Remote analysis of sputum smears for mycobacterium tuberculosis quantification using digital crowdsourcing

Area of Science:

• Medical Diagnostics
• Infectious Diseases
• Computational Biology

Background:

• Tuberculosis (TB) remains a leading infectious cause of death globally.
• While Xpert MTB/RIF is a breakthrough, smear microscopy is still crucial for TB diagnosis in low-resource settings.
• Novel approaches are needed to enhance TB diagnostic capabilities.

Purpose of the Study:

• To assess the feasibility of a crowdsourced approach for analyzing tuberculosis sputum smear images.
• To determine if untrained volunteers can accurately identify acid-fast bacilli (AFB) in digitized microscopic images.
• To evaluate the diagnostic performance of this crowdsourced method.

Main Methods:

• A crowdsourcing strategy was implemented using an online game format.
• 1790 anonymous volunteers with no prior experience analyzed 60 digitized sputum smear images.
• Performance was evaluated using F1 score, sensitivity, and positive predictive value, optimizing based on combined analyses.

Main Results:

• The crowdsourced analysis demonstrated significant feasibility in tuberculosis image analysis.
• Optimized combined analysis from multiple players yielded the best performance.
• High performance metrics were achieved: F1 score of 0.933, sensitivity of 0.968, and positive predictive value of 0.91.

Conclusions:

• Crowdsourcing offers a viable and effective method for tuberculosis image analysis.
• This approach can augment traditional diagnostic methods, particularly in resource-limited areas.
• Volunteer-based analysis shows promise for improving global tuberculosis detection rates.