Other polymers are used in order to generate differing reaction environments; polystyrene giving a hydrophobic environment whereas a polyamide would generate a hydrophilic environment. Tentagel, which is 80% polyethylene glycol grafted to polystyrene has become increasingly popular since it is believed to generate a more "ether-like" environment which lends itself to a greater range of chemical reactions and hence a greater range of possible structures.

The Linker Groups are the chemical groups that link the substrate to the polymer resin. Since it is this group that that must be broken at the end of the chemical sequence, their structure determines the method of cleavage from the resin. It is important that this cleavage reaction does not destroy the new products that the chemist has synthesised.

The real importance, however, of these Linker groups is that they determine the terminal functionality of the product ; thus, by using the same Linker group for an entire chemical library all the resultant products will have the same end group functionality.

Examples of these Linker groups are Wang Resin Linkers (Figure 3) which can be used in the synthesis of carboxylic acids.

Figure 3.Wang resin linkers

Other Linker groups include Rink Amide Linker (Figure 4) for primary amide synthesis and REM (Figure 5), HMB and Oxime Resins (Figure 6) which are for tertiary amine, C-terminally amidated peptide, and alkylamide synthesis respectively.

Figure 4. Rink Amide Linker

Figure 5. REM

 Figure 6. HMB and Oxime resins

There have been several different techniques used in COMBINATORIAL SYNTHESIS to generate large libraries of new structures quickly and efficiently. One such method is described as "Mix and Split" (Figure 7) and was developed by Furka. This is where, for example, three initial materials X,Y, and Z are bound to a resin and mixed together. The resultant mixture is split into three separate reaction vessels and each is coupled with a further set on reagents. Dimer samples from each of the three reaction vessels are saved for subsequent reactions. The remainder of the dimers are mixed, split and put into three separate reaction vessels again. These dimers are reacted again with a further set of reagents resulting in twenty-seven different structures being formed, nine in each of the three vessels.

Figure 7.

At this stage each of the three reaction vessels are tested for activity (Figure 8). On the basis that one of the vessels shows activity this limits the active structure to be one of nine.[This description is simplistic because the yield is likely to be less than 100% and impurities and bi-products may be generated tending to confuse the process involved.]

In the example given it is clear that the active structure must have Y as its terminal group. At this stage the retained dimer structures are reacted only with reagent Y to give nine final products, three in each of the reaction vessels. A further activity test reveals which of these vessels contains the active structure and each of the three compounds in this vessel can be synthesised independently 

Figure 8.

Product analysis can be carried out by cleaving the product from the resin and carrying out a normal classical chemical analysis by some form of spectroscopy. The chemist, however, would like some means of determining the success of his reactions whilst the substrate is still attached to the resin. This would allow him to proceed with the next stage in the sequence without the inconvenience of the cleavage step. For single products this is normally done using IR or NMR whilst for multi-component mixtures HPLC-MS is becoming the method of choice. The remainder of this note will consider the advantages and disadvantages of FT-IR (and Raman) spectroscopy as an analytical technique and also compare the various IR sampling techniques.

It is worth noting at this point that in most cases the synthetic chemist wants to follow a reaction sequence rather than obtain a definitive chemical structure. This means that the monitoring by IR of the appearance or disappearance of specific functional groups is the required objective. This is fortuitous since the resultant IR spectrum from the resin-substrate combination, although dominated by the contribution of the resin, does indicate the presence of the required substrate functionality. It is also worth noting that it is preferable for the analysis to be carried out in the same laboratory as the synthesis and by a non-specialist and therefore the analysis should be as straight forward as possible. Once again it is fortuitous that most synthetic chemists are used to following reaction sequences by following the progress of various functional groups by IR spectroscopy.

The sampling techniques studied were:

1. KBr discs
2. Diffuse reflectance on aluminised reflective pads.
3. Single-reflection (diamond) ATR
4. Microscopy …..both normal reflectance and micro-ATR
5. Raman spectroscopy
6. Transmission through flattened bead

Although photo-acoustic spectroscopy has been used by other investigators, it has not been considered here since the technique requires some degree of expertise to be fully successful.

Figure 9 shows the spectra obtained from a Wang resin bead with substrate attached. The three spectra were obtained using Diffuse reflectance (Spectrum 1), Single Reflection Diamond ATR (Spectrum 2), and Reflectance Microscopy (Spectrum 3).

Although the spectra are dominated by contributions from the polystyrene Wang resin, the important functional group in this synthesis was the carbonyl group whose absorption band is clearly shown at about 1770cm in all three spectra. The same bead sample was analysed by micro-ATR on the IR microscope and Figure 10 compares the normal reflectance spectrum (Spectrum 6) with that using the ATR objective (Spectrum 7).

Figure 10. Spectra from Sample 1 run using IR Microscope

A similar set of spectra was run for a second substrate on a Wang Resin. The diffuse reflectance (Spectrum 4), single reflection diamond ATR (Spectrum 5), and reflectance microscopy (Spectrum 6) spectra are shown in Figure 11.

Similarly, the beads were run using micro-ATR (Spectrum9) and normal reflectance (Spectrum 10) on the IR microscope (Figure 12).

In this particular example the detection of the hydroxyl group (absorption band at ~3300cm) is paramount in the structure identification. It is clear from the above data that this it is extremely difficult to detect this band using single reflection ATR methods either on the Diamond ATR accessory or on the microscope. This may be because these techniques sample such a small area of the sample and hence small amount of sample or it could be due to the wavelength dependence of such ATR techniques; absorption bands at higher wavenumbers do appear less intense than lower wavenumber absorptions. In an attempt to eliminate this later effect an ATR correction was carried out on one of the above ATR spectra and the results are shown in Figure 13.

The top spectrum shows the original data and the bottom shows the result of the ATR correction. Although the hydroxyl absorption band is slightly more distinct it is still a weak feature and could be overlooked. Also the fact that such post-run manipulation is required would tend to compromise the ease-of-use criteria.

Although the KBr disc method of preparing samples can be considered as difficult and time-consuming, it can produce completely satisfactory spectral results. Figure 14 shows the KBr spectra for both the Wang beads analysed above and show all the relevant features required by the chemist.

