Anderson_IDRCv01

Multivariate Image Analysis Processing (Blending Analysis and Coated Beads)

Carl Anderson, Ph.D.James K. Drennen, III, Ph.D.

Robert P. CogdillHua Ma

Roxana MoldovanDuquesne University Center for Pharmaceutical Technology

2 Carl A Anderson, IDRC 9 August 2004, PA

Duquesne University Center for Pharmaceutical Technology (DCPT)


OutlineBlending data

ExperimentData pre-treatmentSupervised discriminate PLS modelAssessment of models used for image enhancement

Coating of tablets and beadsMasking of imagesQuantitative predictionsDistribution of coating on samplesMultivariate image analysis without chemical testing


Description of Blending Experiment

Salicylic Acid (SA) and Lactose 10% SA (w/w)(particle size)

8L Bohle tumble blenderBin modified with windows

Images at both windowscollected every 10 rotations


Modified Matrix NIR Imager

x-axistranslation stage

Blender vessel

Lights

z-axistranslation stage

Camera


Images from BlenderLarge window

Diameter ≈ 33 cm27 images

Small windowDiameter ≈ 5 cm7 images

Data for each rotation(small window only)

7 images256 × 310 pixels(~56 µm resolution)61 wavelengths (1400 – 1700 nm)63 MB/Rotation


Inner diameter of blender

Individual images

Concatenated Single Plane Image(1660 nm)

5 mm

Log (1/r)


Inner diameter of blender


5 mm

Log (1/r)



5 mm

Log (1/r)


0.6

0.5

0.4

0.3

0.2

0.10 500 1000 1500 2000 2500

Row Number

Row

Mea

n (L

og(1

/r))

Mean Log(1/r) for Rows

Blue Line: Mean value across each row in the concatenated image before bias correctionRed Line: Mean value across each row in the concatenated image after bias correction.


Before Bias Correction (After Preprocessing)

After Bias Correction (After Preprocessing)0.

5

0.4

Log(1/r)

Bias Correction on a Single Plane Image (1660 nm)


Typical Bias Correction Between Two Individual Images


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Log(1/r)ScaleFor

Images

10 20 30 6050 70 80 90 100 11040 Rotation

Single Plane Images at 1660 nm

5.5 mm


Processing of Images for Rotation 40(Discriminate PLS Analysis)

Single Wavelength (Plane) 1660 nm

First Iteration of PLS Analysis

Second Iteration of PLS Analysis


Middle of Rotation 40

1 mm

PLSPrediction


10 20 30 6050 70 80 90 100 11040

1

-1

0

1.5

-1.5

-0.5

0.5

Rotation

Predicted Images for Rotations 10-110 using PLS Discriminate Analysis

5.5 mm


Assessment of Images by Qres and Hotelling’s T2

Calculate Qres and T2 for each pixelQres indicates unmodeled residualT2 indicates distance from center of model


10 20 30 6050 70 80 90 100 11040 Rotation

Q Residual Plot for Rotations 10 – 110


10 20 30 6050 70 80 90 100 11040 Rotation

Hotelling’s T2 Plot for Rotations 10-110


NIR Imaging Analysis of Polymer-Coated Granules and Tablets

Rotogranulated Beads &Extruded/Spheronized Beads

EthocelTablets

Sure-releaseEudragit


Single Plane Image (1566 nm) of Extruded/SpheronizedPolymer-Coated Beads (1x Objective)

1 mm


Single Plane Image (1566 nm) of Extruded/SpheronizedPolymer-Coated Beads (10x Objective)

100 µm


Interactive Masking Function Flow Diagram

Open Image Plane

Select (n > 10) “keep” Pixels

Calculate 2 PCA Scores

Choose Operation

Select “keep”Points

Select “omit”Points

Display Masked Image

DONE

Threshold Threshold


PCA Scores for Beads on Holder


PCA Enhanced Image of Extruded/SpheronizedPolymer-Coated Beads (1x Objective)

1 mm


Typical Digital Mask

1 mm

White pixels indicate retained features.

Black pixels are omitted from

analysis


Typical Digital Mask Following Optimization

1 mm

White pixels indicate retained features.

