Automated particle inspection of continuously freeze-dried products using computer vision

Area of Science:

• Pharmaceutical Manufacturing
• Biotechnology
• Computer Vision

Background:

• Continuous manufacturing is advancing in the pharmaceutical industry.
• Continuous freeze-drying offers advantages over batch methods, including enhanced visual inspection potential.
• Manual visual inspection of freeze-dried products is labor-intensive and prone to drawbacks.

Purpose of the Study:

• To develop an automated visual inspection method for freeze-dried biopharmaceuticals using artificial intelligence.
• To compare the performance of AI-driven inspection against human visual inspection.

Main Methods:

• Manually induced particles (50μm to 1 mm) were added to continuously freeze-dried pharmaceutical samples.
• A custom imaging setup was used to capture high-resolution images of the samples.
• Multiple object detection models, including YOLOv7, were trained on the image dataset.

Main Results:

• The YOLOv7 model achieved a particle detection precision of 88.9% and a recall of 81.2%.
• The AI model demonstrated superior quantitative and qualitative performance compared to human inspection.
• Inference time for the YOLOv7 model was less than 1 second per vial.

Conclusions:

• AI-powered computer vision, specifically YOLOv7, offers a highly effective and efficient automated solution for visual quality assessment in continuous freeze-drying.
• This technology can significantly improve the reliability and reduce the labor associated with inspecting freeze-dried pharmaceutical products.