Since it is clear that transmission mode spectra do show the required features without being totally masked by the contribution of the polymer resin, an alternative to KBr disc making was sought. One such option is to use a Diamond Anvil Cell to crush a few beads between the diamond windows and run this directly on the IR microscope (Figure 15).

The quality of this data is very good, the top spectrum showing clearly a strong carbonyl absorption band and the lower showing the hydroxyl grouping.

To determine if similar data could be obtained from other types of beads, the same techniques were used to obtain spectra from a substrate on a Tentagel bead. Figure 16 shows the micro-ATR (top), single reflection diamond ATR (middle), and diffuse reflectance spectra from such a material.

Once again in this example the detection of the hydroxyl group at ~3300cm is important. As with the Wang resin, the diffuse reflectance spectrum shows this absorption clearly whereas in the ATR data this feature is barely detectable.

These Crowns come in a variety of shapes and sizes with the most common shape being shown in Figure 17. They are normally about 3 to 4 cm in length. The initial starting material is bound to the surface of the polymer Crown. In use they are suspended from a tray into a reaction vessel into which differing reacting substrates are found in solution. Between each step in the synthesis the tray is raised from the reaction medium and the Crowns washed with an appropriate solvent.

The IR analysis of these Crowns does present additional difficulties over that for the beads. The shape and their structure rules out performing straight transmission measurements on them and since the surface of one Crown may contain the total yield of the synthesised material the chemist is particularly concerned not to destroy or damage the surface; he will require the Crown for the next stage in the reaction. These criteria tend to rule out the use of Diamond ATR accessories which would require that the Crown be compressed or that a part of the Crown be cut off. The three techniques used for the analysis were Diffuse Reflectance (using the abrasive rods supplied with the PE accessory), Reflectance Microscopy and Micro-ATR on the microscope.

The results for a substrate on such a Crown are shown in Figure 18. Spectrum A shows the diffuse result, Spectrum B shows the reflectance result and Spectrum C the micro-ATR measurement. In this example the product contains a carboxylic acid group and whereas both the diffuse and reflectance spectra show the carbonyl and hydroxyl grouping, the ATR measurement shows only very weak features for both.

Raman Spectroscopy

One of the main advantages of Raman Spectroscopy is its ease of sample preparation. This would be an obvious benefit to the combinatorial chemist wanting data quickly from the beads or crown in a non-destructive manner such that he can easily move on to the next stage in the synthesis. The results shown in Figure 19 are the Raman spectra from two different crowns. The top spectrum is from the crown with the initial starting material on the surface whereas the lower spectrum shows the result after two reaction steps; the result of these reactions is that the new material contains a carbonyl group. Although the spectra are of very high quality and are very easy to obtain, they are totally dominated by the contribution from the polymer crown itself and show no bands to the reactions. Additional Raman data was obtained from other samples with the same result, namely that the intense Raman spectra from the polymer support is much stronger than that of the substrate of interest.

The relative usefulness of Raman compared to IR can be clearly illustrated by taking the above sample containing the carbonyl grouping and comparing it’s Raman with it’s IR spectrum (Figure 20). The top spectrum is the Raman data whereas the lower spectrum shows the IR data in absorbance units for ease of comparison. Although it could be argued that the Raman spectrum is a better spectrum, it is the IR one that shows the required carbonyl absorption.

[It should be noted that no attempt was made to find a system 'ideal' for Raman study. The aim was to compare the usefulness of several analytical techniques applied to the same type of sample].

The relative ease with which FT-IR spectroscopy can determine the success or failure of a combinatorial synthesis appeals to the synthetic chemist. The fact that IR is a do-it-yourself technique means that the results are available immediately to the chemist. It would appear that the best techniques of analysis are diffuse reflectance using the aluminised pads, KBr discs, transmission through a crushed bead, or reflectance microscopy. For crowns, diffuse reflectance or reflectance microscopy would appear to be the best options.

Although ATR sampling techniques result in very good quality spectra, there is a real danger that the relevant absorption bands are too weak to detect easily. Raman spectra seem to suffer in the same way; the absorption bands of interest are very weak compared to the very intense bands due to the supporting polymer.

Norman A. Wright 

Bio-Rad,
Spectroscopy Division,
237 Putnam Avenue,
Cambridge, MA 02139
U.S.A.

Email: Norman_Wright@bio-rad.com

Introduction

This paper provides an overview of an infrared spectroscopic imaging system along with some applications. The technique combines the emerging commercially available technology of focalplane array (FPA) detectors with a step-scan Fourier transform (FT) Michelson interferometer based spectrometer. This coupling provides an optimized method for infrared spectroscopic imaging by simultaneously realizing both a multiplex and multichannel advantage. Specifically, the multiple detector elements enable spectra from all pixels to be collected simultaneously, while the interferometer allows all the spectral frequencies to be measured concurrently. This technique can rapidly generate very high quality, chemically specific images from a wide variety of samples. Early systems were built for special purpose applications. Among these were airborne gas analysis [1,2] and astronomical observation [3]. The instrumentation described here is specifically oriented to chemical analysis.

Infrared microscopy is a tool that lends itself to the examination of heterogeneous samples. Using microscopy a researcher can more easily investigate a sample for defect analysis, domain identification or boundary information. Mapping techniques have been developed for infrared microscopes to expand the capabilities of and ease of use when studying this spatial nature of samples. However, creating a map becomes a time consuming task if both high spatial resolution and high signal to noise is required. Even with the current level of sophistication available in infrared mapping systems, this fundamental limitation cannot be overcome. A mapping system cannot generate, in a reasonable time frame, sufficient data to reconstruct an image with high spatial content as well as good signal-to-noise.

Infrared imaging is the logical replacement of infrared mapping when the need to spatially characterize a sample is required. With array detectors covering suitable spectral regions, this capability is now available. Samples can be viewed either in transmission or reflection and no staining or coating is required. The technique is non-destructive and works well for solids and liquids. The contrast that is seen in these images arises from the inherent infrared absorption of the sample, i.e.; every chemical compound has its own characteristic absorption pattern that creates an infrared signature.

