AutoCellANLS: An Automated Analysis System for Mycobacteria-Infected Cells Based on Unstained Micrograph

Area of Science:

• Microbiology
• Computational Biology
• Medical Imaging

Background:

• Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), is a major global infectious disease.
• Automated detection of Mtb infection is crucial for disease control.
• Current methods often rely on expensive and time-consuming fluorescence staining, hindering automation.

Purpose of the Study:

• To develop a novel automated system (AutoCellANLS) for Mtb detection and morphological feature recognition.
• To overcome the limitations of fluorescence staining in Mtb detection protocols.
• To enable accurate and efficient diagnosis of bacterial infections using AI.

Main Methods:

• Utilized unsupervised machine learning (UML) and deep convolutional neural networks (CNNs) for cell detection and classification.
• Employed an improved level set segmentation model with circular Hough transform (CHT) for adaptive cell detection.
• Developed a Cell-net using transfer learning strategies (TLS) to classify virulence-specific morphological changes.

Main Results:

• Achieved high accuracy: 95.13% for cell detection, 95.94% for morphological classification, 94.87% for sensitivity, and 96.61% for specificity.
• Successfully detected morphological differences between infected and uninfected mammalian cells at various infection time points.
• Demonstrated improved accuracy (>11%) and efficiency compared to previous AI-aided methods, adaptable to different cell lines.

Conclusions:

• AutoCellANLS provides a novel, automated approach for Mtb detection using phase-contrast microscopy, eliminating the need for fluorescence staining.
• The system accurately identifies Mtb-induced cellular changes, offering a promising tool for diagnosing bacterial infections.
• This study opens new avenues for investigating bacterial pathogenesis and improving diagnostic capabilities.