Introduction

There has been a significant andgrowing interest in chemicalimaging of pharmaceutical tabletsin recent years. Technologyimprovements and increased usehas established chemical imagingas a means of providing informa-tion on certain critical qualityparameters that is not readilyaddressable using other tech-niques. In addition to majorpharmaceutical manufacturers,generic manufacturers and evenregulatory authorities1 nowrecognize the value of chemicalimaging both in troubleshooting inmanufacturing and scale-up andin root cause analysis2. Of thevarious imaging techniques, NIR

imaging has probably (andjustifiably) received most atten-tion, and is poised to grow furtheras a PAT-enabling technique3.Other chemical imaging tech-niques such as Raman and mid-IRdiffuse reflectance each have theirown merits and disadvantages,but NIR is attractive from thetechnology, implementation andsampling perspectives. Datagenerated by this method is alsomore familiar to many users inthis area who might have experi-ence in macro-NIR spectroscopyof tablet ingredients. That said,NIR imaging using currenttechnology (be it FT, tunable orfixed filter) suffers some intrinsic

ATR Imaging of Pharmaceutical Tablets

AP

PL

IC

AT

IO

N

NO

TE

FT

-IR IM

AG

ING

www.perkinelmer.com

limitations due to the nature ofthe measurement itself, not thetechnology per se. First, thesample penetration by theincident light is relatively deep(many tens of microns) and theresulting image is effectivelysmeared out by the nature ofdiffuse reflectance. We estimatethat the sample penetration is atleast this order (dependent onparticle size, packing and otherparameters), yet we also knowthat many tablet ingredients areintroduced into the blender witha particle size of less than 10microns. So the technique mustbe sub-optimal for the purpose ofgenerating sharp, distinct, pure,

ingredient images which show theingredient’s spatial distributionunless these ingredients formaggregates. Of the publications onNIR imaging, to the author’s knowl-edge, no publication has shownparticle size distributions for singleparticles of less than 10 microns insize. The examples tend to focus onshowing aggregates of small particlesor larger particles of many tens orhundreds of microns in size, yetparticles of less than 10 micronsmust exist in the distribution. It isalso known that for certain ingredi-ents, such as magnesium stearate,the component is sometimes difficultto detect with NIR imaging even atrelatively high concentrations unlessit forms aggregates of ca 30-40microns or larger, due to the distri-bution of the material throughout thematrix as relatively fine particles.Indeed, some actives at the sub-percent level are not seen at allunless they are present as aggregates.Second, the measured NIR spectra inimages tend to be highly overlapped.In macro-NIR spectroscopy, NIRspectra are generally less discrimi-nating than their correspondingRaman or mid-IR counterparts andthis disadvantage is compounded inNIR imaging, where signal to noiseratios tend to be lower than on themacro-scale. The result of these twofactors is increased reliance onchemometric analysis of the spectraldata in the image. Moreover, subse-quent image analysis also uses pixelcounting statistics, which oftendepend on the requirement toclassify individual image pixels asbelonging to one chemical class orother. In practice, this represents adifficult step in the analysis when itis known that one rarely observespure component spectra at singlepixels in NIR images. This rendersthe overall image analysis moredifficult in the NIR as the decision in

assigning a ‘1’ or ‘0’ to a gray-scalepixel is often somewhat subjective.

ATR imaging in the mid-IR offers theopportunity to provide some majoradvantages in this applicationrelative to other techniques.Compared with NIR imaging, itoffers a large improvement (ca anorder of magnitude) in spatialresolution – now of the order of theindividual particle sizes of manyingredients (less than 10 microns)when they are introduced into theprocess. Better than 5 micronsspatial resolution4 has been demon-strated with ATR imaging. Second,the spectra in the images tend to beof very high contrast, and moreeasily discriminated than in NIR.Third, the technique is effectivelya surface technique, with a samplepenetration depth of ca 1-2 microns,using the optical arrangementdescribed below. The result is thatindividual pixel spectra are relative-ly more pure and the burden on thesubsequent image analysis is lowerthan in NIR. There is no reason whysome of the techniques used inhyperspectral NIR image analysiscannot be used for ATR images.Other advantages of ATR include thefact that large commercial librariesare available and may be used forqualitative identification, and somematerials that are difficult to identi-fy with NIR, such as some anhy-drous inorganic salts, give highcontrast ATR spectra.

