Lung Tumor Localization from Isotropic CT Images By Three Dimensional Visualization

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Lung Tumor Localization from Isotropic CT Images By Three Dimensional Visualization

AbstractCT is the standard method to evaluate abnormalities seen on chest X-ray . Continuous images of chest are provided in a standard CT. For detection of airspace disease (such as pneumonia) or cancer. The isotropic CT images with a thickness of 0.6mm gives detailed information about the lung cavities, which is used for better surgical plan to treat lung cancer. The major challenge for the surgeons is to locate and segment the cancer present lobe. This paper presents a method for segmenting the lobe by finding the lobar fissure, visualizing the segmented lung in three dimensions, locating and analyzing the tumor using software MIMICS (Research version 17.0). This method reduces the surgical planning time for the surgeons.

Keywords— CT (Computed Tomography), lobar fissure, lungs , segmentation

I. Introduction

In India, more than 90,000 men and 79,000 women are diagnosed each year with lung cancer (carcinomas). According to the recent statistics provided by World Health Organization (WHO), around 7.6 million deaths are reported worldwide each year because of lung cancer. In 2030 it is expected to become 17 million deaths worldwide .In most cases of early-stage NSCLC (Non-small cell lung cancer), removal of a lobe of lung (lobectomy) is the surgical treatment of choice. Surgery is the most important method when compared to chemotherapy and radiation therapy in early stage of SCLC (small cell lung cancer).So for an accurate surgical planning it is very important to know about the anatomy of the lung cavities.Fig.1 shows the general anatomy of human lung. Mainly human lung is divided in to five lobes. Right lung has three lobes i.e., superior, middle and inferior lobes separated by boundaries, right oblique fissure and right horizontal fissure respectively. The left lung normally has two lobes, superior and inferior lobes separated by left oblique fissure. The lobe functions are independent to each other without any major airways or vessels crossing the lobar fissure.

CT imaging can be used to study the lobar anatomy Fig.2. A major challenge to the automatic detection of the fissures is the fact that the fissures have low contrast and variable shape and appearance in CT imagery, which sometimes makes it

Fig.1 General anatomy of human lung

difficult even for manual analysts to mark their exact location. For surgical planning the surgeon reads the stack of two dimensional CT images to identify the diseased lung lobes. CT images offer 2-D views from single viewpoint and have different shades of gray. It leads to high surgical planning time, low precision and high work load. Modern scanners allow images generated in the axial or transverse plane, orthogonal to the long axis of the body, to be reformatted in volumetric (3D) representations of structures. Detection of both acute and chronic changes in the lung parenchyma (internals of the lungs) can be done using CT. It is particularly relevant here because normal two dimensional x-rays do not show such defects. Depending on the suspected abnormality, a variety of different techniques are used. High Resolution Computed Tomography (HRCT) [1] technique is used for evaluation of chronic interstitial processes (emphysema, fibrosis, and so forth). In this technique, thin sections with high spatial frequency reconstructions are used. Often scans are performed both in inspiration and expiration. HRCT produces a sampling of the lung and not continuous images since it is normally done with skipped areas between the thin sections.

Fig.2 Original CT images

For detection of airspace disease (such as pneumonia) or cancer, relatively thick sections and general purpose image reconstruction techniques may be adequate. Contrast may also be used as it clarifies the anatomy and boundaries of the great vessels and improves assessment of the mediastinum and hilar regions for lymphadenopathy; this is particularly important for accurate assessment of cancer.

The modern computed tomography produce isotropic CT images with thickness of 0.6mm. Which gives highly detailed information of the lungs compared to clinical CT images whose thickness is 2.5-7mm.In clinical CT images 70% of the fissure details are incomplete and it is even difficult for the experts to observe them[2],[3]. In this paper it deals with isotropic CT images.

It is more important to have an efficient surgical planning for safer surgeries to treat lung cancer. The outbreak in the field of virtual reality technologies made many of the surgeons to prefer 3 dimensional visualization of lung for the surgical planning of lung cancer treatment [4], [5].

