LIDC-IDRI-DRR

Digitally Reconstructed Radiograph

The LIDC-IDRI Digitally Reconstructed Radiograph (LIDC-IDRI-DRR) is a novel dataset based on LIDC-IDRI to improve ribs segmentation through unsupervised domain adaptation. The Lung Image Database Consortium image collection (LIDC-IDRI) consists of diagnostic and lung cancer screening thoracic computed tomography (CT) scans with marked-up annotated lesions. It is a web-accessible international resource for development, training, and evaluation of computer-assisted diagnostic (CAD) methods for lung cancer detection and diagnosis. Initiated by the National Cancer Institute (NCI), further advanced by the Foundation for the National Institutes of Health (FNIH), and accompanied by the Food and Drug Administration (FDA) through active participation.

We proposed a novel methodology for rib cage segmentation in Chest X-Rays (CXRs) by using Unsupervised Domain Adaptation (UDA) from DRRs. Our method leverages the capabilities of Conditional DA (CoDA) to transfer the knowledge learned from synthetically flattened CT-scans to 2D CXRs.

Our pipeline uses higher dimensional 3D data to acquire two sets of flattened 2D images: 1) DRRs that visually resemble real CXRs – serving as training samples for rib segmentation; and 2) bone segmentation semantic maps that can be curated in order to become pixel-level rib segmentation labels. Our rib segmentation pipeline is depicted below:

In this pipeline we make two distinct uses of Conditional DA:

1. CoDA_Lungsfor acquiring lung segmentation predictions Y^ˆ^Lungs_Afor DRRs from labeled CXR source datasets (X_Band Y_B) – highlighted in blue in Figure above;

2. CoDA_Ribsfor translating the knowledge from the filtered rib segmentation masks Y^Ribs_Afor DRRs in order to use them in CXR data (X_B), resulting in the prediction Y^ˆ^Ribs_B– as delineated in orange in Figure above.

Image samples from LIDC-IDRI-DRR

Image

Label

Image

Label

Our dataset can be found in LIDC-IDRI-DRR

Our pre-trained model is available in our GitHub

Please refer to the paper below:

Hugo Neves Oliveira, MSc.; Virginia F Mota, PhD.; Alexei C Machado, PhD.; Jefersson A dos Santos, PhD. From 3D to 2D: transferring knowledge for rib segmentation in chest X-rays. Pattern Recognition Letters, 2020. https://doi.org/10.1016/j.patrec.2020.09.021

Sample of qualitative results in the datasets with labeled test set (JSRT and OpenIST). Visual analysis over the samples further reinforce the tendency to overestimate rib pixels of DNNs pretrained in DRR synthetic samples, while Conditional DA are more conservative in predicting rib pixel labels. It is also evident from the overlayed prediction probability map that the pretrained models have a much sharper decision boundary than CoDAGANs. That is, the deep baseline method predicts either rib or background pixels with more confidence than Conditional DA, which also results in rougher segmentation boundaries.

JSTR

OpenIST

For unlabeled datasets, Montgomery and Shenzhen sets, NIH Chest X-Ray 8, PadChest and NLMCXR, we could also show how CoDAGANs have a more precision segmentation.

Montgomery

NIH

NLMCXR

Padchest

Shenzhen

Probability maps can be found in CoDAGANs probability maps.

Baseline probability maps can be found in Baseline probability maps.

We provide probability maps to the following datasets: JSRT, OpenIST, Montgomery and Shenzhen sets, NIH Chest X-Ray 8, PadChest and NLMCXR.

Please refer to the paper below:

Hugo Neves Oliveira, MSc.; Virginia F Mota, PhD.; Alexei C Machado, PhD.; Jefersson A dos Santos, PhD. From 3D to 2D: transferring knowledge for rib segmentation in chest X-rays. Pattern Recognition Letters, 2020. https://doi.org/10.1016/j.patrec.2020.09.021

BSE-JSRT Labels

We also contribute to label the Bone Supression dataset. This is a major contribution because the granularity of a segmentation label requires massive manual labor and many researchers can benefit from it to train their own methods.