This paper describes the instrumental components of the Bio-Rad FTS Stingray 6000 infrared imaging system configured for microanalysis (shown in Figure1), and a sampling of applications of this emerging powerful analytical tool.

Experimental Set Up

A review of focalplane array detectors and their potential for spectroscopic applications is discussed by Colarusso, et. al. [4]. While much of the current effort in developing focalplane technology is being carried out in the area of uncooled arrays, the LN2 cooled arrays are more sensitive and in general, better suited for spectroscopy.

The initial laboratory description of this technique (5) used a cooled InSb 128 x 128 pixel array detector. The detector has a spectral response window between 3 and 5 micrometers. This allowed collection of data in the narrow spectral region between 4000 and 2000 cm^-1.

Because of the natural interest in mid-infrared spectral information, an MCT array detector was located for this application. The detector is a 64x64 pixel MCT array, with a spectral window of 2.3 to 11 microns (ca. 4000 - 900 cm^-1), which was developed for the Javelin missile program for the U.S. Army. Repackaged, it represents commercial diversification of military technology.

Both the InSb and MCT FPA detectors provide digitized data output. The data is transferred to a host Windows NT based PC that also controls the FT-IR spectrometer. This simplification of the array detector control electronics has created a system rugged and reliable enough to be considered as a commercial product.

The spectrometer system described here is a Bio-Rad FTS 6000 Infrared spectrometer with a UMA 500 Infrared Microscope. The focalplane array detector is mounted on the microscope. The MCT array detector with its 64 x 64 element array views a sample size of approximately 7 mm x 7 mm per pixel or a 400 mm x 400 mm overall sample size using the standard Bio-Rad 15x microscope objective.

The FTS 6000 spectrometer is operated in step-scan mode for the imaging application. Although the integration time for the FPA is less than a millisecond, due to the time required for readout of the entire array, the use of the step scan mode is necessary. The data from the array detector is transferred directly to memory, via a digital frame grabber card that is installed in the single Windows NT based PC, used to control both the spectrometer and the array detector.

The fixed mirror position of the Michelson interferometer is piezo controlled. This allows the interferometer to maintain continuous alignment as well as establish and maintain discrete optical positions thereby allowing data collection from the focalplane array. The spectrometer has a range of step scans speeds, which allows the user to "sit" at a single optical retardation and co-add multiple frames to improve the signal-to-noise of the resulting spectra. The following diagram shows how the data acquisition is synchronized with the steps of the interferometer.

At each step of the interferometer, defined by the summed motion of the fixed and moving mirrors, the array detector is triggered to collect image frames, one or more per step. As shown below, multiple frames are collected and co-added at each step.

The acquired data is an array of interferograms that following a Fourier transform generates a spectral hypercube. Additional processing includes library searches for unknown compositions and the ability to perform quantitative analysis after calibration or reference correction.

Using this technique we acquire a four dimensional array of data. We have a grid of points over the sample (two spatial dimensions) and at each point on that grid is an infrared spectrum (two additional dimensions -- infrared frequency and intensity). We can think of this ‘spectroscopic data hypercube’ in the following ways. First, the data can be thought of as a stack of images of the sample, each obtained at a different infrared frequency. Any card can be pulled out of the stack to ‘see’ what the sample looks like at one particular frequency. Second, an infrared spectrum exists for every point on the sample, and that spectrum can be used to identify a material, or for quantitation (how much is present).

This data is collected in a short time, just a few minutes, and is typically 16-64 MB in size. It can represent over 16,000 spectra, collected at typical analytical spectroscopic resolution, and the individual spectra can be from as small as a 5 mm square area. We obtain an infrared spectrum (a plot of infrared frequency versus energy absorbed) at each pixel, creating a total of 4096 spectra. All the spectra are collected at the same time. The data set can be thought of as a deck of cards. An individual card gives a picture of the sample at one particular infrared frequency (one ‘color’ of infrared radiation). If we drill down through the deck at one particular pixel location, we obtain the infrared spectrum of the sample at that position.

Two simple examples of this imaging chemical analysis follow. The first example is of a nylon mesh sample. The image shown below is generated from the 3350 cm^-1 band (NH stretch). The 3-dimensional image consists of 4096 resolution elements to create the infrared image. The red colour shows the area of highest absorption and the blue area represents no absorption, between the strands. This spectral region is shown selected in the spectrum shown on the right side of the figure. Because the sample consists of a single material, an image at any selected wavelength will result in essentially the same image.

The second example of chemical specific imaging is of a three-layer laminate comprised of an epoxy and a urethane component. The differences between the two materials is shown in the following two figures. The laminate is first imaged using the 1414 cm^-1 band. This band highlights the edge layers of the laminate. The last figure (Figure 5) shows an image at 1467 cm^-1 which highlights the center layer of the laminate.

Other Applications

The areas in which infrared imaging has been applied and reported are varied, including biological systems, polymer systems, material defect analysis, agricultural products, and art restoration investigations.

Some specific applications and their references are, silicone leakage from artificial implants in breast tissue[6], primate brain tissue imaging [7,8], biomineralized tissue, distribution in bone sections [9], curing of polymer blends [10], diffusion rates in Liquid crystals [11], and discrimination of co-extruded polymers [12].

Infrared imaging spectroscopic analysis is an emerging technique that provides chemical specific information as images. From these images, precise information about the chemical composition and material structure of a sample is obtained. This type of information is significant as it allows the investigator to understand more fully a chemically diverse sample, which is generally the rule rather than the exception when studying either made-made or naturally occurring materials. Due to the ease of infrared image generation, and the power of image analysis software tools, the ability to study complex systems is now readily available so that the analysis and applications become the focus rather than the experimental technique itself.

References

1. Michael R. Carter, Charles L. Bennett, David J. Fields and John Hernandez, Gaseous effluent monitoring and identification using an imaging Fourier transform spectrometer, in Substance Detection Systems, Proc. SPIE 2092 16-26 (1993).
   
