Identifying Bacteria Species on Microscopic Polyculture Images Using Deep Learning

Area of Science:

• Microbiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Traditional microbiological diagnosis is time-consuming, taking up to 11 days.
• Current deep learning methods reduce diagnosis time but still require monoculture samples.
• Iterative culture for monoculture samples remains a bottleneck in rapid diagnosis.

Purpose of the Study:

• To present a feasibility study for significantly shortening microbiological diagnosis time.
• To introduce a novel method for analyzing polyculture images, bypassing the need for monoculture samples.
• To evaluate the effectiveness of multi-MIL for rapid bacterial identification.

Main Methods:

• Development of multi-MIL, a novel multi-label classification method based on multiple instance learning.
• Creation of a dataset with microscopic images of all combinations of four bacterial species.
• Evaluation of the multi-MIL approach on the developed dataset.

Main Results:

• Achieved a Receiver Operating Characteristic Area Under the Curve (ROC AUC) above 0.9.
• Demonstrated the feasibility of analyzing polyculture images for microbiological diagnosis.
• Proved the potential for further shortening diagnosis times compared to existing methods.

Conclusions:

• The multi-MIL method is a feasible approach for rapid microbiological diagnosis.
• Analyzing polyculture images with multi-MIL significantly reduces diagnostic time.
• This work paves the way for future research with a larger number of bacterial species.