests to rank our data matrices. Normal skin tissue, melanoma and benign tumors were modeled by four, two and seven principal abstract factors, respectively. We also showed that the spectrum of the first eigenvalue was equivalent to the mean spectrum. The graphical depiction of angular disparities between the first abstract factors can be adopted as a new way to characterize and diagnose melanoma cancer.
|
