Table 4.

Neural network (trained on k-means) prediction of tumor likelihood for five spectra from regions are shown to be rich in tumor tissue by histopathology, but predicted to be “healthy” by k-means.

Neural network prediction of tumor likelihood
NN	s1 (%)	s2 (%)	s3 (%)	s4 (%)	s5 (%)
FPHW	18.5	4.6	0.1	23.8	61.2
FP	4.2	8.6	10.1	11.1	13.4
HW	11.3	13.1	20.4	8.5	82.0
Average NN	11.4	8.7	10.2	14.5	52.2

Likelihoods of tumor > 20% appear in bold. Figure 4 shows these 5 spectra along with average spectra from regions classified as 0% and 100% tumor.