We have developed an ATR accesso-ry for the PerkinElmer® Spotlight™

300 and 400 imaging systems andmeasured a number of pharmaceuti-cal tablets. This note describes ATRimaging measurements of commer-cial pharmaceutical tablets anddemonstrates a number of theadvantages described above. As such, this technique shows a major

advance in state of the art pharma-ceutical tablet imaging. Although,still at the ‘infancy’ stage, it isexpected to grow and to play a verysignificant role in future pharma-ceutical tablet and blend analysis.

Experimental

Sample Preparation andPresentation

For this study, whole uncoatedpharmaceutical tablets (bothprescription drugs and over-the-counter - OTC) were obtained alongwith samples of the raw ingredientsknown to be present in the tablets.For the prescription product, theingredients were active, microcrys-talline cellulose, calcium phosphatedibasic, starch and magnesiumstearate. For the OTC product, theraw ingredients were aspirin,paracetamol, caffeine and a numberof organic and inorganic minoringredients. In this instance, thetablets were 2-3 mm thick, withvertical sides and flat surfaces.It was not necessary to performany special milling/cutting onthe surfaces, although this wouldbe required for curved surfaces(this is one limitation of the ATR technique).

The ATR accessory comprises agermanium crystal with a conicalcross section, Figure 1. The tip ofthe crystal is pressed into the tabletsurface. The accessory is designedsuch that the sample region ofinterest (ROI) can be located usingthe visible microscope camera andthe ATR sampling surface broughtdown onto the ROI without movingthe sample. The crystal tip allows acircular sampling area of diameterca 0.5 mm in which various sizes ofsquare or rectangular image can bedefined. Given an image pixel sizeof 1.56 microns, an area of 500 x 500

2

microns would contain more than102000 individual spectra.

In order to perform the most thor-ough image analysis, it may benecessary to obtain reference ATRspectra from the raw ingredients ofthe tablet. For best results, thespectra should be measured usingthe ATR imaging system, as opposedto extraction from standard refer-ence libraries, although the lattermay be used with some degradationin the image analysis. In this study,simple micro-disks were preparedfor each of the raw ingredients of the

tablets using a standard KBr micro-pellet accessory commonly used inIR spectroscopy. Using a standardmicro-pellet (2 mm die) accessory,hand-pressure is sufficient togenerate a disk for ATR data collec-tion. Once pressed, the disks aretransferred to the ATR accessory andsmall images (50 x 50 microns) ofthe pure components are measured.From these images, single referencespectra are computed using theSpotlight software. Collection ofreference images takes a few min-utes per ingredient. This operationis generally performed only once.

Imaging

Images were collected using astandard PerkinElmer Spotlightmid-IR imaging system equippedwith an ATR imaging accessory. Thesystem uses a small MCT detectorarray combined with a movingsample stage/ATR accessory togenerate images. The ATR crystal isclamped to the sample and does notmove relative to the sample – theentire assembly of sample and ATRcrystal are moved relative to the IRdetectors. Image pixel size is 1.56 x1.56 microns (6.25 x 6.25 micronspixel size is also available) and aspectral resolution of 16 cm-1 was

used for the images. For this study,actual image sizes varied from 100 x 100 microns to 400 x 400microns with total image collectiontimes varying from 1 minute to ca40 minutes.

Image data processing includedspectral derivatisation, normaliza-tion and principal componentsanalysis, and least squares fullspectrum curve fitting performedusing the PerkinElmer Hypersoft 3.0software. Some further graphicalimage enhancement and statisticalanalysis was performed using ImageJVersion 1.36b public domain imageprocessing and analysis software5.

Results and Discussion

A visible image from a tablet surfaceis shown in Figure 2a. Although it ispossible to discern some granules inthe visible image, there is no effec-tive way of assigning the particles toindividual ingredients from thevisible image alone. Figure 2b on theother hand shows how the starchcomponent is distributed. This wasderived from the ATR image follow-ing a number of data processingsteps which will be outlined in thissection. Individual ingredientdistribution patterns may be derivedin a similar manner, allowing a

3www.perkinelmer.com

Figure 1. ATR sampling configuration.

