In this pilot study, a total of 23 samples were collected from five patients and analyzed. The results of histopathologic examination are shown in . Macroscopically, 14 samples were judged malignant and 9 benign. Histologic examination showed that samples from four patients (cases 2–5) had been correctly categorized on the basis of macroscopy as being malignant tissue or benign mucosa; samples deriving from suspected cancerous tissue from one patient (case 1) did not contain malignant cells.
Complete pathology report for 23 biopsy samples from 5 patients
Comparison of malignant vs. benign tissue samples showed significant differences for the three principle components, as listed in , together with their eigenvalues, percentages of variances represented, and cumulative percentages of variances. In addition to this binary differentiating capacity, linear regression analysis showed several principle components had predictive value, summarized in . For instance, PC 2, accounted for 16.1 percent of the variance among the samples, correlated negatively with volume percentage of malignant tissue (P=0.0065, slope=−0.0966, Pearson’s Correlation coefficient r=−0.550) and positively with volume percentage of normal epithelium (P=0.0051, slope=0.0367, and r=0.564)‥
Identification of malignancy according to Student’s t-tests
Results of linear regression analysis
The major loading factors for the regions that most significantly correlated with variations described by the principle components are shown in . For instance, PC 2 reflected changes in several metabolites with known connections to oncologic developments, including choline and phosphocholine.10
Metabolic activity, including choline metabolites in the spectral region from 3.20 to 3.24 ppm, which significantly contributed to the differentiation of the normal from cancerous profile established by PC 2, is shown for one pair of samples (tumor and benign from patient 3) in .
Loading factors and regions corresponding to principle components
Figure 2 A) HRMAS 1HMRS spectrum of Patient 3, sample ii (benign mucosa); entire metabolite spectrum (top) and enlargement of the 3.2–3.7 ppm area, which significantly contributed to the variation in PC 2; B) Sample 3-i contained 0 percent malignant cells, (more ...)
As indicated by , the most significant loading factors of each principle component, i.e., those with larger “+” or “-” values, present the positive or negative contributions of the concerned metabolites represented by their corresponding standardized concentration. For example, if the linear regression fit for a principle component with cancer is negative (slope<0), a negative loading factor for the choline metabolites in the 3.20- to 3.24-ppm region suggests that elevation of these metabolites above the calculated mean from all samples, and the reduction of metabolites in the 3.63–3.65 ppm region from their mean, contributes to the malignant profile, as seen with PC 2 in .