Labels can be found in X-ray bone shadow supression labels .

Please refer to the paper below:

Hugo Neves Oliveira, MSc.; Virginia F Mota, PhD.; Alexei C Machado, PhD.; Jefersson A dos Santos, PhD. From 3D to 2D: transferring knowledge for rib segmentation in chest X-rays. Pattern Recognition Letters, 2020. https://doi.org/10.1016/j.patrec.2020.09.021

References

Users of this data must abide by the Creative Commons Attribution 3.0 Unported License .

If you want to use this dataset, labels and probability maps, please contact oliveirahugo@dcc.ufmg.br and refer to the paper below:

Hugo Neves Oliveira, MSc.; Virginia F Mota, PhD.; Alexei C Machado, PhD.; Jefersson A dos Santos, PhD. From 3D to 2D: transferring knowledge for rib segmentation in chest X-rays. Pattern Recognition Letters, 2020. https://doi.org/10.1016/j.patrec.2020.09.021

Lung Image Database Consortium image collection (LIDC-IDRI)

Data Citation
- Armato III, SG; McLennan, G; Bidaut, L; McNitt-Gray, MF; Meyer, CR; Reeves, AP; Zhao, B; Aberle, DR; Henschke, CI; Hoffman, Eric A; Kazerooni, EA; MacMahon, H; van Beek, EJR; Yankelevitz, D; Biancardi, AM; Bland, PH; Brown, MS; Engelmann, RM; Laderach, GE; Max, D; Pais, RC; Qing, DPY; Roberts, RY; Smith, AR; Starkey, A; Batra, P; Caligiuri, P; Farooqi, Ali; Gladish, GW; Jude, CM; Munden, RF; Petkovska, I; Quint, LE; Schwartz, LH; Sundaram, B; Dodd, LE; Fenimore, C; Gur, D; Petrick, N; Freymann, J; Kirby, J; Hughes, B; Casteele, AV; Gupte, S; Sallam, M; Heath, MD; Kuhn, MH; Dharaiya, E; Burns, R; Fryd, DS; Salganicoff, M; Anand, V; Shreter, U; Vastagh, S; Croft, BY; Clarke, LP. (2015). Data From LIDC-IDRI. The Cancer Imaging Archive. http://doi.org/10.7937/K9/TCIA.2015.LO9QL9SX
Publication Citation
- Armato SG 3rd, McLennan G, Bidaut L, McNitt-Gray MF, Meyer CR, Reeves AP, Zhao B, Aberle DR, Henschke CI, Hoffman EA, Kazerooni EA, MacMahon H, Van Beeke EJ, Yankelevitz D, Biancardi AM, Bland PH, Brown MS, Engelmann RM, Laderach GE, Max D, Pais RC, Qing DP, Roberts RY, Smith AR, Starkey A, Batrah P, Caligiuri P, Farooqi A, Gladish GW, Jude CM, Munden RF, Petkovska I, Quint LE, Schwartz LH, Sundaram B, Dodd LE, Fenimore C, Gur D, Petrick N, Freymann J, Kirby J, Hughes B, Casteele AV, Gupte S, Sallamm M, Heath MD, Kuhn MH, Dharaiya E, Burns R, Fryd DS, Salganicoff M, Anand V, Shreter U, Vastagh S, Croft BY. The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): A completed reference database of lung nodules on CT scans. Medical Physics, 38: 915--931, 2011. DOI: https://doi.org/10.1118/1.3528204
TCIA Citation
- Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P, Moore S, Phillips S, Maffitt D, Pringle M, Tarbox L, Prior F. (2013) The Cancer Imaging Archive (TCIA): Maintaining and Operating a Public Information Repository, Journal of Digital Imaging, Volume 26, Number 6, pp 1045-1057. DOI: https://doi.org/10.1007/s10278-013-9622-7

The authors acknowledge the National Cancer Institute and the Foundation for the National Institutes of Health, and their critical role in the creation of the free publicly available LIDC/IDRI Database used in this study.

Google Sites

Report abuse