Practical Examples of Calibration Transfer and Method Comparison
IFPAC 2004 Robert P. Cogdill James K. Drennen, III January 15, 2004
Calibration Transfer for Imaging Spectrometers
IFPAC 2004 Robert P. Cogdill James K. Drennen, III January 15, 2004
Outline • • • • •
3
Introduction to Chemical Imaging Transferability in Chemical Imaging Calibrated Internal Standardization Results Discussion/Further Investigation
Chemical Imaging: Introduction
Š Spectral Dimensions, Inc. 2004 4
Chemical Imaging: Equipment
5
Chemical Imaging: Current Method Calculation of Reflectance Image Hypercube sample data
dark data
Reflectance Cube
Y X
= (
)
Wavelength
â˜ş Dr. Carl Anderson, 2004, All Rights Reserved 6
reference data
dark data
Chemical Imaging: Application • • • • • •
Microscopy High-throughput analysis Small volume analysis Particle sizing Crystallography Future prospects… …All in Wonder Instrument
7
Chemical Imaging: Limitations • Performance – Spatial – Spectral
• Throughput – Data acquisition – Computation
• Data Analysis – Knowledge extraction
• Transferability 8
• Development/
Continuity of Spectral Databases • Propagation of methods •Upgradeability
Chemical Imaging: Transferability • Image Transfer – Resolution – Format
• Spectroscopic Transfer – Spatial: Transferability among pixels – Temporal: Stability of pixels’ spectral response – Inter-Instrument: imaging and non-imaging transfer
9
Spatial Transferability -3
10 1.2
20 40
1.1
60
1
80
0.9
100 0.8 120 0.7
140
0.6
160 180 50
100
150
200
250
0.5
Standard Deviation (Reflectance)
9
x 10
Standard Deviation vs. Wavelength (mili-reflectance units)
8 7 6 5 4 3 2 1300
1350
1400
1450
1500
1550
Wavelength ( nm )
1600
• Image of Teflon® standard @ 1489 nm • Combination of sample and reference heterogeneity 10
1650
1700
Temporal Transferability Extracted Spectra, 1 Tablet, Imaged during 3 days 1.6
Reflectance
1.4
1.2
1
0.8
Transient errors in reference or sample images
0.6 1300
1350
1400
1450
1500
Source Variation/ Re-collection of References 1550
W a v e le n g th ( n m ) 11
1600
1650
1700
Inter-Instrument Transferability
12
Inter-Instrument Transferability 1.5
9 tablets, Imaged on same day 1.4 1.3
Reflectance
1.2 1.1 1 0.9 0.8 0.7 0.6 1300
1350
1400
1450
1500
1550
W a v e le n g th ( n m ) 13
1600
1650
1700
Chemical Imaging: Current Method
sample data
dark data
Reflectance Cube
Y X
=(
)
Wavelength
reference data
14
dark data
Calibrated Internal Standardization
sample data
Reflectance Image
=
Y
X Wavelength
15
Internal dark signal
x internal bright signal
Internal dark signal
Bias Cube
Slope Cube
+
Calibrated Internal Standardization
16
Calibration Materials 0.62
Teflon® NIR Reflectance Spectrum
0.6
Reflectance
0.58 0.56 0.54 0.52 0.5 0.48
• 2-Point calibration
0.46 1300
1400
1500
– Teflon® (diffuse reflector) – Stainless steel (specular reflector) 17
1600
1700
1800
Wavelength ( nm )
1900
2000
Internal Calibration 1 0.15
20
0.9
20
40
0.8
40
60
0.7
60
80
0.6
80
-0.05
100
0.5
100
-0.1
120
0.4
120
-0.15
140
0.3
140
160
0.2
160
180
0.1
0.1 0.05 0
-0.2 -0.25 -0.3
50
100
150
200
250
Pixel-to-pixel Slope @ 1489 nm
0
180
-0.35 50
100
150
200
Pixel-to-pixel Bias @ 1489 nm
• Light source fixed-pattern, filter aberrations, detector artifacts 18
250
External Calibration (Inter-Instrument) • Relative gain between instruments is matched during internal calibration – Pixel-level regression using “global” references
• Determine relative baseline between instruments using extracted sample spectrum – Internal standardization enables use of “local” reference (e.g.- tablet) for instrument matching calculations
19
Addition of Baseline Correction
sample data
Reflectance Image
=
Y
X Wavelength
20
Internal dark signal
x internal bright signal
Internal dark signal
Bias Cube
Slope Cube
+ + Baseline Correction
Experimental Materials • Spectral Dimensions, MatrixNIR – 1300-1681 nm, 3 nm increment (128 image planes) – 128 ms integration time, 4 co-adds – 256x320 pixels, 2x magnification, ~5.5x7.0 mm
• Brimrose Luminar – Automated tablet transport mechanism – Spectra truncated to MatrixNIR parameters
• Tablet Samples – 1 baseline correction tablet – 9 test tablets 21
Results: Spatial Transferability -3
10 1.2 20 1.1
40 60
1
80
0.9
100 0.8 120 0.7 140 0.6
160 180
50
100
150
200
250
0.5
Standard Deviation (Reflectance)
9
x 10
Standard Deviation vs. Wavelength (mili-reflectance units)
8 7 6 5 4 3 2 1300
1350
1400
1450
1500
1550
1600
Wavelength ( nm )
• Standardized image of Teflon® @ 1489 nm • Reduction in scale, intensity of heterogeneity 22
1650
1700
Results: Temporal Transferability Extracted Spectra, 1 Tablet, Imaged during 3 days 1.6
Reflectance
1.4
1.2
1
0.8
0.6 1300
1350
1400
1450
1500
1550
W a v e le n g th ( n m ) 23
1600
1650
1700
Results: Inter-Instrument Transfer 1.3
9 tablets, Imaged on same day
1.2
Reflectance
1.1 1 0.9 0.8 0.7 0.6 0.5 1300
1350
1400
1450
1500
1550
W a v e le n g th ( n m ) 24
1600
1650
1700
…with additional preprocessing 2 1.5
Arbitrary Units
1 0.5 0 -0 . 5 -1 -1 . 5
SNV Preprocessing -2 1300
1350
1400
1450
1500
1550
W a v e le n g th ( n m ) 25
1600
1650
1700
Results: Inter-Instrument Transfer • Transferability of regression models, PCA example: 2.5 2 1.5 1
PC 2
0.5 0 -0.5 -1 -1.5 -2 -2.5 -4 26
-3
-2
-1
0
PC 1
1
2
3
4
Summary (good news) • Increased ability to merge databases • Diminished need for on-going scanning of reference materials • Fewer sub-scans required for equivalent SNR • Increased robustness • External standards can be replaced (e.g.- loss, breakage, contamination)
27
Summary (not quite as good news) • May reduce active field of view (FOV) • Difficult to implement in some situations – High magnification
• Longevity of Internal Calibration is currently unknown (days? months?..) • Correction is still imperfect – Sub-optimal internal reference, calibration materials – Some nonlinearity not accounted for 28
Further Investigation • Test applicability to samples of significantly different composition • Investigate internal wavelength accuracy measurement • Application of other instrument matching procedures (PDS, Direct Orthogonalization) • Evaluate long-term stability impact on prediction • Develop routines for image transfer 29
Acknowledgements • Spectral Dimensions – www.spectraldimensions.com • Duane Mann (construction of reference stage) • Duquesne Center for Pharmaceutical Technology
– www.pharmacy.duq.edu/DCPT/home.html
30