Deep learning-based framework for Mycobacterium tuberculosis bacterial growth detection for antimicrobial...

Area of Science:

• Microbiology
• Bioinformatics
• Medical Diagnostics

Background:

• Tuberculosis (TB) remains a leading infectious cause of death globally, with drug resistance posing a significant challenge.
• Accurate and accessible diagnostics are crucial for effective TB treatment and control.
• Current high-throughput phenotypic testing methods using 96-well plates can be difficult to interpret, especially with low bacterial growth or poor image quality.

Purpose of the Study:

• To develop and validate a novel deep learning framework, the TB Microbial Analysis System (TMAS), for automated detection of *Mycobacterium tuberculosis* growth in 96-well microtiter plates.
• To enhance the accuracy and efficiency of Minimum Inhibitory Concentration (MIC) determination in tuberculosis drug susceptibility testing (DST).
• To differentiate true bacterial growth from artefacts in plate images.

Main Methods:

• Leveraged state-of-the-art deep learning models within the TMAS framework to analyze images of 96-well microtiter plates.
• Trained and refined TMAS using a dataset of 4,018 plate images from the CRyPTIC consortium.
• Evaluated TMAS performance against established standards for automated growth detection.

Main Results:

• TMAS achieved an essential agreement of 98.8% in detecting *M. tuberculosis* growth, significantly exceeding the International Organization for Standardization (ISO) 90% threshold.
• The system demonstrated robust performance in identifying true bacterial growth and distinguishing it from artefacts.
• TMAS significantly outperformed existing automated algorithms like AMyGDA, particularly in challenging low-growth or low-quality image scenarios.

Conclusions:

• TMAS provides a reliable and automated solution for analyzing microbial growth in TB drug susceptibility testing.
• The deep learning approach enhances the accuracy and efficiency of MIC determination, supporting expert interpretation.
• TMAS has the potential to improve global TB diagnostics, especially in resource-limited settings.