2. Charles L. Bennett, Michael R. Carter and David J. Fields, Hyperspectral Imaging in the Infrared Using LIFTIRS, in Infrared Technology XXI, Proc. SPIE 2552, 274-283 (1995).
3. P. Cox, J. P. Maillard, P. J. Huggins, T. Forveille, D. A. Simmons, F. Riaut, R Bachiller, S. Guilloreau, A. Omont, Astron. Astrophys., 1996.
4. Pina Colarusso, Linda H. Kidder, Ira W. Levin, James C. Fraser, John F. Arens and E. Neil Lewis, Infrared Spectroscopic Imaging: From Planetary to Cellular Systems, Appl. Spectrosc. 52, 106A-120A (1998).
5. E. N. Lewis, P. J. Treado, R. C. Reeder, G. M. Story, A. E. Dowrey, C. Marcott, and I. W. Levin, "FTIR spectroscopic imaging using an infrared focal-plane array detector", Anal. Chem., 67, 3377 (1995).
6. L. H. Kidder, V. F. Kalasinsky, J. L. Luke, I. W. Levin and E. N. Lewis, "Visualization of silicone gel in human breast tissue using new infrared imaging spectroscopy", Nature Medicine, 3(2), 235-237 (1997).
7. E. N. Lewis, A. M. Gorbach, C. Marcott, and I. W. Levin, "High-Fidelity FTIR spectroscopic imaging of primate brain tissue", Appl. Spectrosc. 50, 263, (1996).
8. E. Neil Lewis, Linda H. Kidder, Ira W. Levin, Victor F. Kalasinsky, Joseph P. Hanig and David S. Lester, Applications of Fourier Transfrom Infrared Imaging Microscopy in Neurotoxicity, in Imaging Brain Structure and Function, Annals of the New York Academy of Sciences, 820, 234-247 (1997).
9. Curtis Marcott, Robert C. Reeder, Eleftherios P. Paschalis, Dimitris N. Tatakis, Adele L. Boskey and Richard Mendelsohn, Infrared Microspectroscopic Imaging of Biomineralized Tissues using a Mercury-Cadmium-Telluride Focal-Plane Array Detector, Cellular and Molecular Biology, 44, 109-115 (1998).
10. Sung Joon Oh and Jack L. Koenig, Phase and Curing Behavior of Polybutadiene/ Diallyl Phthalate Blends Monitored by FT-IR Imaging Using Focal Plane Array Detection, Anal. Chem. 70, 1768-1772 (1998).
11. C. M. Snively and J. L. Koenig, Application of Real Time Mid-Infrared FTIR Imaging to Polymeric Systems. 1. Diffusion of Liquid Crystals into Polymers, Macromolecules, 31, 3753-3755 (1998).
    
12. John M.Chalmers, Neil J. Everall, Karen Hewitson, Michael A. Chesters, Martin Pearson, Andrew Grady and Barbara Ruzicka, Fourier transform infrared microscopy: some advances in techniques for characterisation and structure-property elucidations of industrial material, The Analyst, 123, 579-586 (1998).

T. Jawhari ^(a) and A. Pérez-Rodríguez ^(b)

Christopher T. Zugates and Patrick J. Treado

ChemIcon Inc.,
7301 Penn Avenue,
Pittsburgh,
PA 15208,
USA

Abstract

Raman spectroscopy is an efficient means for probing molecular composition, and structure without being destructive to samples. Raman chemical imaging using liquid crystal tunable filter (LCTF) technology is a massively parallel approach that extends traditional Raman spectroscopy to provide information on sample morphology. As a result, Raman chemical imaging is an effective high throughput screening tool for the analysis of pharmaceutical tablet content uniformity. Analysis is performed without sample preparation and characterization of tablets in situ is feasible, even in manufacturing environments as a quality monitoring tool.

Introduction

Rapid, high throughput assessment of the composition, structure and uniformity of active ingredient distribution in drug tablet formulations is of great interest to the pharmaceutical industry. Understanding these parameters is critical to tablet quality monitoring and control. Typical analytical strategies for performing tablet assays often involve invasive sample preparation procedures, including tablet crushing, dissolution and chromatographic separation of active ingredients from excipients. Where tablet content uniformity must be measured, invasive sample staining procedures are often required to generate image contrast between active ingredients and excipients. Active ingredient distribution can then be visualized using optical microscopy, for example. While traditional analytical strategies are effective, they are labor intensive due to the extensive sample preparation required and the techniques are applied to only a very limited number of tablets.

Infrared spectroscopy performed both in the mid IR [1] and near IR [2] provides the potential of rapid determination with little or no sample preparation. Raman spectroscopy also has demonstrated capability for pharmaceutical analysis.[3,4] Vibrational spectroscopic techniques are effective for compositional and structural characterization, as well as quantitation. However, bulk spectroscopy is ineffective for measuring the spatial distribution and architecture of actives which are heterogeneously distributed within intact tablets.

Recent advances in Raman liquid crystal tunable filter (LCTF) imaging spectrometers[5,6] combined with multivariate image processing techniques[7] make Raman chemical imaging a powerful technique for the analysis of wide variety of materials,[8] including pharmaceutical tablet architecture. Some of the materials studied to date include, corrosion systems, [5] polymer blends,[7,9] biological tissues,[10] actinide-contaminated ash samples,[11] surfactants, Martian meteorites [12] and semiconductors.[13] Chemical imaging combines Raman spectroscopy and digital imaging technology to make the assessment of tablet molecular composition and structure a routine analytical procedure. In Raman imaging using LCTFs, thousands of linearly independent, spatially-resolved spectra can be collected simultaneously to differentiate active ingredients even in the presence of complex host matrices. These spectra can then be processed to generate unique contrast intrinsic to analyte species without the use of stains, dyes, or contrast agents. As a result, there is little or no need for sample preparation to characterize heterogeneously distributed active ingredients.

