Source-linked AI summary
A Comprehensive Review of Image Analysis Methods for Microorganism Counting: From Classical Image Processing to Deep Learning Approaches
Jiawei Zhang, Chen Li, Md Mamunur Rahaman, Yudong Yao, Pingli Ma, Jinghua Zhang, Xin Zhao, Tao Jiang, Marcin Grzegorzek
TL;DR
Manual microorganism counting is time-consuming, subjective, and difficult to scale, motivating image-analysis methods. This review synthesizes microorganism-counting research by organism type and segmentation approach, finding established preprocessing and segmentation practices alongside expanding deep-learning methods and identifying segmentation as a key future direction.
Problem
Traditional counting methods are time-consuming, subjective, operationally complex, and difficult to apply at large scale.
Method
The review groups studies into bacteria and other microorganisms, then analyzes methods by segmentation approach and summarizes common image-processing techniques.
Results
The review identifies medial and Gaussian filtering, thresholding and Otsu thresholding, and PCA and neural networks as frequently used techniques across counting systems.
Takeaways & Limitations
Deep-learning-based counting is promising, while semantic segmentation is identified as a future direction for combining microorganism classification with more precise segmentation.
Takeaways & Limitations
Most deep-learning methods reviewed are used for classification rather than segmentation, leaving deep-learning-based segmentation as a stated development opportunity.
Abstract
from arXiv · showhide
Microorganisms such as bacteria and fungi play essential roles in many application fields, like biotechnique, medical technique and industrial domain. Microorganism counting techniques are crucial in microorganism analysis, helping biologists and related researchers quantitatively analyze the microorganisms and calculate their characteristics, such as biomass concentration and biological activity. However, traditional microorganism manual counting methods, such as plate counting method, hemocytometry and turbidimetry, are time-consuming, subjective and need complex operations, which are difficult to be applied in large-scale applications. In order to improve this situation, image analysis is applied for microorganism counting since the 1980s, which consists of digital image processing, image segmentation, image classification and suchlike. Image analysis-based microorganism counting methods are efficient comparing with traditional plate counting methods. In this article, we have studied the development of microorganism counting methods using digital image analysis. Firstly, the microorganisms are grouped as bacteria and other microorganisms. Then, the related articles are summarized based on image segmentation methods. Each part of the article is reviewed by methodologies. Moreover, commonly used image processing methods for microorganism counting are summarized and analyzed to find common technological points. More than 144 papers are outlined in this article. In conclusion, this paper provides new ideas for the future development trend of microorganism counting, and provides systematic suggestions for implementing integrated microorganism counting systems in the future. Researchers in other fields can refer to the techniques analyzed in this paper.
1 Introduction
Microorganisms are important across medical, industrial, environmental, and biological applications, but manual counting is laborious, subjective, and limited at scale. This review introduces image-analysis approaches, tracing digital processing, segmentation, classification, and evaluation methods for microorganism counting.
- Microorganisms and their applications: Microorganisms include bacteria, viruses, fungi, and algae, with both harmful and beneficial roles in health, industry, food production, and environmental processes.Examples include disease-causing viruses, fermentation-related yeast, penicillin-producing fungi, and microorganisms used in wastewater treatment.
- Motivation: Manual counting methods include plate counting, hemocytometry, and turbidimetry, but they require complex operations and can delay or distort quantification.Plate counting requires cultivation, while attached cells may be counted as one colony, producing counts lower than the actual number of living bacteria.
- Motivation: Large samples, weak colony–background contrast, heavy workloads, and observer subjectivity reduce the reliability of traditional counting results.The same observer may obtain different results when examining the same sample at different times.
- Image-analysis basis: Digital image processing converts image signals into digital signals for computer-based enhancement, denoising, restoration, coding, and compression.In microorganism counting, image analysis can automatically and quickly count particles and extract particle characteristics.
- Research development: Since 1995, microorganism-counting research using digital image processing has accelerated, while machine-learning and deep-learning applications have expanded particularly since 2015.The review associates this growth with more accurate segmentation, including segmentation of adherent colonies.
- Review scope and methods: The review organizes counting methods by microorganism type and segmentation approach, covering thresholding, edge detection, region extraction, third-party tools, and deep learning.It also describes preprocessing, feature reduction, classification, and system evaluation; Otsu thresholding is identified as a standard method for many images, while PCA maps n-dimensional features to k-dimensional principal components.
2 Bacteria counting methods
Bacteria counting is a major focus of microorganism research because bacteria are ecologically important and relevant to food and industrial safety. Manual counting remains tedious and subjective, motivating computer image-analysis methods.
- Bacteria play a crucial role in carbon and nitrogen cycles and are closely related to human daily life.
- Bacteria counting examines bacterial numbers and colony size as important aspects of microorganism research.
- Computer image analysis is studied for bacteria counting because manual counting is tedious, redundant, and subjective.