There are many conventional method of lobe segmentation. Using gray –level information depends on local pixel intensity and not on anatomic information. The fissure without sufficient contrast or near the boundary of lungs is difficult to extract [6]. Most studies have mainly focused on lung lobe segmentation and tumor localization but emphasis has not been laid on the determination of tumor volume and size [7].


Mapping of 3D data in Cartesian space is known as 3D visualization. The reason for preferring 3D visualization is , it is said that human have good visual intuition of dynamics and it is easier to communicate interesting features of the simulation to others. Unlike 2D views, 3D visualization provides multi view points, colors, stereoscopic view and no need of mental reconstruction [12] and it makes the process even more simple and interesting.


Previous approaches to lobe segmentation are divided into two classes direct and indirect. The former approaches consist of methods that search for the fissures based on gray-level information present in the data, while the other method use information from other anatomical structures to approximate the location of the fissures. And they have also used (a) lobe segmentation algorithms (b) wavelet transforms (c) Fuzzy c-mean clustering Method (d) computer aided diagnosis.


CT images of lung cancer confirmed cases was taken which is in DICOM Format.

Mimics is an image-processing package that interfaces between 2D image data (CT, MRI, Technical scanner) and 3D engineering applications. This software is used for anatomical measurements, 3D analysis, Finite Element Analysis (FEA), patient-specific implant or device design, 3D printing and surgical planning or simulation. Reconstructing solid model with mimics can minimize the difficulties encountered during contour model and also reduce the disadvantages that complex shape cannot be fully described due to contour in the past. It can generate three dimensional model directly, simplify the modeling process of extracting contours and reduce modeling time [8]. By using image segmentation in Mimics, users can select a specific region of interest from the collected medical data and have the results calculated into an accurate 3D surface model.

Fig.3- Block Diagram Of Proposed Method


Fig.3 shows the overall process of the proposed method. Here we take CT images of cancer confirmed patient and import these images to MIMICS software. Once it is imported to the software it can be viewed as top, front and side view (axial, coronal and sagittal) respectively. And the orientations for the image are set as top, bottom, left and right. One added advantage of using this software is that it is possible to have coronal, axial, sagittal and 3D view simultaneously on the screen and work on the 2D stack images and correct the mistakes simultaneously without any delay.


The histogram of an image is a plot or graph drawn between gray level values (0-255) in the X-axis and the number of pixels having the corresponding gray levels in the Y-axis. The histogram normally refers to the frequency of occurrence of voxel (volumetric pixel) values in a stack of 2-D CT slices.For a dark image, the components of a histogram will be concentrated on the dark (bright) side of the gray scale and for bright image, the histogram components will be biased towards the high side of the gray scale. The lung image consists of soft tissues which are surrounded by rib cage. The voxel values of soft tissues are different from voxel values of the rib cage. The gray values of CT images are expressed according to the Hounsfield (HU) scale. This scale exists out of 4096 values (12 bits) which is mapped on the 256 (8 bits) gray values of your display. The volumetric pixel of lung image is distributed between -825 HU (Hounsfield Unit) and 2500 HU (Hounsfield Unit). For soft tissues, it starts from

-825 HU to 225 HU and for rib cage & bones - 226 HU to 3071 HU. A window that covers the full histogram will visualize all the tissues. A narrow window allows you to better visualize subtle differences in the soft tissue or bone.

Fig.4 Histogram analysis


Threshold is used to separate objects from its background. In MIMICS software for this purpose a threshold tool is used. With this tool the area to be threshold is bone (CT). After this it is noticed that the bone tissue in the scan data becomes highlighted and its 3D model can be viewed. There are several quality options for the highlighted image like low, high and optimum quality. A higher quality setting will require longer calculation time, but will result in a more accurate 3D model. By Converting to three dimensional model, it is created in the 3D window on the bottom right of the working window Fig. To separate the bone from the artery as well as remove any floating pixels in the image, region growing method is used. The same process is continued until the lung tissues are threshold.

