AI-SEED – AI-Based Image Recognition for Macro and Microscopy
Laboratories generate large volumes of images and documents every day. Samples are examined under a microscope, culture plates are evaluated, and order forms are received in many different formats. To this day, much of this is still checked, counted, and recorded by hand by specialized staff. This is time-consuming, and with the increasing volume of samples and a shortage of skilled workers, it is increasingly becoming a bottleneck. Furthermore, assessments can vary slightly depending on the person and their performance on a given day.
This is where automated image analysis using artificial intelligence comes into play. The field is called computer vision, or “machine vision.” In this process, a computer learns to recognize and evaluate image content based on numerous sample images. Three tasks are particularly important here:
- Classification: The system assigns an image or a section of an image to a category, such as a bacterial species.
- Object recognition: The system locates specific objects in the image and marks their position, for example, individual bacteria or cells.
- Image segmentation: The system precisely delineates structures within the image, such as colonies on a culture plate.
This allows routine tasks to be accelerated and standardized. Specialized staff gain time for tasks that require experience and judgment.
The AI-SEED Project: Artificial Intelligence for Selective Experimental Evaluation in Diagnostics
AI-SEED develops AI-based image recognition for microscopy and macroscopy in the field of diagnostics. The focus is on microscopy, but the analysis of macroscopic images is also possible. The underlying technology is the same across all projects. However, each application is individually adapted to the respective conditions, samples, and equipment to ensure it functions reliably in everyday laboratory use.
All subprojects share a common foundation: the automated processing and analysis of image data. This includes classification, object recognition, and image segmentation. The individual projects build upon this common methodology.
Bilateral Projects with Partners
The following applications are currently being developed in collaboration with laboratory partners:
- Text recognition and information extraction from non-standardized order forms: Relevant data is automatically extracted from forms that vary in structure and presentation.
- Differentiation of bacteria based on their shape and Gram stain
- Detection and localization of bacteria in gastric biopsies