Figure 2. Visible (a), derived single component IR image (b) from tablet surface and (c) composite image from 3 major components.

composite image to be constructed.In this case, the three major compo-nents were merged to a single RGB(red/green/blue) composite image(Figure 2c). This kind of representa-tion is common in NIR imaging ofthese products. What is different inthis measurement is the muchimproved spatial resolution com-pared with NIR imaging and thehigher degree of spatial contrastbetween image pixels to yieldsharper, more detailed images fromthe tablet surface–in relatively shortmeasurement times, and the abilityto locate domains which are current-ly not detectable using NIR imaging.

Raw Ingredients Spectra

Figure 3 shows the ATR spectra ofthe raw ingredients for a tablet,alongside their corresponding firstderivative spectra. With ATRimaging, it is observed that the fieldof illumination is not entirely flatacross the entire face of the crystalelement, and this manifests itselfprimarily as a smooth baseline offsetvariation in spectra measured acrossthe crystal area. To a first order, thisis easily handled by working withfirst or second derivative spectra.One point to note is in comparisonwith NIR imaging, the spectra of theingredients are very distinctive,offering a greater possibility for moresimple univariate approaches to theimage analysis. For testing purposes,a common raw ingredient, povidone,known not to exist in the tablet, wasadded as a test raw ingredient toillustrate the effectiveness of thesubsequent data processing.

Testing for the Presence ofTablet Ingredients

The first step is to confirm thepresence of spectral contributionsfrom the raw ingredients in the

image spectra. This was performedusing the ‘Target’ function in theHypersoft software. The spectra inthe image are pre-processed (firstderivative, and blanked in theregions of CO2 absorption) and aprincipal components analysis(PCA) of the spectra performed toextract the major independentspectra contributions. The rawingredient or ‘target’ spectra are thenfitted to linear combinations of theprincipal components from the PCA.The closeness of fit gives an indica-tion of whether the target spectra arepresent in the image spectra. Theresults for the five components ofthe tablet are shown in Figure 4. The quality of fit for all five of theingredients was excellent whereasthe quality of fit for the povidonewas poor, reinforcing the validityof this method.

Estimating Individual IngredientDistributions in the Tablet

Having confirmed that the targetspectra are present in the imagespectra, the next step is to estimatethe relative abundance of theingredients at each pixel in theimage to generate the individual

component distribution maps.There are a number of approachesto this, but for a five componentmixture with the assumption of nochemical interaction between thecomponents and the pure rawingredient spectra available, asimple full spectrum curve fitting isfound to work well. Having exam-ined the residual image after theprincipal components analysisabove to check for any additionalchemical components present, eachpixel spectra is simply modeled as alinear combination of the rawingredient spectra. This techniquecan be less reliable if strong crosscorrelations exist between the targetspectra; that is if some of the targetspectra have many peaks in com-mon. In this example, there isclearly some correlation betweenthe spectra of microcrystallinecellulose and starch – but they arestill sufficiently different to enablea well-conditioned least-squarescurve fitting. Figure 5 shows theresults for the spectrum curvefitting for the five ingredients. Ineach image, the pixel intensityshows the relative abundance of thetarget spectra in the image spectra.

4

Figure 3. Suspected raw ingredients.

In most cases, the ‘hot’ spots in thederived images contain spectra likethose of the pure componentsspectra, with ‘mixture’ pixelsaround the edges of the particles, as expected. Typical spectra fromthe hot spots are illustrated in thefigure, along with the correspondingraw ingredient spectra. There arethree interesting points to note inthese images.

• The spectral contrast and hencesharpness is remarkable – farsharper than most NIR counter-parts. This is easily seen byplotting the histogram distribu-tions in the images.

• The shapes of the hot spots(particles) are not only differentfor the different ingredients, butare actually starting to appearsimilar to those of the raw ingre-dients. For example, scanningelectron microscope (SEM) imagesfor starch particles are shown inFigure 6a, where the starchparticles are seen to be highlyspherical, unlike many otheringredients. The correspondingderived ATR image, Figure 6b,shows particles with similarshape. Similarly, the magnesiumstearate and microcrystallinecellulose domains tend to havehighly irregular shapes not unlikethe SEM images for the rawingredient. This observation,arising from the improved fidelityof the image compared with NIR,could provide very importantadditional information in assist-ing understanding the spatialdistributions and correlating withproduct quality attributes.

