Optical Imaging in Biology: Basics and Applications
655
of an illumination or excitation source, dispersion element, camera, and a moving
stage. The dispersive device is the key component of the set-up. Filters, grating,
and prisms are often employed to disperse the light into different wavelengths and
project it on the detector. The hyperspectral images are often obtained in the form of a
three-dimensional (2 spatial and 1 spectral) cube. In an HSI system, the spatial information can be collected in three different ways namely whisk broom, push broom,
and tunable filter. These are also commonly referred to as point scanning, line scanning, and area scanning, respectively [75]. The point scanning involves gathering
spatial information in two dimensions through the movement of the sample or the
detector. Even though a hyperspectral image can be obtained by moving the sample
systematically in two-dimensions this method is highly time-consuming. In linescan, a slit of spatial information and full spectral information for each spatial point
in the linear field of view can be acquired simultaneously. This method involves an
imaging spectrograph and is relatively faster with respect to point-scan. The third
approach of area scanning does not involve any movement as the images are obtained
only from fixed specimens [76]. Here the light is reflected from the entire surface
and captured on the camera after passing through different optical filters and components. The usefulness of each of these configurations should be considered before
starting the acquisition. For a detailed analysis often, the point scans are preferred
while line or area scans can be employed for real-time analysis.
Depending upon the orientation of the excitation source and the detector (camera,
spectrograph, or lenses) a typical HSI system can be used for three different applications. In reflectance and transmittance configurations, it can be used to collect
information about the surface and internal composition, respectively. In the third
type ‘interactance’ mode, the detector and the light sources are parallel to each other.
These are often used to record information about the changes due to the interaction
of the light with the specimens [75]. After the acquisition of spectra and images, the
data is subjected to spectral data processing and they are visualized through various
methods. The computation processing on 3D image cubes provides information on
the spectral signatures contained, and this technique has been pioneered by David
Landgrebe [77]. These methods are often centered on removing the influence of
dead pixels, spiked wavelengths, or removing unwanted areas (background) from
the image. Some of the well-known methods for pre-processing of the hyperspectral image cube includes de-noising, normalization, smoothening, and derivatives
[78]. However, some of the preprocessing tools such as binning are known to reduce
spectral resolution therefore this should be performed with care considering the
data variability and the target image [76]. The spectral data can be visualized in
three different ways– image space (displays data samples spatially and provides a
picture), spectral space (spectrum corresponding to each pixel) and feature space
(hyperspace) (Fig. 9), where spectral and feature space is used for computational
processing and these methods mostly map between the image, spectral and feature
spaces. Visualization of the image space is one of the most important aspects of
biomedical imaging. There are different strategies for visualization including classification imaging where the pixels are grouped according to their spectral signature. Other methods include quantitative imaging where the image is reconstructed
655
of an illumination or excitation source, dispersion element, camera, and a moving
stage. The dispersive device is the key component of the set-up. Filters, grating,
and prisms are often employed to disperse the light into different wavelengths and
project it on the detector. The hyperspectral images are often obtained in the form of a
three-dimensional (2 spatial and 1 spectral) cube. In an HSI system, the spatial information can be collected in three different ways namely whisk broom, push broom,
and tunable filter. These are also commonly referred to as point scanning, line scanning, and area scanning, respectively [75]. The point scanning involves gathering
spatial information in two dimensions through the movement of the sample or the
detector. Even though a hyperspectral image can be obtained by moving the sample
systematically in two-dimensions this method is highly time-consuming. In linescan, a slit of spatial information and full spectral information for each spatial point
in the linear field of view can be acquired simultaneously. This method involves an
imaging spectrograph and is relatively faster with respect to point-scan. The third
approach of area scanning does not involve any movement as the images are obtained
only from fixed specimens [76]. Here the light is reflected from the entire surface
and captured on the camera after passing through different optical filters and components. The usefulness of each of these configurations should be considered before
starting the acquisition. For a detailed analysis often, the point scans are preferred
while line or area scans can be employed for real-time analysis.
Depending upon the orientation of the excitation source and the detector (camera,
spectrograph, or lenses) a typical HSI system can be used for three different applications. In reflectance and transmittance configurations, it can be used to collect
information about the surface and internal composition, respectively. In the third
type ‘interactance’ mode, the detector and the light sources are parallel to each other.
These are often used to record information about the changes due to the interaction
of the light with the specimens [75]. After the acquisition of spectra and images, the
data is subjected to spectral data processing and they are visualized through various
methods. The computation processing on 3D image cubes provides information on
the spectral signatures contained, and this technique has been pioneered by David
Landgrebe [77]. These methods are often centered on removing the influence of
dead pixels, spiked wavelengths, or removing unwanted areas (background) from
the image. Some of the well-known methods for pre-processing of the hyperspectral image cube includes de-noising, normalization, smoothening, and derivatives
[78]. However, some of the preprocessing tools such as binning are known to reduce
spectral resolution therefore this should be performed with care considering the
data variability and the target image [76]. The spectral data can be visualized in
three different ways– image space (displays data samples spatially and provides a
picture), spectral space (spectrum corresponding to each pixel) and feature space
(hyperspace) (Fig. 9), where spectral and feature space is used for computational
processing and these methods mostly map between the image, spectral and feature
spaces. Visualization of the image space is one of the most important aspects of
biomedical imaging. There are different strategies for visualization including classification imaging where the pixels are grouped according to their spectral signature. Other methods include quantitative imaging where the image is reconstructed