Multi-resolution border segmentation for measuring spatial heterogeneity of mixed population biofilm bacteria
Introduction
Biofilms are communities of microorganisms attached to surfaces that may develop a complex heterogeneous three-dimensional structure. Understanding interactions in multispecies biofilms may contribute to treatment of polymicrobial biofilm infections as well as rational design of engineered biofilms [1], [2]. Analysis of microbial biofilms by confocal laser scanning microscopy (CLSM) yields stacks of digital images that can be combined to give a three-dimensional view of the biofilm. Sophisticated tools are needed to quantitatively analyze and compare biofilms growth under different conditions. Gilbert et al. [2] found that the cultivation of the consortium of bacteria as a biofilm indicated that two species: P. putida KT2440 pSB337 and Escherichia coli SD2 could cohabit as a population of attached cells. Information on the relationships among bacteria in multispecies biofilms can be obtained from CLSM image stacks by combining pattern recognition and image processing techniques such as clustering and segmentation. Previously there was limited work done in this area using automated tools. Our goal is to develop a fully automated method (program, tool) for segmenting and quantifying biofilms, and hence bacteria.
Multi-dimensional clustering methods are simple and serve as powerful tools for classifying and segmenting pixels. Though most of the clustering methods require the number of clusters be known, that requirement does not represent a burden in the case of dual-species biofilms since the maximum number of clusters representing bacteria or void space is three [3]; a unique color label for each type of bacteria and black (the third one) for the void space.
The k-means clustering algorithm and Fuzzy C-Means clustering algorithm are the most popular techniques applied in image processing [3], [4], [5], [6], [7]. Other techniques such as artificial neural networks are particularly useful for classification and clustering [8]. Neural networks can also be used to represent very complex nonlinear systems. Some self-organizing neural networks such as Self-Organizing Maps (SOM) and Learning Vector Quantization neural networks (LVQNN) have been applied for image segmentation and their results are considered satisfying [9], [10].
Our objective is to facilitate better understanding of the relationship of these two microorganisms by the quantitative image analysis of the biofilm. Understanding interactions in multispecies biofilms may contribute to treatment of polymicrobial biofilm infections as well as rational design of engineered biofilms [1], [2].
In our proposed algorithm images are segmented using Fuzzy C-Means approach followed by two-stage SOM or LVQ neural network to identify clusters. The outcome of the segmentation is quantified by the number of cluster objects of each kind of microorganism within sections of the biofilm, and the centroid distances between the identified clusters.
Initial experimental evaluations of our algorithm showed its effectiveness in analyzing the distribution of microbial populations in a model two-member biofilm comprised of an ampicillin-resistant and an ampicillin-sensitive strain of E. coli, each with a distinct red or green fluorescent label.
Section 2 of this paper describes the methods used in our analysis, including the image segmentation techniques and multi-resolution border extraction. Performance evaluation and experimental results are given in Section 3 and conclusions in Section 4, respectively.
Section snippets
Segmentation using Fuzzy C-Means, SOM and LVQ neural networks
Image segmentation is the first step of image analysis for biofilm images. Image segmentation is a process of partitioning an image into disjoint regions such that each region is homogeneous, according to a certain uniformity criterion, and no union of any two adjacent regions is homogeneous. Image segmentation requires processing of a huge amount of data. After segmentation, we can easily obtain the centroids of each bacteria cluster and calculate the distances between them.
The most
Performance evaluation and experimental results
The image segmentation algorithms described in previous sections were implemented using Matlab for 15 bacteria images. The results from the segmentation stage are used for identifying the various levels of mixedness among each patch of biofilm data. The cluster identification process is preceded with the multi-resolution analysis scheme outlined in Section 2.4. The benefits of the multi-resolution border extraction can be seen from Table 1 where the number of clusters is decreasing as the
Conclusions
Multi-resolution clustering algorithms have been developed to segment colored mixed population of biofilm images of bacteria. Image contours have been successfully used to extract boundaries of objects at multi-resolution levels.
Contours of all objects in the image are automatically generated and used to segment and label each object. The contour-based segmentation process is independent of the number of objects on each image slice or their sizes. This procedure has the capability of filtering
Saeid Belkasim is currently an associate professor, Department of Computer Science, GSU. His research interests include digital image processing, image segmentation, compression, image description, image database, feature extraction and shape recognition.
References (15)
- et al.
Phase-based optimal image thresholding
Digital Signal Process
(2003) - et al.
Color image segmentation: advances and prospects
Pattern Recognit
(2001) - et al.
Segmentation of color images using a two-stage self-organizing network
Image Vision Comput
(2002) - et al.
Microbial biofilms
Annu Rev Microbiol
(1995) - et al.
A constructed microbial consortium for biodegradation of the organophosphorus insecticide parathion
Appl Microbiol Biotechnol
(2003) - Zhong D, Yan H. Color image segmentation using color space analysis and fuzzy clustering. In: Proceedings of IEEE...
- et al.
An efficient k-means clustering algorithm: analysis and implementation
IEEE Trans Pattern Anal Machine Intell
(2002)
Cited by (0)
Saeid Belkasim is currently an associate professor, Department of Computer Science, GSU. His research interests include digital image processing, image segmentation, compression, image description, image database, feature extraction and shape recognition.
Gordana Derado is currently a PhD student at Emory University, formerly a PhD student at the Department of Computer Science, GSU, under the supervision of Dr. Saeid Belkasim.
Rizi Aznita is currently a PhD student at the Department of Computer Science, GSU, working to develop software analysis tools for biofilm bacteria under the supervision of Dr. Saeid Belkasim.
Eric Gilbert is currently an associate professor, Department of Biology, GSU. His research interests include using biofilms to combine bacteria for accomplishing useful functions, and cooperative interactions in polymicrobial biofilms.
Heather O’Connell is currently a postdoctoral research associate at the Centers for Disease Control, Atlanta, GA.