Artificial intelligence-aided endoscopic in-line particle size analysis during the pellet layering process
Abstract
In this study, an artificial intelligence-based machine vision system was developed for in-line particle size analysis during the pellet layering process. Drug-layered pellets were produced by coating microcrystalline cellulose cores with an ibuprofen-containing layering liquid until the target drug content was achieved. Drug content increases with pellet size; therefore, particle size monitoring can ensure product safety and quality. The direct imaging system, consisting of a rigid endoscope, a light source, and a high-speed camera, provides real-time information about pellet size and layer uniformity, enabling timely intervention in the case of out-of-spec products. A convolutional neural network-based instance segmentation algorithm was employed to detect particles in focus, ensuring that pellet size could be accurately determined despite the dense flow of the particles. After training the model, the performance of the developed system was assessed by analysing the particle size distribution of pellet cores with variable sizes within the 250–850 μm size range. The endoscopic system was tested in-line at a larger scale during the drug layering of inert pellet cores. The particle size data acquired in real time with the endoscopic imaging system corresponded with the reference methods, demonstrating the feasibility of the proposed machine vision-based method as a process analytical technology tool for in-line process monitoring.
Highlights
- AI-aided endoscopic imaging system was developed for in-line particle size analysis
- CNN-based instance segmentation algorithm provided detailed particle outline
- In-time monitoring of pellet size distribution and shape during the layering process
- Enhanced understanding of the particle size increase dynamics
Introduction
Pellets are small free-flowing agglomerates with a near spherical or cylindrical shape, utilised across various industries due to their narrow particle size distribution and uniform shape, which provide advantages during processing and handling. Applications include animal feeds, fertilizers, catalyst carriers, iron ores, plastics and pharmaceuticals [1]. In the pharmaceutical industry, pellets are spherical particles with a diameter between 0.2 to 2.0 mm, commonly used in the production of oral dosage forms such as capsules and tablets [2]. Pellets offer several advantages, including enhanced flowability, reduced particle segregation, mitigation of excessive dust, improved distribution within the gastrointestinal tract, reduced risk of dose dumping of modified release dosage forms, and enabling the simultaneous administration of incompatible substances [3], [4]. Pharmaceutical pellets are manufactured by various techniques, such as extrusion-spheronization, layering, cryopelletization, freeze pelletization, and hot-melt extrusion [5], [6]. Drug-layered pellets are produced by layering of inert pellet cores with active pharmaceutical ingredient (API) until the desired dosage content is achieved. The API content in these pellets is determined by particle size, with variations in drug load corresponding to changes in layer thickness. The fluid bed apparatus is widely utilised in the pharmaceutical industry for drug layering, as it efficiently integrates multiple process steps—preheating, spraying-layering, and drying—into a single piece of equipment[1], [7], [8]. This technique is also suitable for coating the pellets with rate-controlling film forming polymers to achieve modified-release (MR) profiles [2], [9], [10].
The critical quality attributes (CQAs) of pellets—such as particle size, particle size distribution, particle shape, density, surface area, and moisture content—play a pivotal role in determining the overall performance of the final product. Variations in particle size can cause differences in the surface area available for coating, which may influence the amount of coating material applied. Consequently, a broad particle size distribution could lead to varying coating thickness, potentially affecting the drug content of drug-layered pellets and altering the drug release profile of MR formulations [11], [12], [13], [14]. Additionally, particle morphology plays a crucial role in the flowability of the sample; pronounced surface roughness in pellets can hinder capsule filling [15], [16]. When pellets with non-uniform surfaces are coated, the rate-controlling layers may become uneven, potentially compromising drug release. Therefore it is crucial to monitor these critical quality attributes to ensure the desired product quality.
Various critical process parameters (CPP) influence the aforementioned CQAs, including inlet air temperature, product temperature, relative humidity, spray rate, and atomization air pressure. Elevated air temperatures can cause sprayed droplets to dry too quickly, leading to uneven distribution on the pellet surface and resulting in rough, porous coatings. Conversely, low product temperatures can result in unwanted particle agglomeration because the drug layers remain wet, causing the pellets to stick together. The spray rate influences both the droplet size of the wetting liquid and its distribution uniformity within the solid mass, with higher rates resulting in larger pellets and a broader particle size distribution [17], [18], [19], [20], [21].
