Archives

  • 2019-10
  • 2020-03
  • 2020-07
  • 2020-08
  • br Start br Load LIDC

    2020-08-18


    Start
    Load LIDC-IDRI file Load GT
    Group the data 5 Azacytidine on
    “Series Instance
    UID”
    Match UID data with
    UID ground truth
    N Match?
    The first is elimination of the bed-mat and the second is elimina-tion of the unrelated organs. Extraction of the lung area in 2D can be done with Otsu’s thresholding and morphological filling holes (Ta and Aybars, 2015; Shen et al., 2015). Fig. 5 shows an example of lung area extraction. Finding the largest area in the labelled 3D is conducted to eliminate unrelated areas containing bed-mat and other organs that have low intensities. Fig. 6 depicts the process of lung extraction.
    2.4. Tracheal extraction
    The lung area extracted in the previous stage still leaves the tra-chea region. Several previous works have been conducted to seg-ment airway accurately by region growing (Yim and Hong, 2008; Magalhães Barros Netto et al., 2012; de Carvalho Filho et al., 2014). To simplify and reduce computational, 3D controlled label-ing is applied. Controlled labeling is determined by circularity and area of each labeled object.
    Fig. 5. The lungs extraction in 2D processing. (a) The original CT image; (b) after using Otsu’s thresholding; (c) find the lung area.
    N Juxta-pleural or vascular nodule?
    Y N
    3< diameter nodule in ground truth
    Y
    Sort the 3D CT image
    base on
    “NumberInstance”
    Convert HU to CT
    scale
    Store the selected
    CT
    Finish  Start
    Load the selected 3D CT
    image
    Threshold using Otsu in
    slice i
    Fill the holes using
    morphology
    Take the hole that was
    filled as foreground and
    the other as background
    Label in 3D and select the
    label that has the maximum area
    Store the 3D lung binary
    image
    N Is it the last 3D CT?
    Y
    Finish
    Fig. 4. The flowchart of data selection.
    Fig. 6. The flowchart of lung extraction.
    Please cite this article as: R. Nurfauzi, H. A. Nugroho, I. Ardiyanto et al., Autocorrection of lung boundary on 3D CT lung cancer imagesq, Journal of King Saud University – Computer and Information Sciences, https://doi.org/10.1016/j.jksuci.2019.02.009
    4 R. Nurfauzi et al. / Journal of King Saud University – Computer and Information Sciences xxx (xxxx) xxx
    3D controlled labeling requires an initial point to process. The initial point can be identified by finding the three largest areas. The areas are obtained by scanning all CT slices from the top to half of the number of slices. The first largest area must be greater than 1000 pixels. Moreover, the third largest area must be greater than 100 pixels and has circularity of more than 0.85. The initial area in the slice is shown in Fig. 7. The process to find the initial point is shown in Fig. 8. If the starting point is not found in half of the num-ber of slices, this data is suspected as a reverse CT sequence.
    The cutting intensity in the range 0–170 of the original labeled is applied to make the binary area for tracking (Zhou et al., 2014). 3D controlled labeling will track in all labeled region except is stopped by rules. The rules are that the circularity value in labeled
    region should be more than 0.8 and the differential area from the intersected area in previous slice should be less than 0.3. 3D con-trolled labelling process is shown in Fig. 9.
    2.5. Separation of Lung fusion
    The lungs have two parts, right and left. Mostly in CT images, that parts are close together and even appear to stick together. Only thin membrane is separated them. Thin membrane is called pleural cavity. The occurred lung fusion in binary imagery is caused by the inadequacy of the method used to extract the lungs. The advanced method is necessary applied to separate them.
    In this study, a 2D multi-threshold is proposed for separating the lung fusion. There are three main steps, namely detection of lung fusion, determination of the region of interest (RoI) of lung fusion and separation of lung fusion. This whole step is illustrated in Fig. 10.
    Fig. 7. The lung slice when initial point is found (yellow area). h is half of the number of slices.
    Start
    Load the 3D lung
    binary image
    Calculate area
    and circularity in
    labelled slice i
    Select the three
    largest labels
    N
    N N
    The third largest with
    Suspected as circularity value
    a reverse order CT Y
    Save the initial
    trachea position
    Eliminate the
    labels in B