Automatic Bacillus anthracis bacteria detection and segmentation in microscopic images using UNet+

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Microbiology

Background:

• Anthrax, caused by Bacillus anthracis, remains a significant health concern, especially in developing nations.
• Manual microscopic diagnosis is prone to errors due to human fatigue and visual limitations.
• Accurate and rapid detection of Bacillus anthracis is crucial for timely disease management.

Purpose of the Study:

• To develop an automated system for the rapid diagnosis of anthrax.
• To enable simultaneous detection and segmentation of Bacillus anthracis bacteria in microscopic images using AI.
• To improve diagnostic accuracy and efficiency compared to traditional methods.

Main Methods:

• Utilized deep learning techniques, specifically UNet and UNet++ neural network architectures.
• Applied these models for the automated detection and segmentation of Bacillus anthracis in microscopic images.
• Conducted experiments on a private dataset, evaluating performance on both patch and whole images.

Main Results:

• The UNet++ architecture demonstrated high effectiveness and efficiency in segmenting Bacillus anthracis.
• Achieved 97% accuracy and a 0.96 dice score on patch test images.
• Obtained 98% recall and 97% accuracy on whole raw images, outperforming human specialists in some cases.

Conclusions:

• The proposed AI-based system offers an effective and efficient solution for automated anthrax diagnosis.
• UNet++ shows exceptional performance in segmenting bacteria, overcoming challenges like image artifacts and overlapping objects.
• The system provides benefits of low cost, high speed, and reduced reliance on specialized personnel for diagnosis.