Raman imaging spectroscopy data was collected from a pharmaceutical tablet containing acetylsalicylic acid (aspirin) as an active ingredient using the FALCON Series 3500 Raman imaging microscope system developed by ChemIcon Inc. The FALCON supports a wide variety of laser excitation sources. In the Falcon system shown in Figure 1, a diode pumped Nd:YVO₄ solid state laser source doubled to operate at 532 nm is coupled with a multimode fiber optic relay to an infinity-corrected optical microscope. Coupling is performed via an illuminator assembly that converts the infinity-corrected optical microscope to a Raman imaging platform. The illuminator defocuses the laser source and excites the entire sample field of view through a high numerical aperture (NA) microscope objective. The Raman scattering is collected with the same objective and is transmitted back through the illuminator which houses holographic notch rejection filters to remove the Rayleigh scattering. The Raman signal is filtered with a 9 cm^-1 bandpass liquid crystal tunable filter (LCTF) constructed using the Evans Split-Element geometry. The LCTF is a compact electronically tunable filter that provides spectral resolution over the entire Raman spectrum comparable to a dispersive spectrometer without compromising the diffraction-limited imaging performance of the optical microscope. Raman images are collected using a thermo  electric (TE) cooled (-40 ^oC) slow-scan charge-coupled device (CCD) detector having 512 x 512 (20 mm square) pixels.

Multi-dimensional micro-Raman image data sets were collected under computer control of the LCTF and CCD detector using Acquisition Manager 4.0 software we have developed for commercial use. Multispectral Raman image data sets are processed in ChemIcon's SpecImage 4.0 software to remove contributions from background fluorescence and enhance molecular-specific contrast using univariate and multivariate analysis approaches. Raman spectra for individual pixels or average spectra for larger domains are visualized using ChemImage 4.0 and extracted for data presentation. More traditional morphometric-based image processing (particle counting, sizing and feature segmentation) is also performed using ChemImage 4.0.

Chemical Image Analysis

We employ multivariate image analysis techniques that involve the application of chemometric techniques to spatially resolved spectra. Chemical (i.e. hyperspectral) imaging experiments produce vast amounts of data and are often subjected to data reduction techniques as an initial processing step. The intent of data reduction is to simplify the analysis by preserving the most important structures found within the data. Less important data structures including instrument response, sample background, and noise are ideally not considered during subsequent analyses.

Factor analysis, a first class of multivariate analysis techniques, is often performed to reduce the amount of data. For example, principal components analysis (PCA) is performed on the data to extract factors or scores that best represent the variation in the chemical image data. PCA reassembles the data as linear combinations of the original variables so that the largest variance in the data corresponds to the first principal component. Each subsequent principal component is orthonormal to the previous component and represents the largest remaining variance in the data. The maximum number of principal components allowed is equal to the number of variables measured and maintains the data structure but does not reduce the dimensionality of the data. Typically, the smallest set of principal components necessary to represent some large percentage of the total variance in the data is used for further analyses.[14]

A second class of multivariate analysis, cluster analysis, identifies differences between the spatially-resolved spectra and sorts them into categories based on differences in their measured variables. Each group of objects is treated mathematically as a single object that represents the variable mean of the cluster to a precision described by the variance or standard deviation within the cluster.[14]

Once the data has been preprocessed, the next task is to estimate the composition of the sample. Conventional algorithms, such as partial least squares, classical least squares, and multiple linear regressions, employ a variety of regressions that fit data populations to a predescribed set of observations made on samples of known composition. The data describing the measurements of the known samples are collectively referred to as a training set and are necessary for most multivariate analyses. In practice, a priori information about a specific sample may be unavailable or incomplete making conventional analysis requiring the use of training sets difficult or nearly impossible. In complex materials, training sets may not be adequate for describing all of the complex chemical interactions that arise in heterogeneous samples. Even when possible, constructing spectral image training sets is labour intensive.

Cosine correlation analysis (CCA) is a multivariate technique that assesses similarity in spectral data sets and image data sets.[7] CCA assesses chemical heterogeneity without the need for training sets. CCA identifies differences in spectral shape and efficiently provides chemical based image contrast that is independent of absolute intensity which makes it well suited to Raman chemical imaging.

Performing CCA using one reference vector does not insure image contrast between pixels representing different chemical compositions since many different spectra may be reduced to the same cosine score. For example, the collection of spectral vectors having identical cos q values can be conceptualized as lying along the periphery of the multidimensional cone ( a hypercone) having the reference vector for its axis of symmetry and its vertex at the origin. In order to remove all symmetry, p correlations need to be performed on the image data set using p different reference vectors. Even for large p values, most of the symmetry can be removed using only a few reference vectors. As in principal component analysis, it is often unnecessary to perform all possible correlations in order to achieve a high degree of chemical specificity. Ideally, the reference vector space should span the maximum variance of the data space so that most of the variance of the data is contained along the directions of the first few orthonormal reference vectors.

The choice of reference vectors can influence the efficiency of the data reduction. It is not necessary to know the number of pure components in the sample using this method. In fact, a histogram of the scores often reveals the number of spectral types and is useful in determining the number of components in a chemical system. If the number of components is already known, CCA can be used to determine if unknown species (i.e. foreign contaminants) might be present. Even for small numbers of correlations, all wavelength information about the sample is employed in generating the correlation score allowing unsuspected differences in spectral shape to produce image contrast. Chemical identity can be assigned by matching correlation scores from the image data set to correlation scores calculated for the spectra contained in a spectral library using the same CCA routine.

Figure 2 shows images of an over-the-counter pharmaceutical tablet containing aspirin as an active ingredient. While the pressed tablet appears homogenous in the macroscopic-scale image shown in Figure 2A, the tablet appears heterogeneous in the microscopic-scale reflectance image shown in the inset, Figure 2B. The heterogeneity visible in Figure 2B is not surprising given the rough surface topography of the pressed   tablet. However, it is challenging to correlate the microheterogeneity visible in Figure 2B with microheterogeneity of the active (acetylsalicylic acid) and excipient (calcium carbonate) materials because both species are white powders and cannot readily be differentiated within an intact tablet using conventional contrast enhancement strategies employed in optical microsopy, including color, differential interference contrast (DIC), nomarski and polarized light.

