Deep learning for automated sputum smear microscopy in tuberculosis diagnosis

Area of Science:

• Medical diagnostics
• Artificial Intelligence in Healthcare
• Microbiology

Background:

• Tuberculosis (TB) remains a significant global health challenge, disproportionately affecting low- and middle-income countries.
• Current diagnostic methods like sputum smear microscopy are time-consuming, subjective, and prone to errors.
• Deep learning offers a potential solution to automate and improve the accuracy of TB diagnosis.

Purpose of the Study:

• To evaluate the efficacy of deep learning models for automated detection of Acid-Fast Bacilli (AFB) in sputum smears.
• To compare the performance of different convolutional neural network (CNN) architectures in TB diagnosis.

Main Methods:

• Trained and validated three CNN architectures (custom CNN, ResNet50, EfficientNetB0) on 8000 digitized Ziehl-Neelsen stained sputum smear images.
• Images were annotated by experienced microbiologists, pre-processed, and augmented to handle variations in staining and illumination.
• Addressed class imbalance using class-weighted binary cross-entropy loss and minority over-sampling.

Main Results:

• EfficientNetB0 achieved the highest performance, with 92% accuracy, 89.1% sensitivity, and 94.5% AUC-ROC.
• Deep learning models demonstrated performance comparable to expert microbiologists in microscopy analysis.
• The study highlighted a significant reduction in analysis time compared to manual methods.

Conclusions:

• Deep learning models, especially EfficientNet, can effectively automate TB sputum smear microscopy.
• The proposed automated method can facilitate high-throughput screening, reduce diagnostic delays, and minimize human error in resource-constrained settings.
• Future work involves integrating the model into a portable, point-of-care diagnostic device for broader clinical application.