Automated Mycobacterium tuberculosis Detection in Multivariant Digitized Ziehl-Neelsen Staining Using Faster R-CNN...

Area of Science:

• Medical Diagnostics
• Computer Science
• Infectious Diseases

Background:

• Tuberculosis (TB) remains a significant public health issue, especially in Indonesia.
• Microscopic examination of sputum smears via Ziehl-Neelsen staining is a common TB diagnostic method.
• Manual TB detection faces challenges due to staining variations and subjectivity.

Purpose of the Study:

• To develop an automated system for detecting tuberculosis bacteria.
• To leverage deep learning, specifically the Faster R-CNN algorithm with ResNet-50, for TB detection.

Main Methods:

• Utilized the Faster R-CNN algorithm with ResNet-50 architecture.
• Implemented the system using Python and the TensorFlow Object Detection API.
• Applied data augmentation techniques including rotation, flipping, and color processing.

Main Results:

• Achieved 88% accuracy, 94% precision, 93% recall, and 94% F1-score.
• The model successfully outputs annotated images pinpointing TB bacteria locations.
• Demonstrated the effectiveness of the automated system in identifying TB bacteria.

Conclusions:

• Deep learning offers a promising approach for automating TB detection.
• The developed system can assist medical professionals in TB diagnosis, especially in resource-limited settings.
• Automated TB detection can improve diagnostic efficiency and reliability.