• Individual particles smaller than10 microns can be discerned.Figure 7 shows an expanded areafrom the calculated distributionfor the active ingredient, known

5www.perkinelmer.com

Figure 4. Testing for presence of raw ingredients.

6

Figure 5a. Least squares fitted image for microcrystalline cellulose.

Figure 5b. Distribution for starch.

Figure 5c. Distribution for calcium phosphate.

Figure 5d. Distribution for magnesium stearate.

Figure 5e. Distribution for active (< 1%).

7www.perkinelmer.com

Figure 6. ATR particle image shapes-comparison with SEM.

Starch Granules

Pure microcrystallinecellulose SEMImage

Pure magnesiumstearate SEMimage

Calculated ATRdistribution intablet

Calculated ATRdistribution intablet.

Calculated

Shadow applied

SEM ATR Image in TabletA B

A B

A B

to be present at less than 1%concentration by weight. Thearea shows hot spots around 6-7microns in diameter. From thespectra measured at the hot spots,it is apparent that many pixelscontain almost pure active spectra.It is worth noting that separateresolution studies6 estimate thespatial resolution of the techniqueto be of the order of 3-4 micronsat ca. 1700 cm-1. Again, the

spectral clarity from singleparticles is striking.

A second example showing resolu-tion of small particle sizes is shownin Figure 8. Here, an OTC solubleanalgesic tablet was imaged. Anumber of NIR imaging examplesuse this kind of product to demon-strate the capabilities of NIR imag-ing. The major ingredients, aspirin,paracetamol and caffeine are located

as large domains many tens or evenhundreds of microns in diameter.However, the example here imagesnot only the major componentsabove, but a number of minoringredients present (a) at low levelsand (b) more importantly withdomain sizes of less than 10microns. Many excipients and APIsare introduced with particle sizes ofless than 10 microns. NIR imagingdoes not easily detect individualparticles of this size but rather,domains where these particles haveformed aggregate of much largersize. Here, the figure shows not onlythe major ingredients, but in addi-tion, individual calcium phosphateparticles contained within theaspirin/caffeine/paracetamoldistribution where the individualparticles are not seen by NIR.

Characterizing Tablet IngredientDistributions

With high quality individualingredient distributions like thosedisplayed in Figure 5, the eye cansee obvious differences betweendifferent ingredients in a tablet.However, if a more rapid, lesssubjective characterization isrequired, it is necessary to expressthe images in different manner. Oneway is to express the images ashistograms of intensities of theimage. Usually, for a given image,the pixel brightness is plottedagainst the number of pixels and theresulting histogram can be analyzedto provide useful information whichhas been related to key tabletprocess parameters. With sharp ATRimages, when an ingredient distri-bution is plotted as such, theresulting histogram usually has twodistinct modes corresponding to theparticles of interest and the remain-der Figure 9a. However, the border-line between the two zones of the

8

Figure 7a. Distribution for active (< 1%).

Figure 7b. Active ingredient particles < 10 microns.

histogram is not always distinct.Since the image is a grayscaleimage, there will be a number ofpixels which fall between ‘pure’particle pixels and ‘pure’ back-ground pixels. For pixel counting,

these pixels need to be assigned asforeground or background pixels,i.e. an intensity threshold must beset above which pixels are assignedto the particles of interest, andbelow which particles are assigned

to the remainder or background.There are some mathematicalrecipes for estimating this thresh-old, e.g. ‘maximum entropymethod’5, or it may be estimatedsubjectively. For ATR images, it has

9www.perkinelmer.com

Figure 8a. Analgesic tablet – coarse structure - aspirin. Figure 8d. Revealing minor constituents – calcium carbonate.

Figure 8b. Overlaying paracetamol distribution. Figure 8e. Exploded view of calcium carbonate image.

Figure 8c. Overlaying microcrystalline cellulose. Figure 8f. Calcium carbonate within the aspirin domains.

been found useful to use a combina-tion of information about theparticle size of raw ingredients plusiterative threshold adjustment inconjunction with examination ofunderlying pixel spectra in order tohelp establish intensity thresholds.

