Page 32 - Vibrational Spectroscopic Imaging for Biomedical Applications

0.5 mm Stroma Epithelium 0.7 0.6 0.5 1556 cm –1 : 1652 cm –1
Select Tissue Classes 0.12 0.03 0.06 0.09 Relative Frequency 0.00 0.4 1.6 1.2

(c) 0.8 0.4 0.0 1080 cm –1 : 1238 cm –1 Determine Metric Probability Distributions

Stroma Epithelium 4000 3000 3500 0.24 (d) 0.18 0.12 0.06 Relative Frequency 0.00

1000 1500 2000 2500 Wavenumber (cm –1 ) Select Metrics From Tissue Spectral Features Epithelium Stroma

(b) Absorbance (Offset for Clarity) (a) FT-IR breast TMA image data is acquired and (b) the resulting tissue spectra are analyzed to select spectral metrics. (c) FT-IR stroma and epithelium are determined and (e) each pixel on the spectral image of the breast TMA is classified a
ΔAUC .645 .564 … Sort Metrics by ΔAUC and Repeat Classification Classify IR Image Data and H&E-stained images are then compared to select stroma and epithelium tissue regions. (d) The metric value frequency distributions for

Metric 1 2 … (e)
(h)

Acquire IR images 1.00 0.99 0.98 0.97 Stroma 0.96 Epithelium 0.95 Mean 10 8 6 4 2 80 60 40 20 Number of Metrics Perform Statistical Analysis (f) Compare with Gold Standard converges at AUC ~ 1 at which (i) an optimal set of classification metrics is selected.

(g) 1.00 0.99 0.98 AUC 0.97 0.96 0.95
0.3 0.0 Obtain Final Classification
(a)
(i)

FIGURE 1.2

9