Deep learning-based defect detection in film-coated tablets using a convolutional neural network.

Journal: International journal of pharmaceutics
PMID:

Abstract

Film-coating is a critical step in pharmaceutical manufacturing. Traditional visual inspections for film-coated tablet defect assessment are subjective, inefficient, and labor-intensive. We propose a novel approach utilizing machine learning and image analysis to address these limitations. Here, defects of four types- chipping, breaking, color non-uniformity and speckling, were manually induced in red-orange film-coated placebo tablets. Utilizing a 3-D printed tray and a unique segmentation approach, images of good and defective tablets were collected. A convolutional neural network (CNN) was employed to quantitatively analyze the defects. The model was trained on a comprehensive dataset of 25,200 images of tablets, augmented through various transformations to improve robustness. The CNN's performance was evaluated using metrics such as accuracy, precision, recall, and F1-score. The multi-class classification model demonstrated an accuracy of 99.7% in detection of defects in film-coated tablets, clearly outperforming static rule-based method which had 45%, 45% and 70% error in detecting dimensions- major axis, minor axis, and surface area of the tablets, respectively. This work demonstrates a valuable tool for pharmaceutical manufacturers, providing a standardized, objective, and efficient method for defect detection in tablets and presents a promising solution for ensuring product quality and accelerating the development of new pharmaceutical products.

Authors

  • Kabir A Pathak
    Pharmaceutical Sciences and Clinical Supply, Merck & Co., Inc., Rahway, NJ, USA.
  • Prapti Kafle
    Pharmaceutical Sciences and Clinical Supply, Merck & Co., Inc., Rahway, NJ, USA. Electronic address: prapti.kafle@merck.com.
  • Ajit Vikram
    Process Research & Development, Merck & Co., Inc., Rahway, NJ, USA.