Professional Documents
Culture Documents
ABSTRACT: This study investigated the effects of various process variables on some of
the shape factors most widely used in the morphological characterization of pharmaceu-
tical particles (including circularity, aspect ratio, er, and the recently proposed Vr and
Vp). After optimization of the illumination system and greyscale thresholds for
discrimination of particle background, we evaluated the effects of process variables
within the image capture and analysis system, including the image capture device (video
or digital camera), image color information (24-bits-per-pixel RGB or 8-bits-per-pixel
black and white), file type (JPG or TIFF), and JPG file compression ratio. A key aspect was
evaluation of the effect of scaling factor (mm/pixel), dependent on real pixel size and
optical magnification, on shape factor estimates. The results obtained indicate that
accurate estimation of shape factors requires use of a scaling factor below a certain
maximum; however, use of very low scaling factors will mean that the field of view is very
small, so that it will contain very few particles, implying a time-consuming increase in the
number of images that must be analyzed. Finally, we use statistical procedures to
estimate the minimum number of particles that must be analyzed in order to provide
accurate estimates of shape factors. ß 2005 Wiley-Liss, Inc. and the American Pharmacists
Association J Pharm Sci 95:348–357, 2006
Keywords: image analysis; granulation; shape analysis; extrusion; spheronization;
pixel size; morphology; pellets
studies in pharmaceutical technology to point out Turbula T2 mixer for 15 min, then kneaded with
that IA systems need to be properly validated for appropriate quantities of water in a Kenwood
effective use. This article centered largely on the Chef Classic orbital mixer for 15 min; then extru-
validation of aspects related to light microscopy ded in a Caleva 10 extruder (6 rpm, hole diameter
components of IA systems. Subsequently, in 1995, 1 mm); then spheronized in a Caleva 120 sphe-
Podczeck et al.8 noted that important sources of ronizer (2000 rpm, 15 min); then dried to constant
variation in the morphological characterization of weight in a Heraeus oven.
particles by IA are illumination intensity and Images were obtained with an Olympus SZ-
direction, greyscale thresholding choices, and CTV stereomicroscope with different image-cap-
pixel size (minimum pixel size should be deter- ture devices attached: (1) a JVC TK-S350 black-
mined in view of the optical magnification used). and-white video camera (1/3-inch interline-trans-
Over the last decade, however, there have been fer CCD with effective pixel number of 430000)
dramatic advances in digital photography and in connected to a Matrox Comet video card via PC
computing systems for image treatment and Image VGA 24 software, 92) an Olympus DP11
management. These advances have opened up microscope camera system, or (3) an Olympus
new and interesting applications for IA techni- DP12 microscope camera system. Samples were
ques, and have led to clear improvements in the illuminated perpendicularly from above, against a
quality of the results obtained. At the same time, black background, as recommended by Podczeck
however, there has been an increase in the number and Newton [1995],10 using a Highlight 2000 cold
of potential sources of variation. For example, light (Olympus Europe). A total of 50 particles
modern digital cameras can obtain information from the spherical-pellet batches and 50 particles
with different levels of resolution, and this infor- from the irregular-pellet batches were studied. To
mation can be stored in various file types each with avoid variations associated with the position of
different data compression characteristics, and each particle in the field of view, each particle was
consequent effects on image characteristics. glued to the black background.
Furthermore, depending on the type of camera Shape factor values were calculated from the
used and above all the size and type of the sensor particle measurement data using the SigmaScan
used for image capture (CMOS or CDC), pixel size package and various purpose-written macros in
and quality varies markedly. Microsoft Excel XP.
In view of these considerations, the principal To investigate the influence of file compression
aim of the present study was to identify the most method on shape factors, 20 spherical particles
important sources of variation in a number of were measured in sextuplicate using the Olympus
parameters commonly used in the morphological DP12 camera system, which allows the images to
characterization of pharmaceutical particles. Spe- be stored in either TIFF or JPG format, in the
cifically, we considered two widely used shape latter case with varying compression ratio.
factors, circularity and aspect ratio, as well as er as Shape factors were compared among treat-
defined by Podczeck and Newton [1994] and ments by two-way analysis of variance following
Podczeck et al. [1999]8,9 and the parameters Vr confirmation of data normality and variance
and Vp recently proposed by Almeida et al [2004].11 homogeneity. Means were subsequently compared
These parameters are all widely used in proce- by the Student-Newman-Keuls test. All statistical
dures using recently developed image capture and analyses were performed using the SigmaSat
image management technologies. 1.0 statistics package.