Black pixels are omitted from

analysis


Functions of Masking

Removes spectra of sample holder from data set

Eliminates meaningless dataAllows the analyst to proceed with chemometric analyses of data resulting from the sample


Rotogranulated Bead Spectra

1 1 0 0 1 2 0 0 1 3 0 0 1 4 0 0 1 5 0 0 1 6 0 0 1 7 0 0-0 . 1

-0 . 0 5

0

0 . 0 5

0 . 1

0 . 1 5

0 . 2

0 . 2 5

0 . 3

0 . 3 5

W a v e le n g th ( n m )

Log

(1 /

R)

Direction of IncreasingCoating Concentration


Combined (Rotogranulated + Extruded/Spheronized) Calibration Correlation Plot, Derived from Image Spectra

-5 0 5 10 15 20 25-5

0

5

10

15

20

25

Measured Coating ( % )

Pre

dict

ed C

oatin

g ( %

)


Combined (Rotogranulated + Extruded/Spheronized) PLS Model Basis Vector

1 2 0 0 1 3 0 0 1 4 0 0 1 5 0 0 1 6 0 0 1 7 0 0-0 . 2

-0 . 1 5

-0 . 1

-0 . 0 5

0

0 . 0 5

0 . 1

0 . 1 5

0 . 2

0 . 2 5

0 . 3

W a ve le n g t h ( n m )

Arb

itrar

y U

nits


Multi-Particulate Images of Predicted Coating Concentration for Extruded/Spheronized Beads

2 0 0 4 0 0 6 0 0 8 0 0 1 0 0 0 1 2 0 0 1 4 0 0 1 6 0 0

0% 6% 8% 10% 12% 14% 16% 18%

-5

0

5

10

15

20

25

Unknown Coating (sub-coated), 1x

Coating %

0

5

1 0

1 5

2 0

2 5

3 0

3 5

4 0

Unknown Coating

18% Coating

10x Single-Particle Images


Multi-Particulate Images of Predicted Coating Concentration for Rotogranulated Beads

0% 8% 10% 12% 14% 16% 18% 20%

8% Coating 14% Coating

10x Single-Particle Images


Histogram of predicted coating concentration using particle meanspectra from 2x images of extruded/spheronized beads

-5 0 5 1 0 1 5 2 0 2 5 3 00

2

4

6

8

1 0

1 2

1 4

C o a t in g ( % )

Freq

uenc

y ( n

)

Red(1) – UncoatedGreen – 6%

Yellow(1) – 8%Purple – 10%

Blue – 12%Black – 14%

Red(2) – 16%Yellow(2) – 18%


Histogram of predicted coating concentration using particle meanspectra from 2x images of rotogranulated beads

-5 0 5 1 0 1 5 2 0 2 50

2

4

6

8

1 0

1 2

1 4

1 6

1 8

C o a t in g ( % )

Freq

uenc

y ( n

)

Red(1) – UncoatedGreen – 6%

Yellow(1) – 8%Purple – 10%

Blue – 12%Black – 14%

Red(2) – 16%Yellow(2) – 18%


Coated-Tablet Images of Predicted Eudragit Coating (Time) Using 1x Objective

1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0- 2 0

- 1 0

0

1 0

2 0

3 0

4 0

5 0

6 0

0 Minutes 20 Minutes 40 Minutes 44 Minutes


Coated-Tablet Images of Predicted Eudragit Coating (Time) Using 10x Objective

0 Minutes 20 Minutes 40 Minutes 44 Minutes


Process Understanding UsingNIR Imaging

BlendingBlend behavior

On what scale is homogeneity reached?

Process samplingWhere is the optimal place or places for sensors?

Coating processesDistribution of coating levels on beadsEstimation of complete coating time from images without wet chemical data (process data only)


Acknowledgement

David MolseedDavid Wargo of Mylan Pharmaceuticals


Thank You

Anderson_IDRCv01

Documents

nm12 carl

mbrotation7 carl

mmplsprediction16 carl

mm17 carl

mm14 carl

rotations5 carl

center of model18 carl

coated beadscarl anderson