Figure 9b illustrates thresholding. Acomponent distribution is shown asa calculated grayscale image withintensity thresholds set at two levelsto leave the most intense imagepixels. Which image is correct? Adetailed discussion of this problemis beyond the scope of this note, butthe setting of this threshold value iskey to deriving accurate quantitativeinformation from these images, andmust be used with extreme care.However, when clear distinctparticle distributions are calculated,some of the resulting statistics, (e.g.mean particle size, distributionshape, nearest neighbor information,etc.) have been related to productquality attributes. Figure 9c showscalculated ingredient distributionsfor a given ingredient from twobatches of a product labeled ‘good’and ‘bad’. Five images were recordedfrom different samples of good andbad batches. For all the ingredientsexcept one the calculated averageparticle sizes agreed for the good andbad batches. For the remainingingredient, there was an approximatefactor of two difference in averageparticle size between good and badbatches with all images. Informationof this nature can be extremelyvaluable in helping to establishprocessing problems.

Conclusion

ATR chemical imaging has advancedthe state of the art in chemicalimaging of certain types of pharma-ceutical tablets. It provides a very

10

Figure 9a. Individual ingredient distributions as histograms.

Figure 9b. Setting A-B threshold to produce a binary image.

Figure 9c. Some metrics can be related to quality parameters.

3. Lewis, E. N.; Schoppelrei, W.;Lee, E.; Kidder, L.; in ‘ProcessAnalytical Technology’Blackwell, 2005. ISBN-10 1-4051-2103-3.

4. Canas, A.; Carter, R.; Hoult, R.; Sellors, J.; Williams, S.; ‘Spatialresolution in mid-IR ATRImaging: Measurement andMeaning’ FACCS conference,2006.

5. Rasband, W.S., ImageJ,U. S. National Institutes ofHealth, Bethesda, Maryland,USA, http://rsb.info.nih.gov/ij/,1997-2006.

6. ‘Spatial Resolution in FT-IR ATRImaging’ PerkinElmer TechnicalNote # 007641_03 (2006).

• Small sample penetration depth,so careful experimental design isrequired to ensure the image at thesurface is representative of thesample.

• Given mechanical and opticalconstraints of the system, themaximum sample area is relativelysmall (ca 0.5 mm diameter circlewith current design) comparedwith other imaging techniques.Again, design of experiment shouldtake this into consideration.

Despite these limitations, thetechnique is poised to become a verysignificant tool in the better under-standing of ingredients distributionsin pharmaceutical tablets.

REFERENCES

1. Lyon, R.; Lester, D.; Lewis, E. N.; Lee, E.; Yu, L. X.; Jefferson, E. H.;Hussain, A. S.: AAPSPharmsciTech 2002; 3 (3) article 17 (http//www.aap-spharmscitech.org).

2. Clarke, F.; Vib Spec., 34 (2004)25-35. Available online at www.sciencedirect.com.

For a complete listing of our global offices, visit www.perkinelmer.com/lasoffices

©2007 PerkinElmer, Inc. All rights reserved. The PerkinElmer logo and design are registered trademarks of PerkinElmer, Inc. Spotlight is a trademark and PerkinElmer is a registeredtrademark of PerkinElmer, Inc. or its subsidiaries, in the United States and other countries. All other trademarks not owned by PerkinElmer, Inc. or its subsidiaries that are depicted hereinare the property of their respective owners. PerkinElmer reserves the right to change this document at any time without notice and disclaims liability for editorial, pictorial or typographicalerrors.

007752_01

PerkinElmer Life and Analytical Sciences710 Bridgeport AvenueShelton, CT 06484-4794 USAPhone: (800) 762-4000 or (+1) 203-925-4602www.perkinelmer.com

large improvement in achievablespatial resolution compared withNIR chemical imaging, and sharper,generally easier to interpret images.The technique overcomes the majorlimitation of direct diffusereflectance imaging in the mid-IR,namely problems with high signalattenuation and uncontrolledspecular reflectance contributions inthe image. The technique overcomesa number of limitations of Ramanimaging such as sample overheating,potential fluorescence, limitedspatial resolution (depending onRaman hardware configuration) andrelatively long image collectiontimes.

Considering the limitations of theATR imaging technique, these fallbroadly into three areas:

• As a contact technique, once asample is pressed onto a surface itis effectively contaminated, socrystal cleaning is required beforeeach image is generated. This is,however, relatively fast unlesssome of the particles of the tablethave strong adhesive properties.