The second important aspect of sample illumi- background will be weak, and identification of the
nation is the intensity of the light source. Figure 1 particle edge will be strongly dependent on the
summarizes the procedure used to determine greyscale threshold choice (or equivalent color-
optimal illumination intensity in the present mapping criteria). In contrast, if the light source is
study, taking as an example image capture using too strong, the excessive brightness will hinder
the JVC TK-S350 video camera. As can be seen, the focusing and reduce edge definition. This effect on
importance of illumination intensity depended on Vp is basically attributable to the fact that high
the shape factor considered. Thus Vr, AR, and er light intensities hindered effective measurement
were scarcely affected by illumination intensity, of perimeter but did not markedly affect measure-
whereas Vp increased markedly at light intensi- ments of other parameters (radius, length, width,
ties above 3 (on the arbitrary scale used by the area). On the basis of these results, we conclude
Highlight Olympus cold light). If the light source that for our system the best compromise between
used is too faint, the contrast between particle and brightness is obtained at about light intensity 3,
independently of particle shape.
A Effects of Thresholding
12 Vr
Vp
For measurement of the particle dimensions used
er
10 in morphological analysis, analog images (as
AR
obtained with the video camera in the present
study) must first be converted into a binary image
Vr, Vp, V, er, AR
8
with one bit per pixel. The software must then
6
count the number of pixels making up each
dimension (area, perimeter, etc.) before multi-
plying by the corresponding scaling factor. When
4
the image captured is already digital (i.e., digital
camera or scanner), it must be converted to a bit
2
map comprising a series of pixels whose size will
vary depending on resolution and optical magni-
0 fication. Each of the pixels has an associated color
2.0 2.5 3.0 3.5 4.0 4.5 5.0
value, depending on the color mode being used
Light intensity
(black and white, greyscale, red-green-blue, etc.).
For measurement of particle dimensions, the
image must be converted to a binary image,
B 30
requiring selection of threshold values for con-
Vr
Vp verting greyscale or color images to black-and-
25 er white images with one bit per pixel.
AR Figure 2 shows the results of using different
20
threshold values for analysis of spherical (A) and
irregular particles (B). As with light intensity
Vr, Vp, V, er, AR
Table 1. Basic Characteristics of the Images and Image Files Obtained by the Three Cameras Used in the Present
Study
manage a large number of pixels, which greatly tions used (except for the lowest resolution of the
increases the time required for the analysis. Thus DP12 camera, 640*480 when used with the 0.5
it is desirable to determine the minimum resolu- attachment) gave pixel size after scaling lower
tion at which particle dimensions can be accu- than the maximum of 30 mm recommended by
rately measured. Podczeck et al. [1999]9 for the characterization of
To study and compare the effects of image pellets of around 1 mm diameter.
resolution and image file type on the different Figure 3 shows how number of pixels making up
parameters available for characterization of pellet the perimeter of the particle projection declines
morphology, we obtained images of 50 pellets exponentially with increasing pixel size. Bearing
using two models of Olympus digital camera, with in mind the results obtained by Almeida et al.
different image capture resolutions, different file [2004],11 good precision requires that the radius
formats (TIFF or JPG), and different levels of JPG of the particle should correspond to at last 25–
data compression (see Table 1). In addition, we 30 pixels, or equivalently that the perimeter
obtained images with a wide-angle attachment should be made up of more than 150 pixels. This
(0.5) that increases the field of view and thus the suggests that we must reject resolutions with a
number of particles included in each image. scaling factor of less than 20 mm per pixel.
As can be seen from Table 1, with the three To evaluate this possibility, we have plotted
systems four pixel sizes can be obtained, depend- various commonly used shape factors against
ing on the resolution of the system used. The DP12 number of pixels making up the perimeter
digital camera has the lowest pixel density (Fig. 4). As can be seen from these plots, all shape
(72 72 ppi) but the highest resolution (i.e., factors except AR were affected to greater or lesser
greatest total number of pixels (2,048 1,536)), extent by number of pixels making up the
thus giving the smallest pixel size. The video perimeter, though particle shape (spherical or
camera has a lower resolution (704 420 pixels) irregular) was also important. With spherical
but a high pixel density (176 105 ppi), and so particles, the shape factors Vr and er showed a
pixel size is lower than in most of the digital rapid decline with increasing number of pixels
camera resolutions. In any case, all of the resolu- making up perimeter, until about 150–200 pixels,
550
after which the decline was less marked. Circu-
500 A, spherical particles larity likewise declined with increasing number of
B, irregular particles
pixels making up perimeter, while Vp increased.
