(T1330-06-35) Automated Online Agglomeration Detection Using Shadowgraph Imaging Analysis during Dissolution Testing of Ibuprofen Powder in USP 4 Flow-Through Apparatus

Purpose: Particulate dissolution profiles are affected by the presence of agglomerates during the dissolution test. The purpose of this work is to investigate the potential to develop an automated agglomeration identification method (AIM) through online real-time dissolution imaging analysis.

Methods: Two types of ibuprofen particles – unprocessed ibuprofen (as sieved (106-300 µm) and un-sieved fractions, (Ibu-Upr-S and Ibu-Upr-US, respectively)) and crystals produced through a cooling crystallization method in hexane (Ibu-Hex-S) were used. Shadowgraph imaging (SGI) was carried out on ibuprofen particles dissolving in a single cell flow-through dissolution apparatus (CE1) (Sotax AG) and involved recording and sizing of particles detected in a 11 x 7 mm² imaging window in the cylindrical section of the cell. Dissolution tests were carried out for 2 h at 37 °C in a USP 4 flow-through apparatus with 22.6 mm diameter cells and closed-loop configuration, with a 200 mL reservoir. Three test conditions A, B and C were used, as described in Table 1. Videos were captured every 2-5 minutes in the first 20 minutes and less frequently thereafter corresponding to dissolution test sampling times. A high-speed CMOS Grasshopper 3 (FLIR / Point Grey Research) GS3-U3-23S6C-C camera and a LED light source were used for the SGI, recording videos with a length of 15 s and frame rate of 10 frames/s. Matlab software (The MathWorks, Natick Massachusetts, USA) was used for video post-processing during the AIM development. The AIM was developed in four steps: 1) Manual agglomerate detection to develop a reference image dataset; 2) Testing of parameter combinations on images from dissolution of both sieved particle types and thus with a known size range and potentially different agglomeration tendencies (Ibu-Upr-S and Ibu-Hex-S, 20 mg); 3) Testing of the three best performing parameter combinations from step 2 on particles of unknown size range (Ibu-Upr-US, 100 mg); 4) External validation of best performing parameter combination from step 3 on images from Ibu-Upr-US (40, 50 mg) in new test conditions. Step 1: Videos from the first 10 minutes of the dissolution tests in condition A were selected, manually analysed and agglomerate occurrence times noted. For Ibu-Hex-S and Ibu-Upr-S, 2 videos were selected for each particle type, < 400 particle detections (n) in total in each video ≤ 50 detections above the sieved size; For Ibu-Upr-US: 4 videos were selected with n < 1000 in each video. For the external validation: all videos were analysed. Steps 2-3: The parameters investigated for automated agglomerate detection were particle size and solidity (a dimensionless parameter, expressed as %, describing the concavity of the particle). All tested parameter combinations are shown in Table 2. Mean and percentile values are flexible and relate to the relevant value for the video being analysed. When particles met both criteria from a specific combination, a signal was generated. Performance of parameter combinations was assessed based on highest agglomerate detection rate (step 2) and lowest false positive signals (step 3). When the Ibu-Upr-US signals were < 100, the percentage of false positive results was calculated. Step 4: All videos with any detections from the 2 h dissolution tests of the Ibu-Upr-US in conditions B and C were used.

Results: Manual observation of the videos from Ibu-Hex-S and Ibu-Upr-S revealed up to 30 agglomerates in each video, with many of the agglomerates present in multiple successive frames, equivalent to approximately 100-150 potential agglomerate detections. Three of the tested parameter combinations successfully captured all observedfor both particle types - above mean size/below 96% solidity (D ≥ Mean/S ≤ 96%); size above 75^th percentile/ below 96% solidity (D ≥ Pr75/ S ≤ 96%) and size above 75^th percentile/solidity below 75^th percentile (D ≥ Pr75/S ≤ Pr75). All three combinations succeeded in detecting all agglomerates in the un-sieved fraction (Ibu-Upr-S) with 5-30% false positive results (when signals < 100). The flexible combination D ≥ Pr75/S ≤ Pr75, showed fewer false positive results compared with the other two methods (5-12%, when signals < 100), hence it was chosen for external validation. During the external validation, D ≥ Pr75/S ≤ Pr75 criteria captured all agglomerates except for 3 particles < 150 µm (< Pr75 for the current dataset) (see example Figure 1). False positive results were observed when videos with n < 25 were analysed and/or an air bubble was present in the video. In cases where n = 150-170 with approximately 30 signals, the false results were < 10%.

Conclusion: Results show it is possible to develop a method to automatically detect the presence of agglomerates during dissolution testing using online image analysis. The most successful criterion for detecting agglomerates was a flexible combination of solidity and particle size. Challenges include robust method development in the presence of very large/small numbers of particles, and a combination of signal numbers, detection patterns and overall particle detections should be considered during AIM development.

Acknowledgements:

Funding: Science Foundation Ireland, co-funded under the European Regional Development Fund (Grant number 12/RC/2275_P2) and the TCD Provost’s PhD Award, generously funded through alumni donations and Trinity’s Commercial Revenue Unit.
We acknowledge the contribution of J. Yadav to this work in generating the Ibu-Hex-S particles.


Dissolution conditions used in USP 4 flow-through apparatus during the dissolution tests of Ibu-Hex-S, Ibu-Upr-S and Ibu-Upr-US.


AIM parameter combinations tested to detect agglomerates. Diameter above and solidity below the specified value. The following abbreviations were used: Pr – percentile; D – diameter; S – solidity. The three best performing combinations are highlighted in blue. Note: mean or Pr are flexible parameters calculated from the data present in each video. Specific size in µm or solidity % are absolute values.


Sample particle and agglomerate detections from two 15 s videos, external validation. Particle detections (black dots), particle detections that meet the size criteria (blue circles), particle detections that meet the solidity criteria (red triangles), signals indicating timepoints when both specified criteria have been met (purple stars at 1 (top of image)) and signals when only one criterion is met (purple stars at 0 (bottom of image)), and agglomerates observed in the video at signalled time points with the detected diameter in µm (red circle and number on the grey image).