DeepPulmoTB: Powering Robust Segmentation of Tuberculosis Lesions in Chest Computerized Tomography (CT) with Rich Annotations

Citation Author(s):
zhuoyi
tan
Submitted by:
Zhuoyi Tan
Last updated:
Mon, 07/08/2024 - 15:58
DOI:
10.21227/z95g-cn77
License:
Creative Commons Attribution
43 Views
Categories:
Artificial Intelligence
Standards Research Data
Medical Imaging
Keywords:
tuberculosis, Computerized Medical Image Processing
0
0 ratings - Please login to submit your rating.
ACCESS DATASET CITE SHARE/EMBED

Abstract 

Tuberculosis (TB) remains a major global health problem with high incidence and mortality rates worldwide. In recent years, with the rapid development of computer-aided diagnosis (CAD) tools, CAD has played an increasingly important role in supporting tuberculosis diagnosis. However, the development of CAD for TB diagnosis relies heavily on well-annotated computerized tomography (CT) datasets. Unfortunately, the currently available annotations in TB CT datasets are still limited, which hinders the development of CAD tools for TB diagnosis to some extent. To overcome this limitation, we introduce DeepPulmoTB, a CT multi-category semantic segmentation dataset specifically designed for TB with rich annotations. DeepPulmoTB encompasses three vital segmentation mask categories in TB diagnosis: consolidations, lung cavities, and both lungs. Furthermore, the annotations in DeepPulmoTB undergo meticulous review by professional physicians to ensure their high quality and reliability. 

Instructions: 

DeepPulmoTB is a comprehensive Tuberculosis lesion tissue segmentation dataset. In DeepPulmoTB, a series of images for each patient consists of about 125 slices in the axial projection. For the consolidation and lung cavity segmentation categories, the data are sourced from ImageCLEF2022 TB Cavern detection and cavern report, totaling 618 patient CT images (approximately 77,250 CT slice images). For the lung area mask (lung mask version III) segmentation category, there are 2498 patient CT images (approximately 312,250 CT slice images), combining Mask versions I and II from the ImageCLEF 2020-2022 TB challenge. 

DeepPulmoTB is divided into two parts, Part1 and Part2, in which part1 is a multi-category semantic segmentation task, and part2 is a lung segmentation task. Note: Part 2 only contains the lung region segmentation recognition mask, which does not mean that there is no TB lesion tissue.

After decompressing DeepPulmoTB, you can get the following directory:

├── DeepPulmoTB

│ ├── Training_Mask_Dataset

│ │ ├── Part 1

│ │ │ ├── TRN_00.nii.gz

│ │ │ ├── TRN_000.nii.gz

│ │ │ ├──  …

│ │ ├── Part 2

│ │ │ ├── CTR_TRN_001

│ │ │ ├── ...

 

In Part 1, pixel values 1 to 3 represent consolidation, Lung Mask version III, and Lung cavity. In Part 2, pixel value 1 represents Lung Mask version III.

Dataset Files

LOGIN TO ACCESS DATASET FILES

 

QUESTIONS?