An Automatic Analysis System for High-Throughput Clostridium Difficile Toxin Activity Screening

Abstract

Clostridium difficile infection (CDI) is an increasing global health threat and major worldwide cause of hospital-acquired diarrhea. The development of novel therapies to effectively treat this bacterial pathogen is an unmet clinical need. Here, we describe an image processing and classification algorithm that automatically identifies toxin-induced cytotoxicity to host cells based on characteristic morphological changes. This efficient and automatic algorithm can be incorporated into a screening platform to identify novel anti-toxin inhibitors of the C. difficile major virulence factors TcdA and TcdB, and contains the following steps: image enhancement, cell segmentation, and classification. We tested the algorithm on 504 images (containing 5096 cells) and achieved 93% sensitivity and 91% specificity, indicating that the proposed computational approach correctly classified most of the cells and provided reliable information for an effective screening platform. This algorithm achieved higher classification results compared to existing cell counter and analysis programs, scoring 92.6% accuracy. Compared to visual examination by a researcher, the algorithm significantly decreased classification time and identified toxin-induced cytotoxicity in an unbiased manner. Availability: Examples are available at home.agh.edu.pl/jaworek/CDI.

Pre-Trained Deep Convolutional Neural Network for Clostridioides Difficile Bacteria Cytotoxicity Classification Based on Fluorescence Images

“…Finally, we compared the achieved results with our previous paper [13]. We used the same image dataset as a reference point.…”

“…There is therefore a need to develop new methods of CDI treatment—such as novel therapies focusing on directly neutralising the effects of C. difficile toxins, instead of killing the bacteria [ 12 ]. However, developing a new drug requires extensive testing, which is often carried out manually, making it a slow, tedious and unreliable process [ 13 ]. In order to improve it, microscopic images need to be automatically checked, individual cells segmented and then classified as either dead or alive by an algorithm.…”

“…Although there are many works about different CDI treatments and about predicting CDI based on clinical data using machine learning methods [ 16 , 17 ], to the best of our knowledge there are almost no research reports about deploying computer image processing methods for C. difficile cell segmentation in conjunction with machine learning methods for the classification of these cells (the segmentation method proposed by Memariani and Kakadiaris [ 18 ] is designed for a different modality, namely for scanning electron microscopy). In our previous research [ 13 ] we proposed a preliminary solution for the assessment of cells; however, we observed erroneous results for cells that had irregular borders, uneven intensity, and protruding lines. The following algorithm has been implemented including pre-processing stage and classification part.…”

“…We segmented them, indexed them, and used their position to track whole cells on corresponding GFP images. Finally, single cell windows, extracted from GFP images, were used to train a naive Bayes classifier, which had an accuracy of 91%, sensitivity of 93%, and specificity of 91% respectively [ 13 ]. Furthermore, open source image processing programs like ImageJ contain different plugins like Fiji , Ice or BioVoxxel Toolbox to calculate the cells; however, these are based on a permanent interaction with the user or manual selection.…”

Transfer Learning Methods as a New Approach in Computer Vision Tasks with Small Datasets

“…This infection is one of the most common contagious disease in hospitals. With antibiotics being one of the risk factors, novel forms of therapies are constantly developed [3]. Our algorithms help to speed up the process of testing and make the results more reliable than human subjective approach.…”