Fig.5 Threshold

  1. Front view (b) Top view (c) Side view (d) 3D view


Our approach to the lobar segmentation problem is to use the anatomical information provided by the lung lobe and analyze the fissures in the CT image. To identify the fissures from the CT image its contrast is adjusted. A CT imaging system consists of cross-sectional image or “slices” of anatomy. These slice process can be done either by single slice edit mask or multi slice edit. Multi slice edit interpolate set of slices with similar property Fig.6. Using this method lobe segmentation is done.

Fig.6 Multi slice edit

Until all the five lobes are segmented multi slice process is done. (Boolean operators can also be used to segment other lung lobes). 3D view of the extracted lung lobe is viewed once all the five lobes are segmented Fig.7.

Fig.7 3D View of Segmented Lobes


Lung cancer (tumor) is the uncontrolledcell growthintissuesof thelung. Cancers that begins in the lung is called primary lung cancers, arecarcinomasthat derive fromepithelial cells. Locating the accurate position of the tumor is very important for its treatment. The tumor region is spotted by checking through the 2D CT slices. With the help of dynamic region growing ROI (tumor) is selected and from each slice of CT the tumor region is interpolated. Once tumor interpolation is completely done it’s viewed in 3D. Fig.8 shows the 3D view of the tumor spotted between lobes of left lung.

Fig.8 3D Visualization (tumor spotted)

Once the tumor position is located, it is important to analyze the properties of the tumor to deliver a high dose radiation accurately while sparing normal tissues during radiotherapy. For that the tumor is extracted from the lung lobe Fig.9. Volume and surface area of the tumor is calculated.

Fig.9 Tumor Properties


The proposed system of lobe segmentation and tumor localization was developed using MIMICS Software. These results indicate promising potential in 1) lung lobe segmentation 2) tumor localization 3) 3 D view of lung and tumor 4) measuring the volume and surface area of the tumor. In the previous papers an automatic lobe segmentation framework with atlas initialization was proposed in [6]. The validation results show that some occasional errors exist due to poor image quality. ANOVA analysis indicated that the results produced by the segmentation algorithm are not significantly different from those by manual segmentation [7].


Lung cancer is one of the most difficult cancers to cure and the number of deaths that it causes is generally increasing. The early detection of lung cancer is a challenging problem for the physicians hence it is very difficult to treat. Radiotherapy is the final stage of treating the cancerous tumor. Breathing motion of lungs is the major problem in radiotherapy. Therefore it is very difficult to deliver the radiation dose exactly to the affected tissues. In order to overcome this issue a 3-D lung model was developed by the use of MIMICS software for localizing the lung cancer. To validate the result gold standard was created retrospectively from the radiologists. The experimental results were compared against the gold standard. The obtained results show that the proposed three dimensional lung model using MIMICS software can capable of locating the cancerous tumor by extraction of lobar fissures in human lungs .

Table 1. Lung tumor localization rates of inter observers and the proposed CAD system against the gold standard.

L1 - Left superior lobe L2 - Left inferior lobe R1 - Right superior lobe R2 - Right middle lobe R3 - Right inferior lobe


[1]Qiao Wei, Yaoping Hu, Gary Gelfand, and John H. MacGregor., ―Segmentation of Lung Lobes in High-Resolution Isotropic CT Imagesâ€-, IEEE Transactions on Biomedical Engineering, vol.56, no.5, May 2009.

[2] B. N. Raasch, E.W. Carsky, E. J. Lane, J. P. O’Callaghan, and E.R

Heitzman,“Radiographic anatomy of the inter lobar fissures: A Study

of 100 specimens,” Amer. J. Roentgenol., vol. 138, pp. 647–554,


[3] K. Hayashi, A. Aziz, K. Ashizawa, H. Hayashi, K. Nagaoki, and H.

Tsuji,“Radiographic and CT appearances of the major fissures,”

Radiographics,vol. 21, pp. 861–874, 2001.

