MicrobIA Haemolysis Dataset

If you use the dataset for your research please cite this article as: M. Savardi, A. Ferrari, A. Signoroni, Automatic hemolysis identification on aligned dual-lighting images of cultured blood agar plates, Computer Methods and Programs in Biomedicine, Vol. 156, 2018, pp. 13-24

[Training set + code 2.3GB]

We provide a Python and Jupyter Notebooks implementation of the proposed software, divided in three esay to use notebooks, that follow the presented pipeline. First, the fine alignment software, that automatically takes care of the complete alignment process. In input it needs the back and top light images of the plate, and the segmentation mask, and return an aligned version and debug images in order to visualize how the method is working. Then, the feature extraction and dimensionality reduction phases, and an example of parameter selec- tion in order to tune the SVM classification model. This script loads the dataset and metadata in order to train an SVM classifier. Finally, the haemolysis detection pipeline that, starting from the provided full plate dataset, comprehensive of annontations, counting and segmentation masks, provides the classification results as an overlay over the plate images. The dataset is composed of more that 2400 annotated segments.

[ Full plate images 9.2GB] 

A second part of the dataset (full plate images, for blind test) is provided. The dataset is composed of full plate images, with segmentation masks and additional metadata with counting and pathogen information.

MicrobIA Segments Enumeration Dataset

If you use the dataset for your research please cite this article as: A. Ferrari, et al., Bacterial colony counting with Convolutional Neural Networks in Digital Microbiology Imaging, Pattern Recognition (2016), http://dx.doi.org/10.1016/j.patcog.2016.07.016i

 

[segments enumeration .tar.gz 1.1GB]

The dataset is composed by approximately 29000 images of single bacterial colonies, agglomerates (2-6 bacterial colonies) and confluential growth, on Blood agar plates of urine samples.

In enumeration_segments.json, each image has two relative path, regarding the raw image and the segmentation mask. The label is present on "data" filed, in the form of numerical label and text label:

0 : 1 colony

1 : 2 colonies

2 : 3 colonies

3 : 4 colonies

4 : 5 colonies

5 : 6 colonies

6 : confluential

7 : outlier

 

MicrobIA Images Dataset (Beta ver. 0.1)

 If you use the dataset for your research please cite this article as: A. Ferrari, et al., Bacterial colony counting with Convolutional Neural Networks in Digital Microbiology Imaging, Pattern Recognition (2016), http://dx.doi.org/10.1016/j.patcog.2016.07.016i

 

[sample  .zip ~250MB

[part1  .zip ~5GB]

[part2  .zip ~5GB]  

[part3  .zip ~5GB]  

[part4  .zip ~5GB]  

[part5  .zip ~5GB]  

[part6  .zip ~5GB]  

[part7  .zip ~5GB]  

[part8  .zip ~5GB]  

[part9  .zip ~5GB]  

[part10  .zip ~5GB]  

[part11  .zip ~5GB]  

[part12  .zip ~5GB]  

[part13  .zip ~5GB]  

[part14  .zip ~5GB]  

[part15  .zip ~5GB]  

[part16  .zip ~5GB]  

[part17  .zip ~5GB]  

[part18  .zip ~5GB]  

[part19  .zip ~5GB]  

[part20  .zip ~5GB]   

 

The images database contains a selected collection of bacterial cultures on solid agar plates images. The images have been collected at the Niguarda Hospital (Milan, Italy). They have been selected among real clinical scenarios. The database has the aim to offer a first benchmark to assess image analysis algorithms performances for this application, which is still at an early stage. The dataset is addressed mainly to image analysis engineers interested in develop image analysis algorithms. The database can be interesting also for microbiologists that wants to use it for educational purposes, however it has not been designed for that purpose, so feel free to contact us if you are interested in collaborate to develop this kind of solutions. Currently the database contains images of 209 different bacterial growth.

 

  • The images collection is representative for assessing goodness of segmentation algorithms of complex colonies, even when massive confluential growth occurred. It can be a valid benchmark even for image analysis specialists not actually interested to the clinical application field, but simply interested on evaluating effectiveness of general purpose segmentation algorithms.

 

  • The database contains images of the solid agar plate before specimen inoculation and after that bacterial growth occurs, those are  co-registered in order to reduce clutter disturbs, by means of the difference of the images  before and after bacterial growth.

 

  • When available, the Maldi Tof identification of the colonies on the agar has been included.

 

  • Moreover, for each image qualitative or quantitative, whether possible, bacterial load estimation or colonies counting has been performed by a clinical specialist.

 

Together with the images is included a json file (easily accessible by matlab or python code) that includes informations relatively to the plate contained in each image, more details about the metadata and the research application of the dataset are specified in the README.txt file.

 

MicrobIA Hyperspectral Dataset (Beta ver. 0.1) 

[dataset] 

The hyperspectral database contains a selected collection of spectral signatures from bacteria colonies on solid blood agar plates. The hypercubes have been jointly collected by University of Brescia (Information Engineering Department) and Copan Italia (Brescia, Italy). They have been selected among ATCC strains. The database has the aim to offer a first benchmark to assess image analysis algorithms performances for this application. The dataset is addressed to researchers and professionals in the fields of digital microbiology, hyperspectral imaging (especially short-range application domains), machine learning and image analysis. The database can be interesting also for microbiologists that wants to use it for educational purposes, however it has not been designed for that purpose, so feel free to contact us if you are interested in collaborate to develop this kind of solutions. Currently the database contains spectral signature of 9 different pathogens.

An early work has been published based on data contained in this database and another more complete study is under review for a scientific journal.

If you use this database for research purposes and scientific publication please cite our works.

G. Turra, N. Conti, A. Signoroni, Hyperspectral Image Acquisition and Analysis of Cultured Bacteria for the Discrimination of Urinary Tract Infections, IEEE EMBC 2015 (Milano, Italy), pp.759-762