(T1230-09-62) Enhanced Mechanistic Understanding of Drug Product Performance Via High-Resolution Imaging and AI Image Analytics

Purpose: Microstructures directly impact tablet performance and understanding these structures is critical in optimizing dissolution behavior. When exposed to humidity and heat, the interplay of domains can evolve significantly and impact the underlying microstructure and performance. Understanding these changes provides insight into tablet dissolution but demands a high-resolution technique to evaluate their role in performance and shelf life. Employing innovative microstructure imaging and artificial intelligence (AI)-based quantitative tools, this study sought to visualize and quantify tablet microstructure quality attributes in an unstressed/un-dissolved tablet and a tablet that underwent environmental stress via controlled humidity and heat exposure. This study additionally sought to track the evolution of tablet microstructure and porosity under stress conditions such that correlations with variations in dissolution could be further investigated at a microstructure level.

Methods: Cross-sections of an unstressed and stressed tablet of the same formulation were both imaged via mosaic field of view scanning electron microscopy (MFV-SEM). Energy Dispersive X-ray Spectroscopy (EDX) was then performed on several regions of the tablet MFV-SEM cross-section for phase confirmation and evaluation of ingredient distribution. Artificial intelligence (AI) and deep learning methods were utilized to classify distinct material phases in the tablet cross-sections. Quantitative image analytics were then applied to quantify microstructural critical quality attributes including porosity fraction, pore size distribution, and pore spatial distributions.

Results: MFV-SEM revealed distinct differences in the microstructure of the solid phase and micro-porosity networks in the unstressed and stressed tablets. In conjunction, EDX confirmed the identity of chemically unique solid domains. The solid phases were found to show less structural variation than that of the micro-porosity networks of the two tablets. Variations in the distinct micro-porosity networks of the samples were quantified with AI-based phase segmentation of the solid material and pore domains in both tablet cross-sections. AI image analytics further quantified these observed variations in micro-porosity. The unstressed tablet cross-section was found to have a pore fraction of 20.7% and this value decreased to 18.0% in the stressed tablet cross-sections. The unstressed tablet was observed to have a smaller and tighter pore size distribution, with a D₅₀ of 1.5µm as compared to the stressed tablet cross-section which showed a wider size distribution and an increase in D₅₀ to 2.2µm. In addition to variation in pore size, the unstressed cross-section displayed a more uniform spatial distribution as compared to the stressed sample.

Conclusion: MFV-SEM uniquely allowed for visualization at nanometer resolution on a large representative field of view millimeters in size such that the broader macroscopic understanding of the sample is not forfeited. Going beyond the unprecedented visualization offered by MFV-SEM, the application of AI-based image analytics is demonstrated here to be capable of extracting endless quantitative microstructure information including; area fractions, particle morphology, particle size and spatial distributions, and highly precise quantification of complex microporosity networks. In this study distinct microstructure changes resulting from storage under stress conditions were visualized and quantified with cutting edge imaging and analysis techniques. This allowed for further investigation into correlations of the microstructure changes with stress conditions and downstream tablet performance and informing manufacturability, stability attributes, and packaging control strategies. This characterization technique can be broadly applied to formulation development, quality control and assurance, microstructure bioequivalence, forensic analysis, reverse engineering, stability assessment, and regulatory support.

Figure 1. Imaging and spectroscopy-based tablet analysis workflow

Figure 2.: (a,b) MFV-SEM Grayscale Images of an unstressed and stressed tablet, respectively. (c,d) Corresponding images with AI-based segmentation overlay. (e,f) EDS elemental map of regions of interest for each tablet. (g) Image-based cumulative pore size distribution plot. (h) Image-based pore spatial distribution along the X axis of the tablet images. (i) Image-based pore spatial distribution along the Y axis of the tablet images.