Identification of Urinary Tract Infection Causing Bacteria from Microbial Colony Images using Deep Learning

Abstract

Diagnosis of microbiological and bacterial infections involves extensive procedures of sample culture and microscopic examination. The diagnosis process begins with the identification of symptoms followed by collection of test samples in the form of blood, scraps of skin lesions, mucus, sputum, urine etc. The entire process of bacteria routine culture and bacteriological examination may take as long as 10 to 11 days. Due to the long time required for the standard process of species identification, high costs and need of human expertise it is beneficial to use methods that do not rely on conventional methods. In this work we propose a comprehensive Deep Learning based approach to identify urinary Tract Infection (UTI) causing bacteria from microbial colony images of bacteria cultured on agar plates. The approach explains the use of VGG-19 and Inception-v3 deep learning architectures, in bacteria identification evaluated on The Annotated Germs for Automated Recognition (AGAR) dataset, an image dataset of microbial colonies cultured on agar plates. The findings affirmed the significant potential of employing deep learning techniques for the identification of microbial colonies and their classification using Petri dish images.

Introduction

I. INTRODUCTION

Urinary tract infections (UTIs) stand as one of the most prevalent bacterial infections affecting millions worldwide. Characterized by the invasion of bacteria into the urinary system, UTIs pose as a significant health burden. The causative agents, often bacteria like Escherichia coli and even fungi, can cause inflammation in various parts of the urinary tract, causing discomfort, pain, and, if left untreated, potential complications such as kidney infections. UTIs  not only diminish the  quality of life for affected individuals but also strain healthcare systems globally. The ubiquity and recurrent nature of UTIs underscore the pressing need for efficient diagnostic methods to ensure timely and accurate intervention [4].

The traditional diagnostic path for identifying the bacteria responsible for UTIs is complex. It starts with identifying symptoms, which can range from frequent and painful urination to lower abdominal pain. When a UTI is suspected, samples are collected, most commonly in the form of urine. However, the following steps involve routine culture and bacteriological examination, which is a time-consuming and resource-intensive process. Culturing bacteria from collected samples and analysing them under a microscope can take 10 to 11 days, delaying diagnosis and impeding treatment initiation. In addition, this procedure necessitates a team of proficient microbiologists, which adds to the difficulty and expense of diagnosing UTIs . One of the biggest barriers to providing timely and efficient healthcare is the length of time needed for traditional species identification procedures for bacteria that cause UTIs [1]. This issue is made worse by the resource-intensive nature of these methods and the scarcity of skilled labour. It becomes essential to find alternative approaches that can overcome the limitations of traditional diagnostic techniques. The difficulties presented by current UTI diagnostic methodology necessitate the investigation of novel approaches to address the critical issues of time inefficiency, high costs, and reliance on skilled human personnel. The financial burden of extensive culture and examination procedures necessitates the development of cost-effective alternatives that can streamline the diagnostic process without sacrificing accuracy. The intersection of medical science and advanced technologies, particularly deep learning, offers a promising avenue in this context. The search for a more efficient and accessible diagnostic solution for UTIs is part of a larger effort to improve global healthcare outcomes and reduce the burden of preventable complications associated with bacterial infections.

This project's main goal is to develop and put into practice a Deep Learning-based method for the quick and precise identification of bacteria that cause urinary tract infections [2]. This method aims to lower the time, expense, and reliance on human skill associated with diagnosis. The project employs The Annotated Germs for Automated Recognition (AGAR) dataset, a collection of microscopic images capturing microbial colonies cultured on agar plates. This dataset ensures the representation of diverse bacterial species relevant to UTIs. In particular, the research uses VGG-19 and Inception-v3, two cutting-edge deep learning architectures, for bacteria identification.

II. LITERATURE SURVEY

III. AGAR DATASET

The dataset comprises of 18,000 photos with 336,442 annotated colonies of 5 standard microbes. Notably, the dataset exhibits a balanced distribution of instances for different microbes, crucial for building robust deep learning models [13].

The dataset includes five classes, namely Staphylococcus aureus, Bacillus subtilis, Pseudomonas aeruginosa, Escherichia coli and Candida albicans, while annotations are stored in JSON format with information about the number and type of microbe, environment, and coordinates of bounding boxes. The dataset consists of higher and lower-resolution images. The higher-resolution images are further classified into bright, dark, and vague subgroups which are a result of variations in lighting conditions during data acquisition as seen in Figure 1. This enhances the dataset's versatility and ensures its effectiveness across different experimental setups. This balance is crucial for the development of a resilient deep learning model. The AGAR dataset was chosen for this work keeping in mind the most common bacteria which cause urinary tract infections. Even though there are some bacteria like Klebsiella pneumoniae, Proteus mirabilis which might present themselves as more infectious than the ones in the AGAR dataset, the AGAR dataset is the closest we can get, to satisfy the suitable conditions for our work [3].