Molecular compositional heterogeneity does exist within the sample at the microscopic level. Raman data shown in Figure 3 provides convincing evidence for the presence of compositional heterogeneity. Figure 3A is the brightfield reflectance image shown in Figure 2B. Raman microspectra are shown at random sampling points, 1-3. Figure 3D shows Raman microspectra captured through the LCTF at the sampling points 1-3. It is obvious from the Raman spectral band shapes that the local composition is varying. For example, the active ingredient, acetylsalicylic acid, has a band at 1044 cm^-1 while the calcium carbonate excipient has a band centered at 1060 cm^-1. The relative intensities of these two bands indicates the relative concentration of the materials which are clearly heterogeneously distributed at the spatial resolving power of the experiment.

While the randomly sampled Raman microspectra confirm heterogeneity, Raman imaging is required to visualize the distribution of the materials. Figure 3B shows the Raman band intensity collected by tuning the LCTF to 1044 cm^-1 and Figure 3C is collected at 1060 cm^-1. Note the similarity between the two. Due to the fluorescence background emitting from the tablet, image contrast is dominated by the surface topography of the sample.

Susceptibility to surface roughness is a general limitation of Raman imaging based on scattered intensity. However, this problem can be avoided by making use of the Raman spectrum at each linearly independent pixel to generate images based on other spectral band parameters, including subtle differences in spectral shapes. A very effective strategy that compensates for varying sample topography is cosine correlation analysis (CCA), as shown in Figure 4. Figure 4A is a 1044 cm^-1 correlation image revealing active ingredient distribution, including the presence of two particles between 20-40 µm in diameter. Figure 4B is a 1060 cm^-1 correlation image revealing excipient, which are more uniformly distributed than the actives. Note that CCA is spectral shape dependent, but insensitive to local concerted variations in spectral intensity due to surface topography. As a result, the CCA images shown in Figure 4 differ substantially from the Raman intensity images of Figure 3. Figure 3 and Figure 4 both show microstructure from a small region of the tablet surface. While Figure 3 is showing surface topography, Figure 4 is showing surface chemical composition.

The two active ingredient particles are visualized in Figure 5A, a chemical surface map derived from the two-dimensional image of Figure 4A. However, the compositional texture of the tablet becomes more obvious when seen as a surface map. Figure 5B shows the chemical surface map for excipient distribution. As anticipated, where the active ingredient is localized, the excipient is absent.

It is striking that the 1044 cm^-1 and 1060 cm^-1 spectral features can be used to generate molecular chemical image contrast, as the band centers vary by only 16 cm^-1. The images shown in Figures 4 and 5 would be very challenging to obtain using alternative imaging spectrometer technologies. The spectral resolving power of the LCTF is very high due to the narrow spectral bandpass (<9 cm^-1) and fine tunability, especially when combined with the excellent sampling statistics available in a typical imaging experiment. For example, it has been demonstrated that the LCTF has a resolving power of greater than 0.03 cm^-1 in the study of semiconductors.[15] With regard to other sampling characteristics, LCTF Raman microscopy provides diffraction-limited spatial resolution of 250nm with appropriate high numerical aperture microscope objectives. To capture the raw image data of Figure 3, an integration time of 5 sec/spectral band was employed. Typically, integration times range between 1-30 secs/spectral band and depend on the scattering cross-section of the analyte of interest, its relative concentration and background emission (fluorescence) levels. Typical sample areas that are analyzed from square mm down to square m.

Conclusions


The recent advancement in LCTF Raman imaging technology is changing the way Raman spectroscopy is being employed for materials characterization, including pharmaceutical tablet analysis. A driving force for change is the desire to understand how sample morphology, composition and structure influence the physical and chemical properties of complex materials. LCTF Raman chemical imaging is gaining wide acceptance by providing users unprecedented capability for molecular characterization of materials, even materials that traditionally have been difficult to assess. LCTFs coupled to CCD detectors facilitate wide field Raman image collection and allow experiments requiring excellent image fidelity to be performed in less time than conventional scanning techniques. By harnessing multivariate image processing, it is becoming possible to analyze materials without sample preparation even in manufacturing environments.

Raman chemical imaging is not limited to microscopic analysis and can be performed on large samples as well. Macroscopic Raman imaging requires lasers to have sufficient power to maintain high power densities across large sample areas. For sample areas of a few square inches, it is possible to attain sufficient laser power with commercially available lasers. Wide field Raman chemical imaging of larger samples requires greater laser power and becomes impractical until mainframe solid state lasers providing higher powers (>10W) become commercially available. However, new generations of diode laser technologies are continually being developed which will facilitate LCTF Raman chemical imaging becoming a universal imaging technique.

References

1. J. A. Ryan, S. V. Compton, M. A. Brooks and D. A. C. Compton, J. Pharm. Biomed. Anal. 9, 303 (1991).
2. J. D. Kirsch and J. K. Drennen, Appl. Spectrosc. Rev. 30, 139 (1995).
3. C. J. Petty, D. E. Bugay, W. P. Findlay and C. Rodriquez, Spectroscopy 11, 41 (1996).
4. T. M. Niemczyk, M. Delgado-Lopez, and F. S. Allen, Appl. Spectrosc. 52, 513 (1998).
5. Morris, H. R.; Hoyt, C. C.; Miller, P.; Treado, P. J. Appl. Spectrosc. 50, 805 (1996).
6. Turner, J. F., II; Treado, P. J. Proc. SPIE – Int. Soc. Opt. Eng. 3061, 280 (1997).
7. Morris, H. R.; J. F. Turner, Munro, B.; Ryntz, R.; Treado, P. J. Langmuir 15, (1999) in press.
8. M. D. Schaeberle, H. R. Morris, J. F. Turner II, and P. J. Treado, Anal. Chem. 71, (1999) in press.
9. H. R. Morris, B. Munro, R. Ryntz, P. J. Treado, Langmuir 14, 2426 (1998).
10. N. J. Kline and P. J. Treado, J. Raman Spectrosc. 1997, 28, 119-124.
11. G. J. Havrilla, J. Schoonover, F. Weesner, M. C. Sparrow, P. J. Treado, Appl. Spec. 1998, 52, in press.
12. Treiman A. H. and Treado P. J., Lunar Planet. Sci. XXIX, 523 (1998).
13. Schaeberle, M. D. Treado, P. J. Proc. XVth ICORS, S. A. Asher, Ed., (Wiley, Chichester, 1996) 1188.
14. P. Geladi, and H. Grahn, Multivariate Image Analysis, John Wiley and Sons, Inc., New York, NY, 1996.
15. M. D. Schaeberle, Raman Chemical Imaging: Development and Applications (Doctoral Dissertation, University of Pittsburgh), 1998.