Even minor deviations in the CPPs can significantly impact product quality, potentially leading to batch rejects and recalls. Real-time monitoring of the CQAs enables the timely adjustment of the CPPs, ensuring that the CQAs remain within the specified target range. Traditionally, CQAs are assessed at the end of the fluid-bed process by taking a sample from the apparatus for off-line analysis; however, these methods are destructive, time-consuming and do not allow for real-time feedback control [22]. A recent shift in pharmaceutical manufacturing has led to the adoption of the Process Analytical Technology (PAT) initiative, which aims to improve product safety and quality by monitoring the CQAs and CPPs with in-process sensors, coupled with the corresponding data analysis methods and control strategy [23], [24]. Many PAT tools have demonstrated the capability of monitoring the pellet layering and coating process, including near-infrared (NIR) spectroscopy [25], [26], [27], Raman spectroscopy [28], [29], spatial filter velocimetry (SFV) [30], [31], [32], focused beam reflectance measurement (FBRM) [33], photometric stereo imaging [34], Terahertz pulsed imaging [35] and acoustic emission spectroscopy [36]. Machine vision stands out as an extremely versatile PAT tool for real-time analysis. Digital imaging has seen a tremendous progress in the last decade, driven by the development of high frame-rate cameras capable of real-time data acquisition and the increased efficiency of graphical processing units (GPU) that reduce processing time [37]. Literature describes numerous machine vision-based methods to monitor the pellet layering and coating process. Digital imaging systems consisting of an industrial camera, light source and lens were employed for pellet size and shape analysis. Možina et al. [38] implemented a digital imaging system on-line using a vibratory feeder, which minimized particle overlap and ensured continuous particle flow. For in-line application, Kadunc et al. [39] positioned a camera-based imaging system externally on the observation window of the fluid-bed apparatus. In order to extract quantitative information from the captured pictures, the authors employed image segmentation to separate the pellets from the background. The performance of the aforementioned image analysis method declines with dense material flow due to the extensive particle overlap. Furthermore, the changing image background and out-of-focus particles often make segmentation more difficult [40].
Convolutional neural networks (CNN) have revolutionized image-based object recognition. CNNs are a class of deep learning models designed to process and analyse visual data through multiple layers of convolutional operations. Object detection algorithms classify and locate objects within an image using bounding boxes, while instance segmentation outlines the exact contours of each object [41], [42]. Hosseinzadeh et al. [43] developed an off-line camera-based imaging system to investigate the shape and size of iron oxide pellets, utilising a mask region-based convolutional neural network (Mask R-CNN) algorithm to detect particles. Mehle et al. [44], [45] have mounted an in-line camera-based visual inspection system on the observation window of the fluid-bed apparatus and a CNN-based instance segmentation algorithm was used to detect pellet agglomeration during pharmaceutical pellet coating.
However, since not all fluid-bed apparatuses are equipped with observation windows, integrating these systems across different setups remains a challenge. Endoscopes can be easily integrated and automated with existing laboratory hardware and software, and have been utilised for in-line monitoring of pharmaceutical processes. Simon et al. [46] presented the first endoscopy-based particle imaging system for the crystallization process, where video data was analysed using the mean gray intensity method and digital image processing to detect the initial formation of crystals during nucleation. Our previous work demonstrated the feasibility of an AI-based endoscopic imaging system for the particle size measurement of granules [47]. An endoscope can be easily installed through a sampling port without requiring modifications to the apparatus. The imaging system was tested under laboratory conditions using a 3D-printed device that simulated particle flow during fluidization. Bounding box-based object detection was utilised to identify the granules in focus, which limited the system’s ability to investigate particle morphology.
In the current study, a machine vision based PAT system was developed for the in-line monitoring of a pellet layering process. A convolutional neural network-based instance segmentation algorithm was employed to detect the outline of pellets, allowing a more precise particle size measurement and accurate determination of particle shape. The endoscopic system was integrated in-line into a fluid-bed apparatus through the sampling port with no modification of the equipment. This application marks a significant scale-up from our previous proof-of-concept experiments, where testing was conducted during particle fluidization in a small 3D printed apparatus. In the current experiments, the drug containing coating liquid was sprayed on the pellets, thus the particle size and drug content increased throughout the pellet layering process. The performance of the AI-based technique was evaluated by comparing the particle size obtained in real-time with off-line reference measurements. Additionally, the dynamics of particle growth during the pellet layering process were assessed. To the authors’ knowledge, this is the first in-line particle size and shape measurement method employing an artificial intelligence-based endoscopic system during the pellet layering process.
Download the full article as PDF here: Artificial intelligence-aided endoscopic in-line particle size analysis during the pellet layering process
or read it here
Materials
Sugar spheres (various sizes within the 250–850 μm range, pharm-a-spheres®, H.G.Werner GmbH, Kiel, Germany) and MCC spheres (355–500 μm; Ethispheres® 850, NPPharm Ltd., Bazainville, France) were used as inert cores. Hydroxypropyl methylcellulose (HPMC; Pharmacoat 606, Shin-Etsu Chemical Ltd., Tokyo, Japan) was used as binder for the drug-layering process. Ibuprofen sodium (Sigma-Aldrich Chemie GmbH, Steinheim, Germany) was used as model drug.
Orsolya Péterfi, Nikolett Kállai-Szabó, Kincső Renáta Demeter, Ádám Tibor Barna, István Antal, Edina Szabó, Emese Sipos, Zsombor Kristóf Nagy, Dorián László Galata, Artificial intelligence-aided endoscopic in-line particle size analysis during the pellet layering process, Journal of Pharmaceutical Analysis, 2025, 101227, ISSN 2095-1779, https://doi.org/10.1016/j.jpha.2025.101227.
See our next webinar:
“How to predict the impact of excipient variabilities in your tablet design?“
Date: 18th of March, Time: 4:00 PM (Amsterdam, Berlin)