A. Types of Images

IV. IMPLEMENTATION

In order to identify the bacteria, the first step was to identify microbial colonies. To achieve this, simple image segmentation was performed on the images.

Upon image segmentation, an ensemble of two deep learning architectures was applied on the dataset namely VGG19 (Visual Geometry Group) and Inception-v3 (Iv3). The VGG-19 is a Convolutional Neural Network that is 19 layers deep (16 convolution layers, 3 Fully connected layer, 5 MaxPool layers and 1 SoftMax layer) as seen in Figure 2. The network takes in RGB images as input of size (224X224) pixels. The top layer used in VGG-19 is trained on ImageNet weights. The pretrained top layer is capable of classifying images of upto 1000 different categories [14].

In which × is the convolutional function which describes the connection between the weights of the ith and jth features in the (l−1)th, lth layers, bj is the bias value, and φ is the activation function [15].

Inception-v3 is convolutional neural network that is 48 layers deep. It takes in RGB images of size 299x299 pixels as input. In the case of Inception-v3 however, the top layer with ImageNet weights was omitted owing to different experimental observations of induces bias. However, the exclusion of the top layer forces a trade-off between the induced bias and decreasing variance [16].

Despite the availability of more efficient, scalable and robust image recognition architectures like Inception-v5, EfficientNet, and Exception Model, the architectures VGG-19 and Inception-v3 were used owing to their efficiency in region specific convolution output, based on modified CNN architecture as seen in Figure 3. EfficientNet is a recognition specific architecture and Exception model is the latest general and highly accurate model. These architectures were not chosen as they did not adhere entirely to the critical aspects of the experiment. As the bacteria colony detection was critical in the classification process, segmentation and models which concentrate on region specific output were chosen. Another reason for choosing Inception-v3 and VGG-19 was their low accuracy loss and suitability for categorical data.

Maxpooling was used in both the individual architectures, as region specific features were needed for colony identification. In the first stage, of training 150 images of each bacterium consisting of both low and high resolution were chosen for model training. The models were individually trained on the data with a split of 80% and 20% for training and validation respectively [5].

Conclusion

In our experiments conducted with the deep architectures VGG-19 and Inception-v3 on the AGAR dataset to classify UTI causing bacteria, we could infer that there is excellent scope for modern deep learning techniques in the field of healthcare, with medical image processing being the greatest example. Both the deep learning architectures performed well in identifying the class of bacteria from images of microbial colonies on petri dishes with high accuracies. Also, the results clearly showcase that an ensemble of two or more such architectures would certainly perform just as well and is better for new data and images of various resolutions. The challenge in bringing out such an approach was the limited availability of image data, other than the AGAR dataset and the manual labelling of random training data. The models require the images to be reasonably clear and of good resolution with ample illumination. Such a requirement would hinder the direct application of these approaches in the real diagnosis process. However, these deep learning techniques are more than capable of identifying abnormal and mixed colonies grown on petri dishes usually small and located at the edge of the dish, which even a trained technician may sometimes overlook. Overall, such automated and technology assisted approaches can help experts in the diagnosis process, and shorten the time required. These methods can be employed in other diagnostic procedures for microbial infections, with the anticipation of yielding favourable outcomes.

