(T1130-02-09) Optical Coherence Tomography as a Real-Time Tool to Monitor Sublimation Properties of Lyophilized Cakes

Purpose: Quantitative analysis of optical coherence tomography freeze-drying microscopy (OCT-FDM) data can significantly enhance the utility of OCT imaging for characterizing microstructural evolution during pharmaceutical lyophilization processes. OCT-FDM enables non-invasive, time-lapse 3D characterization under conditions that closely replicate industrial freeze-dryer environments. This facilitates precise determination of critical temperatures associated with key structural transitions, including sublimation front progression and cake collapse. These parameters are essential for robust lyophilization cycle development and optimization. This study sought to expand the analytical value of OCT data through advanced image segmentation to extract quantitative metrics, providing deeper insights into failure mechanisms and final product characteristics. Additionally, given OCT's capability for early detection of microstructural changes, this analysis explores its potential as an at-line process analytical technology (PAT) for real-time monitoring of critical process parameters, ensuring consistent product quality during lyophilization manufacturing.

Methods: OCT images were acquired during lyophilization of three formulations: mannitol with bovine serum albumin (BSA), trehalose with BSA, and mannitol alone. Time-lapsed imaging was performed using an OCT-FDM system (Physical Sciences Inc., Andover, MA) throughout the lyophilization process. The imaging parameters included a scan width of 0.75 mm, height of 4 mm, and depth of 1.5 mm from the inner vial surface. The 3D imaging system provided lateral and axial resolution of approximately 5 μm, enabling detailed visualization of microstructural changes as temperature and pressure profiles evolved during the freeze-drying cycle. The volumetric datasets were managed and analyzed using digiM I2S cloud-based software (digiM Solution LLC, Woburn, MA). Machine learning algorithms were employed to segment three distinct phases: cake solid matrix, pore space, and vial boundaries. To accurately isolate the visible pore space within the cake from the external air surrounding the vial, a smoothed convex hull algorithm was applied to the solid phase, establishing precise boundaries for internal porosity analysis. From the segmented image volumes, quantitative measurements of pore size distribution and cake solid wall thickness were calculated to characterize the structural properties of the lyophilized matrix. Intensity surface plots and spatial vector analysis were generated to visualize microstructural evolution and track material movement during the lyophilization process. Analysis provided comprehensive quantitative data on the relationship between formulation composition and structural development during freeze-drying.

Results: A 3D time-lapse video was generated showing the change in time over the entire imaged region, instead of the typical 2D time-lapse at a particular cross section. A frame of the image data and its associated segmentation are shown in Figure 1 for each sample. Qualitatively, the sublimation front is seen traveling from left to right (top to down in the lyo cake) where the solid phase expands as temperature changes. Shrinkage occurs when the entire structure moves upwards away from the vial, but collapse is not easily seen in the 3D view. From pore size analysis, the average pore increases significantly during the sublimation phase as shown in Figure 2 for mannitol. In mannitol, the average pore size is larger in the dried state than in the frozen. However, the other two samples showed similar pore size distributions before and after the sublimation front. As for cake wall thickness, the average solid wall grows thicker during sublimation, but the resultant dried solid walls are only thicker than they were in the frozen state for the mannitol samples (Figure 2). Spatial vector videos show the displacement of the solid phase from one image volume to another, and selected frames are shown in Figure 3 to highlight sublimation, macrocollapse, and shrinkage. The sublimation process is easily identified by a front of vectors moving from left to right (top of the imaged region to the bottom). Shrinkage can be seen when all the vectors align upwards towards the inside of the cake, while macrocollapse is displayed through large vector values in localized regions. Changes in structure can be easily noticed whenever the sum of vector magnitudes exceed the normal noise level. Calculated curl and divergence also tend to increase and have higher values during large structural changes.

Conclusion: This study demonstrates that advanced image segmentation and spatial vector analysis significantly enhances OCT's value for quantitative characterization of lyophilized cake microstructure. This provides critical insights into failure modes and structural evolution during manufacturing. The observed increases in pore size and wall thickness during sublimation front progression directly impact cake resistance properties, while spatial vector analysis successfully identified early structural changes that could serve as quality indicators. These quantitative analytical capabilities elevate OCT from a qualitative imaging tool into a robust at-line and in-line PAT platform enabling real-time monitoring of critical quality attributes. Enhanced microstructural characterization facilitates deeper process understanding and may transform lyophilization from an empirical process to one guided by quantitative, real-time structural monitoring. This advancement ultimately supports improved manufacturing efficiency and enhanced product quality assurance.

Figure 1: A frame of the 3D time-lapse video of optical coherence tomography freeze-drying microscopy data and its associated segmentation where the pore space is red, cake solid is green, and glass vial as yellow.

Figure 2: The left side shows pore size and solid wall thickness analysis for the mannitol only sample. Images in the middle column are from YZ cross section of the image data volume and its corresponding segmentation to the right of it.

Figure 3: Frames from spatial vector analysis where each vector indicates movement. Examples of sublimation front, macrocollapse, and shrinkage are shown.