A sputum smear microscopy image database for automatic bacilli detection in conventional microscopy.

Area of Science:

• Medical Imaging
• Microbiology
• Computer Vision

Background:

• Sputum smear microscopy is crucial for diagnosing diseases like tuberculosis.
• Accurate automated bacilli detection requires high-quality, well-annotated image datasets.
• Existing datasets may lack diversity in background content and detailed object annotations.

Purpose of the Study:

• To present a novel image database for the development of automatic bacilli detection algorithms.
• To provide a resource for training and validating computer-aided diagnosis systems in microscopy.
• To establish a gold standard for evaluating the performance of bacilli recognition algorithms.

Main Methods:

• Creation of two distinct image databases: an autofocus database (1200 images) and a segmentation/classification database (120 images).
• Images acquired from Kinyoun-stained sputum slides with varying background content (high and low).
• Annotation of bacilli in the segmentation database using geometric shapes (circle for true, rectangle for agglomerated, polygon for doubtful) by trained technicians.

Main Results:

• The database includes high-resolution images (2816 × 2112 pixels) suitable for detailed analysis.
• Image classification based on background content (high/low) is provided.
• Precise object-level annotations serve as a gold standard for performance evaluation.

Conclusions:

• The presented image database is a valuable resource for advancing automated bacilli detection in sputum microscopy.
• This dataset facilitates the development of more accurate and reliable diagnostic tools.
• The annotated images enable robust validation of algorithm performance metrics such as accuracy, sensitivity, and specificity.