Quantitative Imaging of Pharmaceutical Compacts
Carl Anderson, DCPT
2 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Acknowledgements Cody Peer
David Molseed
3 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Duquesne University Center for Pharmaceutical Technology (DCPT)
http:// http://www.pharmacy.duq.edu/DCPT/home.html www.pharmacy.duq.edu/DCPT/home.html
Introduction • Quantitative calibration to enhance contrast in images
4 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
– Quantitative calibration from single point spectra – Quantitative calibration from images
• Predict the concentration of regions of a compact
Description of Samples • Powder blends for this study
5 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
– Starch and salicylic acid (SA) – SA concentration from 5% to 50% (w/w) – Blended
• Single component compacts – Blends pressed into 13 mm diameter compacts – One compact for each blend
Description of Samples (cont’d)
6 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
• Two component compacts – Two blends used for each compact – Die partitioned into two halves – Different blend in each side of partition % salicylic acid – Remove partition Label Side 1 Side 2 – Press compact 5% 50% A Blend 1
Blend 2
B
10%
35%
C
15%
25%
D
20%
45%
E
5%
40%
Description of Measurements
7 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
• Single point spectrum – – – – –
Perkin-Elmer FT-NIR Reflectance measurement Spot size ~8 mm 1100 nm – 1700 nm Resolution reduced to 10 nm spacing
• Images – – – –
‘Matrix’ by Spectral Dimensions Same range and resolution (61 wavelength planes) 256 × 320 pixels/image plane Field of view 5.5 mm × 6.9 mm
Description of Data • 15 Samples – 10 homogeneous compacts – 5 heterogeneous compacts
8 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
• 30 Spectra – Front and back of each sample – Log (1/r)
• 30 Images – Front and back of each sample – Mean image spectrum (N = 81,920) – Dark subtracted Log (1/r)
Image of Compacts @ 1660 nm
9 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
50% SA Single component compact
15/25% SA Two component compact
Comparison of Spectra, 30% SA 0.80 0.70
Single point Mean Image Spectrum
0.60
log(1/r)
10 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
0.50 0.40 0.30 0.20 0.10 0.00 1100
1200
1300
1400 Wavelength (nm)
1500
1600
1700
Comparison of Spectra, 30% SA (After MSC) 7.00E-01
6.50E-01
Single Point Spectrum Mean Image Spectrum
log(1/R) (after MSC)
11 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
6.00E-01
5.50E-01
5.00E-01
4.50E-01
4.00E-01
3.50E-01 1100
1200
1300
1400 Wavelength (nm)
1500
1600
1700
Comparison of Spectra, 30% SA (After MSC and First Derivative) 0.030
Second Derivative of Log(1/R)
12 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
0.025 0.020
Single point spectrum Mean Image Spectrum Series2
0.015 0.010 0.005 0.000 -0.005 -0.010 -0.015 -0.020 1100
1200
1300
1400 Wavelength (nm)
1500
1600
1700
Comparison of Spectra Pre-treated mean image spectra
0.05
0.05
0.04
0.04
0.03
0.03
Second Derivative of Log(1/R)
Second Derivative of Log(1/R)
13 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Pre-treated single point spectra
0.02
0.01
0.00 1100
1200
1300
1400
1500
1600
1700
0.02
0.01
0 1100
-0.01
-0.01
-0.02
-0.02
Wavelength (nm)
1200
1300
1400
Wavelength (nm)
1500
1600
1700
Calibration Built from Single Point Spectra of Single Component Compacts 60
Slope Intercept R RMSEC
50
0.98 0.42 0.992 1.83
NIR Prediction
14 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
40
30
20
10
NIR Prediction y=x 0 0
10
20
30 % SA
40
50
60
NIR Prediction
15 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Prediction of Mean Image Spectra Using Single Point Model 50
RMSEP
0
5.07
40
30
20
10
0
Y y=x
10 20 30
% SA 40 50
Application of PLS Model to Images Reference Data
Single Point Spectra
PLS Model Log(1/r) Image
Wavelength
16 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Model Building
Y X
PLS Image
Enhanced Contrast Using a Quantitative Model Two component compact 15% and 25% PLS Prediction Image
Image Plane 1660 nm
1.1
1.1 Ratio 3.2
Ratio 1.3 2.3
3.2
3.2
4.3
4.3
5.4
5.4
17 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
2.3
1.1
0.18
2.3
0.19
3.2
0.20
4.3
0.21
5.4
0.22
0.24
6.5
1.1
2.3
12
18
3.2
24
4.3
30
5.4
36
6.5
42
Enhanced Contrast Using a Quantitative Model Two component compact 5% and 50% PLS Prediction Image
Image Plane 1660 nm
18 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
.1
1.1
.3
2.3
3.2
3.2
.3
4.3
5.4
5.4 1.1
0.17
2.3
3.2
4.3
5.4
6.5
0.19
0.21
0.24
0.26
0.28
1.1
9
2.3
3.2
21
34
4.3
47
5.4
60
6.5
72
PLS Calibration from Images Image Data from Single Component Compacts ...
Reference Data Model Building
Log(1/r) Image
Wavelength
19 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Mean Image Spectra
Y X
PLS Model
PLS Image
Calibration Built from Mean Image Spectra of Single Component Compacts 60
Slope Intercept R RMSEC
50
0.996 0.12 0.998 0.97
NIR Prediction
20 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
40
30
20
10
NIR Prediction y=x 0 0
10
20
30 % SA
40
50
60
Images Calculated Using Models Built from Single Point and Mean Image Spectra
21 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Two component compact 15% and 25% Prediction based on Prediction based on model from model from single point spectra mean image spectra
16
20
24
28
32
36
10
18
26
34
42
50
22 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Images Processed Using a PLS Model Developed from Mean Image Spectra Two Component Compact 10% and 35% 31%
46%
17%
19%
Two Component Compact 20% and 45%
Comparison of Histograms Derived from Two Different Models Two Component Compact 10% and 35% Prediction from Single Point Spectra
Prediction from Mean Image Spectra 3000
4500
4000 2500
3000
2000
F re q u e n c y
Frequency
23 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
3500
2500
2000
1500
1500
1000
1000 500
500
0
0
4
7
10
13
16
Predicted SA Content
19
22
25
0
2
5
8
11
14
17
Predicted SA Content
20
22
25
24 Quantitative Imaging, Carl Anderson, DCPT, IFPAC, 14 January 2004 Washington D.C.
Conclusions • PLS calibration developed from single point spectra of compacts enhanced contrast • PLS calibration developed from mean image spectra improved image quality over single point calibration