2.1 Classic counting methods
Classic bacteria-counting methods combine image enhancement, filtering, thresholding, segmentation, feature extraction, and object separation to detect and enumerate colonies. Reported results include low counting errors and strong precision, recall, or agreement with manual counting in several studies.
- Image enhancement: Gray-level contrast, colony area, and within-colony gray distribution support bacteria counting, with less than 3% error reported by Niyazi et al. (2007).
- Image enhancement: Histogram equalization, convolutional filtering, median filtering, and connected-component counting achieved accuracy above 95% in Shenglang and Yongguang (2005).
- Image enhancement: Fourier and Mellin transforms provide scale- and rotation-invariant analysis, with differences from manual counts ranging from 1 to 3% across twelve samples.The reported standard deviation was 4.51.
- Thresholding: Global thresholding commonly combines color-space conversion, denoising, morphological enhancement, and circle detection for bacterial or colony enumeration.The cited studies use YCbCr, Lab, adaptive median, flatten, top-hat, Gaussian, moving-average, and median filters.
- Thresholding: Iterative local, adaptive, and Otsu thresholding methods segment colonies after filtering or enhancement, often followed by region analysis or morphological processing.Extended minima, connected neighborhoods, color similarity, and object-size distributions assist colony detection or counting.
- Segmentation and colony separation: Otsu and watershed segmentation produced precision, recall, and F-measure values of 0.97, 0.96, and 0.96 on blue medium plates.For another dataset, the proposed counter reported 0.61±0.29 precision, 0.94±0.06 recall, and 0.69±0.20 F-measure, compared with Clono-Counter values of 0.22±0.25, 1.00±0.00, and 0.29±0.31.
- Segmentation and colony separation: Watershed, edge detection, region growing, concavity analysis, and connected-region processing separate touching or aggregated colonies before counting.The reviewed approaches include plate and colony masks, edge filters, multiple threshold values, and region-growing segmentation.
2.2 Machine learning and deep learning counting methods
Machine-learning and deep-learning methods combine image preprocessing, segmentation, and learned classification to automate microorganism counting. Reported systems include PCA, SVM, ANN, and CNN approaches, with CNN-based counting reaching 92.8% accuracy and substantially reducing analysis time.
- Machine learning methods: PCA separates biological patterns from surrounding areas, while nearest-neighbor searching, distance transforms, and watershed divide touching or overlapping colonies.Yoon et al. apply local absorbance maxima detection and masking before colony separation, followed by bottom-hat extraction from rim images.
- Machine learning methods: SVM-based systems segregate colonies from culture backgrounds using chromatic information and supervised descriptions of the background.The reviewed SVM studies address bacteria counting and classification, including uncertainty regions formed from background information.
- Machine learning methods: ANN processing combines Marr-Hildreth edge detection, thresholding, rank filtering, and neural classification for automated bacteria-image analysis.The network uses 6 input nodes, 5 intermediate nodes, and 3 output nodes; images are analyzed at 100 images per hour with 5% count variation between filters.
- Deep learning methods: CNN systems augment and normalize colony images before classification and counting with four convolutional layers and one fully connected layer.Training accuracy flattened after 15,000 iterations; 50,000 iterations required approximately 3 hours on an Nvidia Titan Black GPU.
- Deep learning methods: 92.8% accuracy was obtained for CNN-based bacterial colony classification and counting.The reviewed workflow applies image transformations, normalization or contrast enhancement, CNN classification, and subsequent watershed processing.
2.3 Third-party tools
Third-party image-analysis tools support bacterial counting through specialized preprocessing, segmentation, fluorescence analysis, and object measurement. Their reported outcomes range from close agreement with manual counts to low average error and rapid processing.
- Specialized systems: Real-time bacterial counting reconstructs time-series high-resolution images before segmentation and measurement of colony diameter and number.The processed images are illustrated as reconstructed high-resolution images from time-series data.
- Specialized systems: Cellstat independently detects and measures surface-associated microcolonies and single cells by changing object-recognition parameters.The system was developed to determine bacterial biovolume and supports simultaneous quantitative RNA and DNA staining.
- Reported agreement: ProtoCOL counts were highly correlated with true counts but slightly lower than the true count.The result was obtained from processing digital-camera images of bacterial colonies.
- Reported agreement: Hua et al.’s method measured connected regions as bacteria and reported an average error of no more than 1.6% against manual counting.Image enhancement and median filtering removed noise, while local binary fitting performed segmentation.
- Reported agreement: SigmaScan Pro found no significant difference between software thresholding and manual counting for fluorescently stained biofilm cells.The passage reports r>0.05 and attributes discrimination between bacteria and background to Live/Dead staining.
2.4 Summary of image analysis based counting for bacteria
The reviewed bacteria-counting literature uses recurring combinations of filtering, segmentation, and classification techniques. The field expanded from early digital image processing toward more rapid growth of machine-learning and deep-learning research after 2010.