Perkin Elmer's New Instrument
- Spectrum One


Your Editor’s Totally Prejudiced description

Immediate reaction – small,  neat and looks smart. Totally computer controlled with no equivalent of the built-in control panel found on the PE1600 and Paragons and the machine comes with several accessory pods. The Nicolet Avator but coloured P-E grey, I hear you ask? No I don’t think it is. I think it’s significantly different. The specification looks excellent and quite clearly the machine would be useful in a research laboratory.

Next reaction – has several clever features. Not features aimed at me – an old spectroscopist, but rather at the young chemist or materials student  and occasional user or the QC analyst. One feature I really like is a touch screen set up to make spectroscopy absolutely idiot proof.

But then let me tell you about it

Basic Layout
Spectrum One is set out in the now contemporary way – a very small footprint on the bench with a full-size sample area symmetrically situated in the front. It’s about a foot high (300mm) and has a depth and width of around 2 ft sq., so it’s area even allowing for the cables at the back is considerably less than a square metre. It weighs around 30kg (60lbs), not light but unlikely to bend the bench top. A computer controls the instrument, but this doesn’t have to be adjacent to the optical bench. The computer can be in a separate room or even miles away.

The sample area (and let’s face it, this is the bit the user actually wants to know about) in it’s transmission form, has a horizontal plate onto which is mounted the industry standard 2" x 3" slideway stand and the whole lot is closed by a detachable lid.

To the right of the sample area, you will see a rectangular panel. In strip-down trim this acts as a simple status indicator with diodes to tell you that the source and laser are working and whether the interferometer is scanning or not.

In it’s higher specification, Spectrum One has a clever touch sensitive liquid crystal display in this position. The touch screen enables occasional and non-expert users to use a custom-made procedure especially prepared for them and to do so with minimal skill.

In Figure 2, you will see the touch screen set up to carry out a horizontal ATR based routine. Once called up, the screen requires the user to insert the appropriate accessory. Pressing the touch screen takes the user through all the essential steps - checking the accessory, insertion of the sample, optimization of the system, recording the spectrum and its presentation in an agreed format on the computer screen, and if required on paper through a printer. This feature seems to me to be ideal for use in teaching, in service laboratories and of course, in routine analysis. In the past, instrument makers have tended to produce instruments that require the user to understand how the system and accessories work. However, highly expert chemists or pharmacists, materials scientists or skilled analysts want to use the instruments and interpret the results. Their cross-section of skills do not include experimental spectroscopy so the approach adopted in Spectrum One is timely and highly appropriate.

I have already said above that the computer can be remote – it can be – it can be networked to control several instruments, but if it is, you need a keyboard at the machine to key in the sample identification. I was even more intrigued to discover that for QC and similar measurements, you don’t even need a keyboard – you can use a bar code reader to identify the sample!

When the spectrum is running, a very attractive icon – a tiny spectrum wriggles before your very eyes at the bottom of the screen. See Figure 2. When the spectrum is acquired it appears in miniature form on the touch screen, see Figures 2 & 3. In many installations a whole gang of users will want to use the machine each with a different set of requirements. This too can be accommodated through a password system.

In the end, people buy on performance. I saw some impressive results on stability and signal:noise ratio (essentially the quality of the spectrum) significant if you are looking at weak bands. The performance is better by more that two in both respects than the P-E Paragon, the machine that Spectrum 1 replaces. Range 7800-350cm^-1 but extension to 200cm^-1 is promised very shortly and I suspect nir will follow soon. Resolution - ½cm^-1, so its more than adequate unless you are seriously interested in gases. Almost all FTIRs have the source and interferometer sealed inside a desiccated base but the manufacturers usually leave the detector area open to the atmosphere. Spectrum One encloses everything other than the sample area. Must be a good idea.

Accessories

Three accessory 'pods' are on offer at launch. You remove the sample area lid, slide out the standard absorption base and then slide in the accessory of your choice (limited only by your chequebook). As you do, a small processor on the accessory identifies it in detail (down to its traceability for QC work) and sets up alignment motors built into the pod. Fit the top plate (in the case of ATR) of your choice and another processor defines this. The machine then automatically optimizes so the accessory alignment is ready for you to proceed. In the diffuse reflection accessory, this automatic alignment routine is repeated when the sample or reference are introduced.

Horizontal ATR

The ATR crystal (ZnSe or Ge) is pretty standard in size and can be illuminated at 30, 45 or 60º. To achieve this you need a set of top plates one for each combination of crystal and angle. The top plate is instantly interchargeable and as I have explained, each plate contains a processor, which tells the machine the crystal material and the angle. Troughed and flat plates are, of course, available. Needless to say, you can press the sample on the crystal if it’s a flexible solid. Another processor lower down in the pod controls the alignment routine so the user has to do little or nothing other than putting the sample on the crystal.

Many users may not be aware that changing the angle of the horizontal ATR can be beneficial. By changing the angle the number of reflections can be altered and hence the sensitivity of the technique adjusted. An example is shown in Figure 5.

Universal ATR

P-E’s euphamism for diamond ATR – your learned Editor’s favourite technique! Again, its in the form of a plug-in pod and comes with several diamonds allowing 1, 3 or 9 reflections. As with the horizontal ATR, the processor in the top plate sets up the instrument and the second processor in the lower section sorts out the alignment.

