ImageCLEFmed Tuberculosis

Motivation

Welcome to the 5th edition of the Tuberculosis Task!

Tuberculosis (TB) is a bacterial infection caused by a germ called Mycobacterium tuberculosis. About 130 years after its discovery, the disease remains a persistent threat and a leading cause of death worldwide according to WHO. This bacteria usually attacks the lungs, but it can also damage other parts of the body. Generally, TB can be cured with antibiotics. However, the different types of TB require different treatments, and therefore the detection of the TB type and characteristics are important real-world tasks.

Task history and lessons learned:

• Tuberculosis task exists in ImageCLEF since 2017 and it was modified from year to year.
• In the first and second editions of this task, held at ImageCLEF 2017 and ImageCLEF 2018, participants had to detect Multi-drug resistant patients (MDR subtask) and to classify the TB type (TBT subtask) both based only on the CT image. After 2 editions we concluded that the MDR subtask was not possible to solve based only on the image. In the TBT subtask, there was a slight improvement in 2018 with respect to 2017 on the classification results, however, not enough considering the amount of extra data provided in the 2018 edition, both in terms of new images and meta-data.
• In the 3d edition Tuberculosis task was restructured to allow usage of uniform dataset, and included two subtasks - continued Severity Score (SVR) prediction subtask and a new subtask based on providing an automatic report (CT Report) on the TB case. In the 4th edition, the SVR subtask was dropped and automated CT report generation task was modified to be lung-based rather than CT-based.

Because of fairly high results achieved by the participants in the CTR task last year, the task organizers have decided to discontinue the CTR task at the moment and switch to the task which was not yet solved with high accuracy. It was decided to bring back to life the Tuberculosis Type classification task from the 1st and 2nd ImageCLEFmed Tuberculosis editions. The task dataset is updated in this year's edition. The dataset was extended in size, and some additional information is available for part of the CT scans.

We hope that utilizing the recent Machine Learning and Deep Learning methods will allow the participants to achieve much better results for the TB Type classification compared to the early editions of the task: 2017 and 2018. Here we encourage the participants to use any kind of methods and additional data which can be useful for the automatic classification of the TB Type. The participants are free to use general-purpose pre-trained models (e.g. using ImageNet), or to pre-train the models with the use of a 3rd-party dataset of CT-images, or to use a 3rd-party solution to pre-process the CT-scans of the current subtask.
Here are some resources that may be worth your attention:
* Catalog of trained models by NVIDIA: https://ngc.nvidia.com/catalog/models?orderBy=scoreDESC&pageNumber=0&que...
* Segmentation of anatomical structures in different modalities including CT: https://www.smir.ch/VISCERAL/Start
* Pulmonary Fibrosis Progression Challenge on Kaggle: https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression
* Segmentation of COVID-19 lesions on Grand-Challenge: http://covid-segmentation.grand-challenge.org/
* Other Kaggle CT-related datasets and competitions: https://www.kaggle.com/search?q=ct+in%3Adatasets+in%3Acompetitions
Please pay attention to the distribution license for each resource if you use it.

The task is available on the AIcrowd.

Preliminary schedule

• 13 January 2021: Task web-page goes live
• 26 February 2021: Release of the development data
• 26 April 2021: Release of the test data
• 30 April 2021: Registration closes
• 7 May 2021: Run submission deadline
• 17 May 2021: Release of the processed results by the task organizers
• 21 May 2021: Submission of participant papers [CEUR-WS]
• 21 May – 11 June 2021: Review process of participant papers
• 11 June 2021: Notification of acceptance
• 2 July 2021: Camera ready copy of participant papers and extended lab overviews [CEUR-WS]
• 21-24 September 2021: The CLEF Conference, Bucharest, Romania.

Task description

The goal of this subtask is to automatically categorize each TB case into one of the following five types: (1) Infiltrative, (2) Focal, (3) Tuberculoma, (4) Miliary, (5) Fibro-cavernous.

Data

In this edition, a dataset containing chest CT scans of 1338 TB patients is used: 917 images for the Training (development) data set and 421 for the Test set. Some of the scans are accompanied by additional meta-information, which may vary depending on data available for different cases. Each CT-image can correspond to only one TB type at a time. In this edition, there is each CT-scan corresponds to one patient.

For all patients we provide a single 3D CT image with an image size per slice of 512×512 pixels and number of slices being around 100. All the CT images are stored in NIFTI file format with .nii.gz file extension (g-zipped .nii files). This file format stores raw voxel intensities in Hounsfield units (HU) as well the corresponding image metadata such as image dimensions, voxel size in physical units, slice thickness, etc. A freely-available tool called "VV" can be used for viewing the image files. Currently, there are various tools available for reading and writing NIFTI files. Among them there are load_nii and save_nii functions for Matlab and Niftilib library for C, Java, Matlab and Python, NiBabel package for Python.

Moreover, for all the CT images we provide two versions of automatically extracted masks of the lungs. These data can be downloaded together with the patients CT images. The description of the first version of segmentation can be found here. The description of the second version of segmentation can be found here. The first version of segmentation provides more accurate masks, but it tends to miss large abnormal regions of lungs in the most severe TB cases. The second segmentation on the contrary provides more rough bounds, but behaves more stable in terms of including lesion areas. In case the participants use the provided masks in their experiments, please refer to the section "Citations" at the end of this page to find the appropriate citation for the corresponding lung segmentation technique.

Evaluation methodology

The results will be evaluated using unweighted Cohen’s Kappa (sample Matlab code).

Results

CEUR Woring Notes

for segmentation #2 (via non-rigid image registration):
• Liauchuk, V., Kovalev, V. ImageCLEF 2017: Supervoxels and co-occurrence for tuberculosis CT image classification. In: CLEF2017 Working Notes. CEUR Workshop Proceedings, Dublin, Ireland, CEUR-WS.org http://ceur-ws.org (September 11-14 2017) (URL: http://ceur-ws.org/Vol-1866/paper_146.pdf)
• BibTex:

  @inproceedings{CLEF2017WN,

  Title = {ImageCLEF 2017: Supervoxels and co-occurrence for tuberculosis CT image classification},
  Author = {Liauchuk, Vitali and Kovalev, Vassili},
  Keywords = {Supervoxels, Dictionary, Co-occurrence, Image Classification},
  Booktitle = {CLEF2017 Working Notes},
  Series = {{CEUR} Workshop Proceedings},
  Year = {2017},
  Publisher = {CEUR-WS.org $<$http://ceur-ws.org$>$},
  Month = {September 11-14},
  Address = {Dublin, Ireland}
  }

Organizers

• Serge Kozlovski <kozlovski.serge(at)gmail.com>, Institute for Informatics, Minsk, Belarus
• Vitali Liauchuk <vitali.liauchuk(at)gmail.com>, Institute for Informatics, Minsk, Belarus
• Yashin Dicente Cid <yashin.dicente(at)warwick.ac.uk>, University of Warwick, Coventry, England, UK
• Vassili Kovalev <vassili.kovalev(at)gmail.com>, Institute for Informatics, Minsk, Belarus
• Henning Müller <henning.mueller(at)hevs.ch>, University of Applied Sciences Western Switzerland, Sierre, Switzerland

Acknowledgements