References

[1] Shallu Kotwal, et-al.: “Automated Bacterial Classifications Using Machine Learning Based Computational Techniques: Architectures, Challenges and Open Research Issues”, 12 October 2021, https://doi.org/10.1007/s11831-021-09660-0 [2] Priya Rani, et-al.: “Machine Learning and Deep Learning Based Computational Approaches in Automatic Microorganisms Image Recognition Methodologies, Challenges, and Developments”, 31 August 2021, https://doi.org/10.1007/s11831-021-09639-x [3] Sylwia Majchrowska, et-al.: “AGAR a microbial colony dataset for deep learning detection”, 3 Aug 2021, https://arxiv.org/abs/2108.01234 [4] Shimaa A. Nagro, et-al.: “Automatic Identification of Single Bacterial Colonies Using Deep and Transfer Learning”, 14 November 2022, https://ieeexplore.ieee.org/document/9950247 [5] Adriana Borowa, et-al.: “Identifying Bacteria Species on Microscopic Polyculture Images Using Deep Learning”, 26 September 2022, https://ieeexplore.ieee.org/document/9903318 [6] Natalia Czerep, et-al.: “Deep Neural Networks Approach to Microbial Colony Detection—A Comparative Analysis”, 27 July 2022 https://link.springer.com/chapter/10.1007/978-3-031- 11432-8_9 [7] Hedieh Sajedi, et-al.: “Image-processing based taxonomy analysis of bacterial macromorphology using machine-learning models”, 29 August 2020 https://link.springer.com/article/10.1007/s11042-020- 09284-9 [8] Francesco Asnicar, et-al.: “Machine learning for microbiologists”, 15 November 2023, https://link.springer.com/article/10.1038/s41579-023-00984-1 [9] Yuandi Wu, S. Andrew Gadsden, “Machine learning algorithms in microbial classification: a comparative analysis”, 19 October 2023 https://www.frontiersin.org/articles/10.3389/frai.2023.1200994/full [10] Jaken Whipp, et-al.: “YOLO-based Deep Learning to Automated Bacterial Colony Counting”, 29 December 2022, https://ieeexplore.ieee.org/document/9999125 [11] Problem Solving Strategies (Problem Solving in Mathematics) by Arthur Engel [12] Z. Cai and N. Vasconcelos. Cascade r-cnn: Delving into high quality object detection. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6154–6162, 2018. [13] AGAR Dataset, https://agar.neurosys.com/ [14] Understanding the VGG19 Architecture, https://iq.opengenus.org/vgg19- architecture/ [15] A VGG-19 model with Transfer Learning and Image Segmentation for Classification, https://www.mdpi.com/2624-7402/4/4/56 [16] Inception V3 CNN Architecture Explained, https://medium.com/@AnasBrital98/inception-v3-cnn-architecture-explained- 691cfb7bba08 [17] VGG16 and VGG19, https://keras.io/api/applications/vgg/ [18] InceptonV3, https://keras.io/api/applications/inceptionv3/ [19] T. Shaily and S. Kala, ‘‘Bacterial image classification using convolutional neural networks,’’ in Proc. IEEE 17th India Council Int. Conf. (INDICON), Dec. 2020, pp. 1–6, doi: 10.1109/INDICON49873.2020.9342356. [20] T. Ching and D. S. Himmelstein, ‘‘Opportunities and obstacles for deep learning in biology and medicine,’’ J. Roy. Soc. Interface, vol. 15, no. 141, 2018, Art. no. 20170387. [21] F. Cong, S. Lin, H. Wang, S. Shang, L. Long, R. Hu, Y. Wu, N. Chen, and S. Zhang, ‘‘Biological image analysis using deep learning-based methods: Literature review,’’ Digit. Med., vol. 4, no. 4, p. 157, 2018, doi: 10.4103/digm.digm_16_18. [22] L. Huang and T. Wu, ‘‘Novel neural network application for bacterial colony classification,’’ Theor. Biol. Med. Model., vol. 15, no. 1, pp. 1–16, Dec. 2018, doi: 10.1186/s12976-018-0093-x. [23] H. Sajedi, F. Mohammadipanah, and A. Pashaei, ‘‘Image-processing based taxonomy analysis of bacterial macromorphology using machine-learning models,’’ Multimedia Tools Appl., vol. 79, nos. 43–44, pp. 32711–32730, Nov. 2020, doi: 10.1007/s11042-020-09284-9. [24] H. Wang, H. Ceylan Koydemir, Y. Qiu, B. Bai, Y. Zhang, Y. Jin, S. Tok, E. C. Yilmaz, E. Gumustekin, Y. Rivenson, and A. Ozcan, ‘‘Early detection and classification of live bacteria using time-lapse coherent imaging and deep learning,’’ Light: Sci. Appl., vol. 9, no. 1, p. 118, Dec. 2020, doi: 10.1038/s41377-020-00358-9. [25] B. D. Satoto, M. I. Utoyo, R. Rulaningtyas, and E. B. Koendhori, ‘‘Classification of features shape of Gram-negative bacterial using an extreme learning machine,’’ in Proc. IOP Conf., Earth Environ. Sci., 2020, vol. 524, no. 1, Art. no. 012005, doi: 10.1088/1755-1315/524/1/012005. [26] S. Patel, ‘‘Bacterial colony classification using atrous convolution with transfer learning,’’ Ann. Romanian Soc. Cell Biol., vol. 25, no. 4, pp. 1428–1441, 2021.

Copyright

Copyright © 2024 Dr. D. R. Ramesh Babu, Sunanda , Ayush Kumar Manas, Dhruv Rathi, Tejas R, Tushar Pathak. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.