Page 26 - Vibrational Spectroscopic Imaging for Biomedical Applications

4 Cha pte r O n e

A major impediment to clinical translation, however, has been both
the lack of fast imaging methods and the lack of robustly validated
protocols that are ready for implementation. 18
The motivation for developing imaging methods is now clearly
accepted. In the past 20 years several research groups have investi-
gated the application of mid-IR spectroscopy for automated disease
diagnosis. 20,21 The first attempt involved simply measuring the FT-IR
22
spectrum of an extracted tissue sample. Noticing that histologic
composition provided a stronger variance than the benign-malignant
23
differences, practitioners moved quickly to employing microscopy
24
approaches. Since each pixel required a spectrum to be scanned,
these approaches were very slow. Consequently, studies typically
measured only a few spectra from a small number of samples. Thus,
the validity and robustness of these studies were not clearly estab-
lished due to the low statistical power of the studies. It is only dur-
ing the last ~5 years that microscopy approaches became routine in
rapidly providing high-quality data. Consequently, it is now increas-
ingly recognized that this technology has the potential to provide an
objective method for histopathology. The crucial question, however,
has remained one of accuracy while being robust. In this manuscript,
we discuss the key steps and large population feasibility in the
development of a practical algorithm for clinical translation. We
employ two-class models for breast histopathology to provide the
essential features of, first, breast histology and, second, breast
pathology.

1.1.1 FT-IR Imaging
Early efforts in providing spatially resolved FT-IR data involved a
25
point-mapping approach. Briefly, a target sample region was identi-
fied and radiation restricted to this region with an opaque aperture to
achieve the desired spatial localization of the beam. A single-element
detector measured the spectrum at each point and the entire sample
could be measured in a mapping sense by raster scanning. Though
useful for small numbers of samples at a low-spatial resolution, this
technique is prohibitively slow and produces noisy data for high-
26
resolution images of large samples. A typical field of view for a
sample of pathologic interest is about 0.5 × 0.5 mm. Point mapping of
this size of sample at diffraction limited spatial resolution (~5 μm at
the center wavelength) would require almost a day. Thus, even
though promising results were shown by many, a practical approach
to translating developments to the clinic was lacking.
The use of focal place array (FPA) detectors provided a multi-
channel detection advantage that allowed simultaneous measure-
ment of interferometric data from large fields of view and significantly
increased data acquisition rates. The first FPA detectors employed
27
step scanning interferometry, which have been almost exclusively