In diamond ATR (a bad description because you can use a Geranium [Sorry Germanium –  I couldn’t resist that one! -Assistant Editor]  prism for some purposes and P-E will oblige. You must press the sample really firmly into the crystal and the force is important. It is essential you apply enough but not too much. In all the existing diamond ATRs available from accessory manufacturers, a torque wrench is normally provided. You wind the pressure screw to a recommended set torque. I love torque wrenches but then I’m a freaky DIY auto mechanic(!), but many people find the procedure clumsy. On the Spectrum One, P-E supplies a force SENSOR. Twiddle the knob to press down on the sample and a green bar grows from left to right on the computer screen, so you know exactly what you are doing. Overdo it and the bar changes to red and flashes angrily at you to tell you to ease off. Have a look at Figure 7. A similar force sensor is fitted to the horizontal ATR.

Diffuse Reflection DRIFTS

DRIFTS is a good technique. Mix a solid with dry KBr at around 5%, put it in a stainless steel cup and reflect mid i.r. off it. Alternatively, rub a small abrasive pad over a specimen and look at the reflection spectrum of the 'loaded' pad. The problem with the technique has always been "skill". You have always had to align the sample in the accessory – get the focus right (focus the incident radiation accurately onto the sample surface) and get the alignment ‘spot on’ so that the reflected light is efficiently collected. All very fiddly! The P-E Spectrum One DRIFTS accessory is fully motorised – when you shove the accessory into the spectrometer, the mirrors adjust the set up to produce a maximum in intensity. So, how do you load the sample? Well, you either fill a cup with KBr/sample mixture and scrape off a nice flat surface OR you rub the sample on the disposable aluminised abrasive pads OR you rub the provided disposable "sticks" on an immovable sample. You then drop the sample into a sample slide rather like a microscope accessory slide, push it into the first position and run the background. Push the holder the whole way in and run the sample. As the reference or sample is shoved into the accessory, the machine re-aligns to maximise the quality of the spectrum. (Even I, after a bottle of Australian Chardonnay couldn’t get it wrong!)

Software

Two bits of software took my eye, but there was an enormous amount to consider. When you run a spectrum on any F-T instrument the spectrum you actually record has a distorted band-shape and band position. The same problem applied in the dispersive days but no one noticed because the spectra were so bad. The problem is that the band-shape and position vary unpredictably from experiment to experiment. This difficulty is fairly subtle and most people don’t notice it, but it does spoil accurate quantitation measurement.

Absolute Virtual Instrument (AVI)


The P-E team has devised an Absolute Virtual Instrument (I know a Virtual Instrument is all you can afford!) The optical bench incorporates a tiny cell containing methane. When instructed, it runs the methane spectrum at whatever resolution you happen to have selected. The machine then generates within memory a spectrum it SHOULD have produced based on the high-resolution data available in the literature. It then compares its real spectrum with the theoretical one and generates a correction algorithm – very clever! Does AVI work? Superbly! If a sample is recorded by several methods, it is inevitable that the perturbation caused by the DRIFTS accessory is not identical to that caused by the ATR pod etc. [AVI does not allow for effects arising from the reflection process itself]. Or, spectra recorded on one instrument from those on another or your instrument has been serviced and hence spectra recorded before and after the service visit are not identical. The errors are not too bad but they are significant in quantitative analysis. Apply the Virtual instrument process and hey presto the spectra all look the same. Look at the example in Figures 10a and 10b.

Atmospheric Absorption Suppression

When you run an infrared on a single beam instrument (and all FTIRs are such) you record a background, then a spectrum and the computer then point by point divides one by the other, the spectrum displayed in transmission (0-100%) being a plot of
( ^{I sample}/_{I reference}) vs v. Theoretically atmospheric CO₂ and water vapour should give no bands because they appear equally in both the sample and reference. In reality, the CO₂:H₂O ratio is always changing and the H₂O level varies continuously and hence real spectra often contain positive or negative CO₂ bands and also weak water bands.

These spurious features really are a nuisance particularly if you are looking for weak V_C=O or V_OH bands. As part of the Spectrum One package you can have a software set to automatically remove these bands. It works really well and is fully automatic.

IR Expert


This package appealed to me. In essence it looks at the spectrum you have recorded, tells you how well you have done and then applies corrections to get the best out of your data. In Figure 12, you see a spectrum with a high noise level and a quality Report. Note the yellow and red lights at the top. This spectrum didn't earn a green light. The copy tells you what is wrong. If you then issue the appropriate instruction, the computer will tidy up your handywork. See Figure 13.

To carry out the corrections only a single click is needed - the processes are fully automated. As well as identifying the problems and correting them the folks at P-E offer advice on how to do better next time - ideal for students and occasional users.
The advice is intoned by a lady with a Scottish accent. At the launch meeting an American marketing man asked - "Why Scottish?",   "because no-one hates the Scots" was the reply! After all, they never embarass anyone by getting to the Final group in the World Cup!

Price

Impossible to give one because as is normal in the instrument industry, the options are huge and the price reflects what you decide to have and how good you are at haggling with the salesman. Put in CD parlence, the Spectrum One is a mid-priced option. It’s certainly not budget price or full price.

Summary

Now the crunch - I accept you would like me to tell you whether to buy Spectrum One. This is beyond the Journal's remit because we have nothing to compare it with. This report is NOT a road test - it's a description.

I am prepared to tell you what I liked.

I thought it would be ideal in my research lab because we don't do high resolution spectroscopy on gases. The S:N ratio is superb and it's dead stable, so the machine is ideal for working on weak bands. I think Spectrum One is just what is wanted for teaching and service labs - the touch pad and the totally sealed and dessicated optics are attractive. Being able to set up a routine for a particular problem is ideal especially because I don't have to teach the student or synthetic chemist how to drive the computer.

QC and routine analytical applications - Validation standards can be built in. Search and Quantitative packages are all on offer and the stability must be a very important consideration in this field.

So, there it is - my immediate impression.

[IJVS Home]  [Hot Sources]  [Links]  [Bookshelf]  [Contents]
[Publication Details]  [Help]  [Back Issues]  [Feedback]

Copyright © 2004 John Wiley & Sons, Ltd