- Counting techniques: Median and Gaussian filters are the most frequently used preprocessing methods in reviewed bacteria-counting studies.Other preprocessing approaches include contrast enhancement, color processing, adaptive filtering, and morphological operations.
- Counting techniques: Thresholding, distance transform, and watershed are the most frequently used image-segmentation methods.The review identifies threshold segmentation, edge detection, and region extraction as broad segmentation categories.
- Counting techniques: SVMs and ANNs are the most frequently used classifier algorithms in the reviewed bacteria-counting literature.The table summarizes studies combining these classifiers with thresholding, watershed, edge detection, and other segmentation methods.
- Literature summary: The review organizes bacteria-counting studies across preprocessing, segmentation, classification, and evaluation categories.Table 1 lists related work, microorganism type, preprocessing, segmentation, classification, and evaluation information.
3 Other microorganism counting methods
Counting methods for other microorganisms combine thresholding, enhancement, filtering, segmentation, feature extraction, and classification to separate objects and estimate counts. Reported systems include specialized algorithms and third-party software, with performance varying across organisms and image conditions.
- Counting methods based on thresholding: Thresholding-based methods count fungi, mycelia, yeast, protozoa, algae, pus cells, and other microorganisms after filtering, labeling, or feature extraction.Some methods also measure filament lengths or separate single cells from overlapping objects.
- Counting methods based on thresholding: Histogram thresholding combined with filtering, enhancement, morphology, and connected-region labeling achieved a 2.56% average relative error for fungi counting.
- Third-party tools: Third-party tools automated counting across applications, including bacteria and archaea enumeration, marine fouling measurements, and colony-forming-unit estimation.AnalySIS produced automated-to-manual ratios of 97.7% for bacteria and 92.2% for archaea, while COVASIAM estimated 95.47% of manual colony counts on average.
- Third-party tools: Automated cell enumeration was highly correlated with manual counts, with Bennke et al. reporting r2>0.9 using measured object and intensity features.
- Summary: Counting research for non-bacterial microorganisms remains relatively limited because their complex structures and fewer available datasets make precise segmentation and evaluation more difficult.The review notes that bacteria datasets are more abundant, encouraging evaluation on bacteria images.
- Summary: The review identifies median and Gaussian filtering, thresholding and Otsu thresholding, and PCA and neural networks as frequently used components.
4 Analysis of image processing based counting methods
The review identifies preprocessing, segmentation, classification, connected-region detection, and feature extraction as central components of microorganism image counting. It traces practical methods from color and illumination correction through thresholding, watershed, and CNN-based analysis, while highlighting remaining segmentation gaps.
- Image preprocessing, segmentation, classification, connected-region detection, and feature extraction are identified as effective approaches for microorganism image counting.
- Image preprocessing methods: Color conversion, illumination correction, denoising, and contrast enhancement prepare microorganism images for segmentation.RGB-to-gray-scale or HSI conversion, background subtraction, filtering, and histogram equalization address color, shading, noise, and weak contrast.
- Image segmentation methods based on thresholding: Thresholding is widely used for microorganism segmentation, with global and Otsu methods performing best under suitable contrast or target-background conditions.Otsu thresholding is simple and effective when target and background area differences are negligible, but overlapping gray levels reduce separation accuracy.
- Image segmentation methods based on thresholding: Local and multilevel thresholding address uneven illumination and detail loss that limit single-threshold segmentation.Local thresholds adapt to regional brightness, while multilevel thresholding can preserve more detailed image information.
- Other image segmentation methods: Edge detection and distance-transform watershed methods provide additional routes for colony segmentation.Sobel and Laplacian operators are used for edge detection, while distance transform is commonly combined with watershed.
- Image classification methods: CNNs automatically extract image features and process high-dimensional data quickly, with 92.8% accuracy reported for bacteria colony counting.Pooling layers may lose valuable information during training.
- Analysis of potential methods: Deep learning automates classification, but microorganism segmentation still commonly requires manual adjustment or traditional techniques.The review identifies semantic segmentation as a future direction for simultaneous classification and segmentation.
5 Conclusion and future work
The review organizes microorganism-counting methods by organism type and segmentation approach, covering more than 144 related papers and commonly used processing techniques. It concludes that image-analysis methods show substantial research potential, with future work focused on deep learning, precise segmentation, and integrated counting systems.
- The review groups microorganism-counting studies by bacteria and other microorganisms, then separates them by segmentation approach.Covered approaches include thresholding, edge detection, third-party tools, and deep learning.
- Classic methods developed from the 1980s to the 2000s, while deep learning since the 2010s has produced high-accuracy microorganism-counting results.
- The review finds that image-analysis microorganism counting has substantial research potential and that its methods can be referenced in other digital image-analysis fields.
- Future development is directed toward deep learning for precise segmentation and integrated systems combining sampling, imaging, and analysis.The review also notes increasing attention to microorganism analysis following the COVID-19 outbreak.