Bacterial image analysis using multi-task deep learning approaches for clinical microscopy

Area of Science:

• Microbiology
• Computer Science
• Biotechnology

Background:

• Bacterial image analysis is crucial for structural biology, disease diagnosis, and drug discovery.
• Automating bacterial image analysis with deep learning (DL) enhances accuracy, efficiency, and standardization.
• DL enables rapid, reliable analysis for better understanding and control of bacterial phenomena.

Purpose of the Study:

• To develop and evaluate DL object detection networks for automated bacterial image analysis.
• To detect and classify Escherichia coli (E. coli) bacteria based on their growth stages.
• To compare the performance of SSD-MobileNetV2, EfficientDet, and YOLOv4 for multi-task bacterial detection.

Main Methods:

• Three DL object detection networks (SSD-MobileNetV2, EfficientDet, YOLOv4) were implemented.
• A multi-task DL framework classified bacteria into rod-shaped, dividing, and microcolony stages.
• Models were trained using preprocessed data (augmentation, annotation, splitting) and evaluated using mAP, precision, recall, and F1-score.

Main Results:

• All DL models showed high detection accuracy on test images.
• YOLOv4 achieved the highest confidence scores and used distinct bounding boxes for different growth stages.
• YOLOv4 outperformed other models with a 98% mAP, 86% precision, 97% recall, and 91% F1-score.

Conclusions:

• DL approaches are effective for multi-task bacterial image analysis.
• The proposed models automate the detection and classification of E. coli from microscopic images.
• YOLOv4 demonstrated superior performance, highlighting its potential for bacterial image analysis applications.