CrysAI: How to Harness the Power of AI for Faster, Smarter Particle Analysis

For pharmaceutical leaders, delays in microscopy don’t just slow science they add regulatory risk, extend timelines, and increase the cost of bringing new medicines to patients.

Microscopy has long been central to process development in pharmaceuticals and biotechnology. From monitoring crystallization to assessing cell cultures, microscopic images provide critical insight into morphology and variability. Yet the process of analyzing these images often creates bottlenecks. Manual segmentation is slow, results can vary between analysts, and these limitations restrict teams from making timely, confident decisions. 

Recent advances in deep learning have begun to shift this reality. One example is CrysAI™, a tool developed by our Innovation Team for microscopy image analysis and deployed within the iAchieve® WebApp. Designed with particle and cell systems in mind, it highlights a broader movement; using machine learning to accelerate and standardize image-based decision-making in development programs. 

How CrysAI™ Analyzes Images: From Pixels to Decisions

Behind the scenes, CrysAI™ uses deep learning to recognize and measure features in microscopy images. Whether data comes from SEM, TEM, or optical systems, the software translates them into actionable metrics like particle size distributions and morphology insights.

At its core, CrysAI™ transforms images into reproducible, decision-ready data in seconds:

1. Image Ingestion
   Microscopy images of particle or biologics systems are uploaded via the iAchieve® WebApp.

2. Image Analysis
   Each image is run through deep learning models (Mask R-CNN) for mapping and segmentation.

3. Insight Generation
   Morphology metrics are quantified (size, shape factors, d10/d50/d90) and correlated with process parameters to detect shifts, with outputs provided in visual and tabular form for decision-making.

Because models can be retrained on project-specific data, teams can extend CrysAI™ across modalities and scales - from complex particle morphologies to diverse cell systems.

Why It Matters for Development and Tech Transfer

CrysAI™’s defining advantage is speed. Manual segmentation of microscopy images can take 30–40 minutes per image. Automated approaches reduce this to a matter of seconds. By replacing manual segmentation with automation, CrysAI™ enables reproducible, real-time insights that accelerate development and reduce risk.

Reported outcomes include:

• More than 90% reproducibility across sample types

• Up to 98% reduction in analysis time

• 3x improvements in detecting process deviations

• Decision cycles cut in half

For scientists working on process development or tech transfer, this means bottlenecks are reduced, and microscopy can play a more active role in monitoring and control strategies.

To illustrate how automated microscopy delivers value in practice, the following case studies highlight its impact across real-world applications.

Case Study: Cell Cluster Analysis

In practice, automated microscopy was put to work monitoring cell clusters across multiple bioreactors, offering insights into both speed and reproducibility. Cell clusters were monitored across five bioreactors over 15 days. Hundreds of images at different magnifications were collected daily. While offline microscopy provided valuable data, manual analysis was too slow - delaying insight into variability and limiting process confidence.

Automating segmentation using CrysAI reduced analysis time from around 40 minutes per image to roughly 10 seconds. This not only accelerated workflows but also improved reproducibility and allowed microscopy to scale into a more routine monitoring tool. The team gained clearer visibility into cluster dynamics and greater confidence in transferring processes between sites.

The result: clearer visibility into cluster dynamics, improved reproducibility, and reduced risk when transferring processes across sites.

Case Study: Crystallization Monitoring

In crystallization particle morphology can determine yield and reproducibility. In one project, CrysAI™ was integrated with process analytical technologies such as ParticleTrack™ and EasyViewer™. By automating particle characterization, real-time morphology metrics were fed into process control decisions.

The approach allowed earlier detection of form conversions and tighter control over cooling and agitation conditions. Reported outcomes included a 95% reduction in analysis time, improved reproducibility, and greater robustness across batches.

This shifted microscopy from a passive diagnostic step to an active lever in control strategies directly impacting yield and reproducibility.

From Bottleneck to Driver of CMC Decisions

The case studies highlight a wider theme: embedding digital tools into experimental workflows can remove hidden bottlenecks and shift microscopy from an after-the-fact analysis into a real-time decision driver.

For researchers, the benefits go beyond speed. Standardized analysis improves reproducibility, a key concern in both development and regulatory contexts. Linking image attributes to process parameters opens opportunities for deeper understanding of system behaviour. And by automating routine segmentation, scientists can focus on higher-level interpretation rather than manual tasks.

Of course, limitations remain. As with any model-driven approach, results depend on training data quality. But with human oversight and retraining, CrysAI™ strengthens rather than replaces expertise.

Looking Ahead: Digital Microscopy in Tomorrow’s CMC

The adoption of machine learning in image analysis is part of a broader digital shift in Chemistry, Manufacturing, and Controls (CMC). By integrating microscopy with real-time data streams and analytical platforms, development teams are creating workflows that are faster, more reproducible, and better aligned with regulatory expectations.

Whether applied to crystallization, bioreactor monitoring, or other image-intensive processes, CrysAI™ is part of a larger digital transformation in CMC, where AI-enabled tools are aligning development with regulatory expectations for speed, reproducibility, and data integrity.