Number of pixels in perimeter
450
400
An increase in the number of pixels making up
the perimeter (due to smaller pixel size) gives a
350
larger number of effective radii for calculation of
300
Vr, and allows better definition of the surface
250 roughness of the particles, thus causing a reduc-
200 tion in the values of those shape factors that are
150 sensitive to surface roughness (circularity and er),
100
and increased Vp. In view of the results obtained,
we can confirm that the use of scaling factors lower
50
5 10 15 20 25 30 35 40 than 20 mm/pixel should be sufficient for precise
Scaling factor (µm/píxel) estimation of most of the shape factors considered
here. This indicates that the DP11 and DP12 digi-
Figure 3. Number of pixels making up the perimeter tal cameras, with 1 optical magnification, can be
of the particle projection, plotted against the scaling used at any resolution except the lowest (640*512
factor used (mm/pixel). A, spherical particles; B, irregu-
and 640*480 in DP11 and DP12 respectively),
lar particles.
11 0.88
15
Circularity
10
0.86
Vp
Vr
0.80
7
5
0.78
0.76
6
4
0.74
100 200 300 400 500 100 200 300 400 500 100 200 300 400 500
No. of pixels in perimeter No. of pixels in perimeter No. of pixels in perimeter
1.58 0.54
1.56
0.52
AR
er
0.25
1.11
0.20
1.10
100 200 300 400 500 100 200 300 400 500
No. of pixels in perimeter No. of pixels in perimeter
although use of the 0.5 attachment on DP12 they may become apparent at higher compression
would require the highest resolution (2048*1536). ratios. In any case, color information is not
necessary for characterization of particle shape,
so that a priori the JPG compression algorithm
File Format and Compression Ratio
would not be expected to affect shape factor
Another important aspect to be taken into account estimates, except at high compression ratios. The
is file format. In the case of the system based on results obtained are summarized in Table 2.
the video camera, this aspect is not so critical, Analysis of variance indicated significant differ-
since it is not necessary to save the images before ences in Vr (a < 0.01) between the 2048*1536 and
analysis; rather, the signal can simply be frozen 1024*768 resolutions, but no significant effects of
and the particles measured directly from the file format or JPG compression ratio. In the case of
frozen image. With the digital cameras, however, Vp, er, and circularity, however, not only resolu-
it is necessary to store the images Under these tion but also file format and JPG compression ratio
conditions image size is important, since it affects had significant effects. Finally, AR was not
download time, storage space requirement and significantly affected by any of these factors.
subsequent image treatment. To identify the causes of these effects, we
To empirically investigate the effects of file performed a similar study of the parameters
format on shape factor values, we obtained several directly measured by the IA program, namely
identical photographs of a single field of view area, perimeter and mean radius. Analysis of
containing 20 particles, using different resolutions variance indicated that both area and mean radius
and file formats. In each case sextuplicate images are basically independent of resolution, file format
were obtained. This part of the study was carried and compression ratio; however, perimeter was
out using the Olympus DP12 camera, which allows significantly affected (a < 0.05) by both file format
various resolutions, storage in TIFF or JPG and JPG compression ratio. This explains why Vp,
format, and various compression ratios in JPG er, and circularity (which are based on the
format (see Table 1). TIFF (Tagged Image File perimeter measurement) are affected by these
Format ) is a high-quality uncompressed format factors. The effects of file format and JPG compres-
offering up to 64-bit color, and thus basically sion ratio on perimeter are related to the way in
complete image information; the disadvantage is which the IA program identifies the pixels that
that the files may be very large (up to 4 GB). JPG or make up the perimeter. Under the illumination
JPEG (Joint Photographic Experts Group) takes and thresholding conditions used here, an exces-
into account the imperfections of the human visual sive increase in resolution or in amount of
system as regards sensing of certain colors and information stored greatly hinders the process of
other information, and eliminates this information identification of the perimetral pixels, because an
to reduce file size. Notably, the JPG compression increase in image information (i.e., more pixels
algorithm eliminates minor color differences with more color and/or intensity information)
between adjacent pixels; at low compression ratios means that the transition between particle projec-
these changes are not visually detectable, though tion and background is more gradual. This means
Table 2. Mean Radius and Shape Factor Values Obtained in Analysis of Particle Batches Using Different File
Formats (TIFF or JPEG) and Different JPEG Compression Ratios (Low or High)
Resolution-File-Compression
in turn that, as can be seen from Table 2, the information, that is 24-bit-per-pixel RGB format
number of particles for which the perimeter can be or 8-bit-per-pixel greyscale. A priori, and bearing
completely identified declines markedly with in mind that conversion to greyscale maintains all
increasing image quality. To attempt to resolve color and intensity information, with only color
this problem it would be necessary to re-optimize information lost, we would expect only slight
illumination and thresholding for high-resolution effects.
