Fissure Segmentation
Jingbin Wang, Margrit Betke, Jane P. Ko

Approach

Figure 1: Pulmonary fissure a.) In 1.25mm thickness data,
b.) In 5mm thickness data, diagnostic CT

We describe a new approach to detect the fissures between the lobes of the lungs in computed tomography (CT) scans. We focus on the detection of the major ssures in diagnostic CT. Our approach is to first find a region containing the fissure and then extract the fissure within this region by analyzing local features. We developed a "marching-bar method" to segment the region containing the fissure. After this region is computed for an initial slice, fissure extraction in consecutive slices is performed using the location information inherited from previous slices.

The framework of our two-step algorithm is shown in Fig. 2. In the initial step, an automatic thresholding method is used to produce a binary image of the lung parenchyma, which was segmented by a previously described method [2]. We developed a "marching-bar method" to mark the fissure ROI by sweeping a "bar" to the lateral and medial borders of the lungs. In this process, the length of the bar can adaptively change at each step, so that the vertically longest local region that does not contain large gradient changes is covered by the bar. From the given fisure ROI, the fissure is then segmented.

Figure 2: Method Overview

The fissure ROI is detected in two phases -- a thresholding phase and a marching-bar phase. In the thresholding phase, the grayscale values of the pixels in the segmented lungs are automatically converted into binary values. The resulting binary image contains white pixels for lung structures, such as bronchi, vessels, and nodules, and black pixels for the fissures and the aerated lung. The marching-bar phase of our algorithm is then applied to the binary image of the lungs, which is shown as Fig.3a.


Figure 3: Marching Bar Method a.) Fissure ROI Detection b.) Fissure path extraction inside the fissure ROI

Once the fissure ROI is computed for some slice, additional local feature analysis is performed to determine the location of the fissure inside the region, which is illustrated as Fig. 3b.

Preliminary Results

We tested a CT scan with a resolution of 512 * 512 pixels per slice and a slice thickness of 10 mm with a change in slice thickness to 5 mm through the hila. Some of our test images and results are shown in Fig. 4.


Figure 4: Fissure Segmentation -- Premilinary Result

Read our paper for more detailed description

Ongoing Work

Our method has shown its promising result on the diagnostic CT data. We are currently extending it to a general framework to deal with various CT scan data of different resolutions. And our final objective is to create 3D lung lobe structures based on the segmented fissure result. This 3D lobe information could be very useful in many fields, such as anatomical analysis of lung structure, visualization, and nodule registration of lung.

Figure 6: Fissure Segmentation Framework

References

M. S. Brown, M. F. McNitt-Gray, J. G. Goldin, J. W. Suh, R. D. Sayre, and D. R. Aberle. Patient-specic models for lung nodule detection and surveillance in CT images. IEEE Trans Med Imag, 20(12):1205-1208, December 2001.
J. P. Ko and M. Betke. Chest CT: Automated nodule detection and assessment of change over time--preliminary experience. Radiology, 218(1):267{273, 2001.
M. Kubo, Y. Kawata, N. Niki, K. Eguchi, H. Ohmatsu, R. Kakinuma, M. Kaneko, M. Kusumoto, N. Moriyama, K. Mori, and H. Nishiyama. Automatic extraction of pulmonary ssures from multidetector-row CT images. In Proc. IEEE International Conf. on Image Processing (ICIP'01), pages 1091-1094, 2001.
L. Zhang and J. M. Reinhardt. Detection of lung lobar fissures using fuzzy logic. In C.-T. Chen and A. V. Clough, editors, Medical Imaging 1999: Physiology and Function from Multidimensional Images, SPIE Proceedings Vol. 3660, pages 188-198, San Diego, CA, February 1999.

[ Jingbin's Home | IVC Home ]

Last Updated 01/14/03 byJingbin Wang

Fissure Segmentation Jingbin Wang, Margrit Betke, Jane P. Ko