A hybrid CNN-Random Forest algorithm for bacterial spore segmentation and classification in TEM images.

Area of Science:

• Microscopy
• Computational Biology
• Bacteriology

Background:

• Accurate segmentation and classification of bacterial spore layers in Transmission Electron Microscopy (TEM) images are crucial for understanding spore structure and function.
• Current methods can be labor-intensive and prone to human bias, necessitating automated and robust approaches.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning approach for segmenting and classifying bacterial spore layers from TEM images.
• To assess the model's performance against existing classification algorithms and its ability to identify spores with damaged cores.

Main Methods:

• A hybrid model combining a Convolutional Neural Network (CNN) for feature extraction and a Random Forest (RF) classifier for classification was developed.
• The model was trained and tested on TEM images of bacterial spores, including those exposed to chemical treatments to induce core damage.

Main Results:

• The proposed CNN-RF model achieved 73% accuracy, 64% precision, 46% sensitivity, and 47% F1-score on test data.
• The model demonstrated superior robustness and generalization ability for non-linear segmentation compared to AdaBoost, XGBoost, and SVM classifiers.
• The model successfully identified bacterial spores with damaged cores, validated by TEM analysis.

Conclusions:

• The hybrid CNN-RF approach offers a valuable, automated method for identifying and characterizing bacterial spore features in TEM images.
• This method has the potential to significantly reduce manual labor and mitigate human bias in spore analysis.
• The model's ability to detect core damage provides a new tool for studying spore viability and response to environmental stressors.