Application of AI/ML in recognizing bacterial strains based on Fresnel patterns

A research and diagnostic organization was seeking an innovative solution for the automatic and rapid identification of bacterial strains. Their goal was to reduce costs, shorten analysis time, and enable the detection of multiple species in a single test—eliminating the need for expensive and time-consuming lab procedures.

Traditional bacterial identification methods had significant limitations: 

• Results could take up to 7 days and required highly skilled personnel. 

• Modern techniques like PCR were expensive and couldn’t detect all bacterial strains in a single sample. 

The client needed an effective system to automate the identification process and significantly accelerate analyses. 

The project was executed in three main phases: 

1. Design of the Imaging System and Initial Concept
   A dedicated optical system was developed to capture Fresnel patterns of bacterial colonies. These image data formed the foundation for further analysis. 
2. Data Processing and AI/ML Model Development
   We created a pipeline for extracting numerical morphological and textural features, enabling the differentiation of bacterial strain patterns. An AI/ML model was then trained to recognize dozens of bacterial species or indicate no match when similarity was low. 
3. Validation and Optimization
   The model achieved high identification accuracy—over 96%—as confirmed by an independent UK-based laboratory. The system was designed for easy integration with routine microbiological procedures.

• We developed a system that was enthusiastically received by the client’s lab teams. It reached an identification accuracy of over 96%, validated by a certified lab. 

• Analysis time was reduced from 7 days to a maximum of 24 hours. The system can identify multiple bacterial strains in a single test. 

• All hardware and reagents used in the solution comply with lab standards, with just one additional component. The same sample can also be analyzed using other methods, increasing diagnostic flexibility. 

• The development process and collaboration with the client ran smoothly and were continuously monitored. Our team actively shared knowledge, making implementation and future use of the system easier.