images. To empirically investigate the magnitude of
In fact, the effects of file format and JPG these effects, we again used the Olympus DP12
compression ratio, although statistically signifi- camera, at three resolutions (640*480, 1024*768,
cant for some shape factors, were of scant practical and 2048*1536), obtaining photographs of 50
relevance: the modifications were in all cases of spherical particles and 50 irregular particles in
less than 5% (maximum 1.3%, 4.0%, 0.4%, and both color and greyscale. The results obtained
2.9% for mean radius, er, Vr, and Vp, respectively). (Table 3) confirm that loss of color information has
As noted, however, the number of particles for minimal impact of shape factor values.
which the perimeter is correctly identified dropped
markedly at higher resolutions, making it neces-
Repeatability and Precision
sary to have a larger total sample size, and
requiring operator intervention to ensure that Having identified the illumination and image-
only particles with correct perimeter identification processing factors with most marked effects on
are being included: thus we consider that it is shape factor values, we next proceeded to evalu-
better to use lower-resolution images. ate method repeatability and precision. First, we
evaluated variability among particle batches
characterized on different days. To this end, we
Color Versus Black-and-White Images
randomly selected 10 particles from each batch of
Another relevant aspect is whether parameter particles used in the study. Images were obtained
estimation is significantly affected by color daily for 3 days with a video camera (in this
Table 3. Influence of Color Information (Color or Black-and-white) and Image Resolution on Estimates of Mean
Radius and the Various Shape Factors
Spherical Particles
Color B and W
Resolution Resolution
640 480 1,024 768 2,048 1536 640 480 1,024 768 2,048 1,536
Number of 77.55 0.33 115.44 0.06 225.36 3.05 75.10 0.14 112.72 0.14 224.99 1.65
pixels in
perimeter
Mean radius 546.39 0.34 549.29 0.56 531.24 0.75 530.81 0.24 536.00 0.11 525.34 1.78
(mm)
Vr 5.71 0.05 4.75 0.06 4.32 0.08 5.52 0.05 4.95 0.05 4.34 0.01
Vp 6.10 0.17 5.98 0.08 7.09 0.22 5.52 0.29 6.05 0.10 6.55 0.11
er 0.57 0.004 0.55 0.001 0.55 0.002 0.57 0.005 0.56 0.001 0.55 0.001
Irregular particles
Number of 90.01 0.32 134.07 0.16 266.53 2.91 89.65 0.58 133.93 0.24 270.09 1.34
pixels in
perimeter
Mean radius 607.10 0.04 611.41 0.31 595.38 2.83 604.87 1.69 609.42 0.27 601.06 2.39
(mm)
Vr 15.08 0.004 14.89 0.001 15.34 0.22 15.23 0.13 14.99 0.01 14.97 0.24
Vp 9.99 0.007 10.62 0.14 11.33 0.21 10.09 0.06 10.59 0.20 11.45 0.28
er 0.28 0.001 0.25 0.001 0.24 0.009 0.26 0.008 0.25 0.002 0.24 0.008
Values shown are means standard deviations.
8. Podczeck F, Newton MJ. 1994. A shape factor to assessment of the sphericity and surface roughness
characterize the quality of spheroids. J Pharm of pellets. Int J Pharm 124:253–259.
Pharmacol 46:82–85. 11. Almeida-Prieto S, Blanco-Méndez J, Otero-Espinar
9. Podczeck F, Rahman SH, Newton MJ. 1999. FJ. 2004. Image analysis of the shape of granulated
Evaluation of a standardised procedure to assess powder grains. J Pharm Sci 93:621–634.
the shape of pellets using image analysis. Int J 12. Walpole RE, Myers RH, Myers SL. 1998. One and
Pharm 19:123–138. two sample test of hypotheses. In: Heath A, editors.
10. Podczeck F, Newton MJ. 1995. The evaluation of a Probability and stastatistics. 6th Ed., New Jersey:
three-dimensional shape factor for the quantitative Prentice Hall Inc. pp 290–357.