Automatic identification of mycobacterium tuberculosis with conventional light microscopy

Area of Science:

• Medical diagnostics
• Microbiology
• Image analysis

Background:

• Distinguishing Mycobacterium tuberculosis (tuberculosis bacilli) in conventional microscopy is challenging due to poor contrast with the background.
• Conventional light microscopy lacks the clarity of fluorescence microscopy for identifying bacilli.

Purpose of the Study:

• To develop an automated method for identifying tuberculosis bacilli using conventional microscopy images.
• To overcome the limitations of conventional microscopy in distinguishing bacilli from the background.

Main Methods:

• Utilized Red and Green color channels from RGB images to create Red minus Green (R-G) images.
• Applied global adaptive threshold segmentation on R-G images, where bacilli appear as white regions.
• Removed artifacts using morphological, color, and size filters for improved segmentation.

Main Results:

• The R-G image transformation effectively highlighted tuberculosis bacilli against a dark background.
• Post-processing filters successfully reduced artifacts in the segmented images.
• Achieved a best sensitivity of approximately 76.65% for bacilli identification.

Conclusions:

• This work presents the first automated method for tuberculosis bacilli identification using conventional light microscopy.
• The proposed R-G image analysis and filtering technique offers a viable approach for improved automated diagnosis.