This website describes a collection of feature datasets, derived from chest computed tomography (CT) images, which can be used in the diagnosis of chronic obstructive pulmonary disease (COPD).
The images in this database are weakly labeled, i.e. per image, a diagnosis (COPD or no COPD) is given, but it is not known which parts of the lungs are affected. Furthermore, the images were acquired at different sites and with different scanners. These problems are related to two learning scenarios in machine learning, namely multiple instance learning or weakly supervised learning, and transfer learning or domain adaptation.
These problems are receiving quite a lot of interest from the machine learning community, but are more recent in medical image analysis. However, the problems is very important in practice, if machine learning methods are to be translated to the clinic. By publicly releasing these feature datasets, we hope that the problem of classifying COPD will receive more attention from both communities.
Conditions
The database can be used free of charge for research and educational purposes. Redistribution and commercial use is not permitted. If you publish using data from this website (journal publications, conference papers, abstracts, technical reports, etc.), please cite the following paper:
Veronika Cheplygina, Isabel Pino Peña, Jesper Holst Pedersen, David A. Lynch, Lauge Sørensen and Marleen de Bruijne. Transfer learning for multi-center classification of chronic obstructive pulmonary disease. In Journal of Biomedical and Health Informatics 22(5): 1486-1496, 2018. DOI 10.1109/JBHI.2017.2769800. https://arxiv.org/abs/1701.05013
We kindly ask you to register your details by sending an email to Veronika Cheplygina (vech (at) itu (dot) dk ) before you can download the data. This is to keep you updated should there be any changes to the data, any related publications, etc, that you might want to be informed of.
Data description
The dataset contains derived features (320-dimensional feature vectors) from CT images of patients and controls scanned at two different centers, with different scanners and scanning parameters. The two datasets are referred to as DLCST and Frederikshavn.
Each image is represented by 50 feature vectors, where each feature vector describes a volumetric ROIs of size 41 x 41x 41 voxels, extracted at random locations inside the lung mask. Two different feature types are available:
_gss.txt Gaussian scale space features, or histograms of intensity values in the ROI after filtering the image. Here we use eight filters (smoothed image, gradient magnitude, Laplacian of Gaussian, three eigenvalues of the Hessian, Gaussian curvature and eigen magnitude), four scales (0.6, 1.2, 2.4 and 4.8 mm), and histograms of ten bins. There are 320 features in total. The file feature_names.txt specifies the ordering of the filters and the scales.
_kdei.txt Kernel density estimation features, or a histogram of intensity values between -1100 and -600 Hounsfield units in the ROI. There are 256 features in total.
Per data subset, two additional files are available:
_index.txt The ID of the subject that each ROI belongs to
_labels.txt The label of each subject, determined by the subject’s diagnosis: COPD (1) or non-COPD (0). COPD diagnosis is determined according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria (FEV1/FVC < 0.7).
Publications
Publications using this database:
V. Cheplygina, I. Pino Peña, J. H. Pedersen, D. A. Lynch, L. Sørensen and M. de Bruijne. Transfer learning for multi-center classification of chronic obstructive pulmonary disease. In Journal of Biomedical and Health Informatics 22(5): 1486-1496, 2018.
Other relevant publications:
V. Cheplygina, L. Sørensen, D.M.J. Tax, M. de Bruijne, and M. Loog, Label Stability in Multiple Instance Learning, Medical Image Computing and Computer Assisted Intervention (MICCAI), 2015
V. Cheplygina, L. Sørensen, D.M.J. Tax, J.H. Pedersen, M. Loog, and M. de Bruijne, Classification of COPD with Multiple Instance Learning, International Conference on Pattern Recognition (ICPR), 2014.
L. Sørensen, M. Nielsen, P. Lo, H. Ashraf, J.H. Pedersen, and M. de Bruijne, Texture-Based Analysis of COPD: a Data-Driven Approach, IEEE Transactions on Medical Imaging 31(1): 70-78, 2012.
L. Sørensen, M. Loog, P. Lo, H. Ashraf, A. Dirksen, R.P.W. Duin, M. de Bruijne, Image Dissimilarity-Based Quantification of Lung Disease from CT, Medical Image Computing and Computer Assisted Intervention (MICCAI), 2010
L. Sørensen, P. Lo, H. Ashraf, J. Sporring, M. Nielsen, M. de Bruijne, Learning COPD Sensitive Filters in Pulmonary CT, Medical Image Computing and Computer Assisted Intervention (MICCAI), 2009
Pedersen, J.H., Ashraf, H., Dirksen, A., Bach, K., Hansen, H., Toennesen, P., Thorsen, H., Brodersen, J., Skov, B.G., Døssing, M. and Mortensen, J., 2009. The Danish randomized lung cancer CT screening trial—overall design and results of the prevalence round. Journal of Thoracic Oncology, 4(5), pp.608-614.
Acknowledgements
We thank Jesper Holst Pedersen (Department of Thoracic Surgery, Rigshospitalet, University of Copenhagen), Morten Vuust (Department of Diagnostic Imaging, Vendsyssel Hospital, Frederikshavn) and Ulla Møller Weinreich (Department of Pulmonology Medicine and Clinical Institute at Aalborg University Hospital) for their assistance in the acquisition of the CT images. We also thank the Danish Council for Independent Research.
Contact
If you have any questions or comments, please contact Veronika Cheplygina (vech (at) itu (dot) dk )