[4] B.M.Hemminger, P. L. Molina, T. M. Egan, F.C.Detterbeck, K.E.Muller, C.S.Coffrey, and J. K. Lee, “Assessment of real-time 3D visualization for cardiothoracic diagnostic evaluation and surgery planning,” J. Digit.Imag., vol. 18, pp. 145–153, 2005..

[5]J.-M. Kuhnigk, V. Dicken, S. Zidowitz, L. Bornemann, B. Kuemmerlen,S. Krass, H.-O. Peitgen, S. Yuval, H.-H. Fend, W. S. Rau, and T.Achenbach, “Newtools for computer assistance in thoracic CT. Part 1. Functional analysis of lungs, lung lobes, and bronchopulmonary segments,” Radio-Graphics, vol. 25, pp. 525–536, 2005.

[6] Li Zhang, Member, IEEE, Eric A. Hoffman, Member, IEEE, and Joseph

M. Reinhardt*, Senior Member, IEEE., ―Atlas-Driven Lung Lobe

Segmentation in Volumetric X-Ray CT Imagesâ€-., IEEE Transactions on

medical imaging, vol. 25, NO. 1, January 2006.

[7] Qiao Wei, Yaoping Hu, Gary Gelfand, and John H. MacGregor., ―

Segmentation of Lung Lobes in High-Resolution Isotropic CT

Images â€-, IEEE Transactions on Biomedical Engineering, vol.56, no.5,

May 2009.

[8] Zhang Han , Cui Jianwen Fu Bin,. -Study on Three-Dimensional Finite

Element Model of Maxillary Dentition., Proceedings of the 2011 IEEE

IICME International Conference on Complex Medical Engineering May

22 - 25, Harbin, China.

[9]Al-Mayah, J. Moseley, and K.K.Brock, ―Contact surface and material nonlinearity modeling of human lungs.â€-,Phys Med Biol 53 (2008), no. 1, 305–317.

[10]Jiang S.B, ―Technical aspects of image guided respiration-gated radiation therapyâ€-.,Med Dosim 31 (2006), no. 2, 141–151.

[11]K. Hayashi, A. Aziz, K. Ashizawa, H. Hayashi, K. Nagaoki, and H. tsuji,“Radiographic and CT appearances of the major fissures,” Radiographics,vol. 21, pp. 861–874, 2001.

[12]Rene Werner,, Jan Ehrhardt , Rainer Schmidt, and Heinz Handels , ―Modeling Respiratory Lung Motion – a Biophysical Approach using Finite Element Methodsâ€- Medical Imaging 2008: Physiology, Function, and Structure from Medical Images, Proc. of SPIE Vol. 6916, 69160N, (2008).

[13]Keall P.J, S. Joshi, S.S.Vedam, J.V.Siebers, V.R. Kini, and R. Mohan, ―Four-dimensional radiotherapy planning for dmlc-based respiratory motion trackingâ€-., Med Phys 32 (2005), no. 4,942–951.

[14] Yoping Hu, ―The role of three dimensional visualization in surgical

planning of treating lung cancer,â€- 27th annual IEEE conference on

engineering in medicine and biology society. pp. 646-649, 2006.

[15]Zehtabian.M, R.Faghihi, et al.,―A fast model for prediction of respiratory lung motion for image-guided radiotherapy : A feasibility studyâ€-, Iran. J. Radiat. Res., 2012.

[16] Zhang Han , Cui Jianwen Fu Bin,. -Study on Three-Dimensional Finite

Element Model of Maxillary Dentition., Proceedings of the 2011 IEEE

IICME International Conference on Complex Medical Engineering May

22 - 25, Harbin, China.

[17] Y. Hu and R. A. Malthaner, “The feasibility of three-dimensional

Displays of the thorax for preoperative planning in the surgical treatment

of lung cancer,” Eur. Cardio-Thorac. Surg., vol. 31, pp. 506–511, 2007.