Desktop version

Home arrow Health

Lung and Pleura

With the development of CT technology, complete volumetric chest images can be acquired over a single breath hold. With increasing resolution, the data load has substantially increased. A lung CAD system may help radiologists to deal with these data loads more effectively. Accurate lung segmentation is fundamental for quantitative analysis of lungs by CAD systems.

  • 3 Understanding Medical Images Based on Computational Anatomy Models
  • 221

The lung is divided into lobes by interlobar fissures that are potential spaces lined by visceral pleura. Extraction of the interlobar fissures is essential, because it make it possible for a lung CAD system to determine the lung lobe of the pathology and calculate volume rate of pathological region in the lung lobe. Clinically, extraction of the interlobar fissure is often important to determine whether a disease affects one or more lobes, when lobar resection is considered.

Anatomy of Lungs and Interlobar Fissures The right and left lungs are covered by a tightly attached layer of visceral pleura. The veins, arteries, airways, and lymphatics comprising the lung roots attach to the mediastinum in the center of the chest (Fig. 3.56a). The entire thoracic cavity is lined by an outer layer of pleura called the parietal pleura. A potential space is located between these two pleural layers and can be enlarged by fluid (pleural effusion), air (pneumothorax), or disease. The lungs consist of sections called lobes. The left lung usually contains two lobes, the upper and lower, while the right lung usually consists of three, the upper, middle, and lower lobes. An accessory right upper (azygous) lobe is not uncommon, and there are other variations that are less common. These lobes are separated by interlobar fissures lined by visceral pleura. The fissures can be of varying depth, extending down to the lung root, or incompletely dividing the parenchyma. Both lungs have a major fissure dividing the upper and lower lobes, and the right lung has a minor fissure dividing the upper and middle lobes. Examples of lung lobes are shown in Fig.3.57. The lobes are divided into pulmonary segments, and each segment contains multiple secondary pulmonary lobules divided by thin connective tissue septa. In pathological cases, such as pleural effusion, lung cancer, interstitial pneumonia, and severe emphysema, features of the lungs and pleura are usually altered. Pleural effusion in a CT image is shown in Fig. 3.56b.

Lung Segmentation Many methods for automatically extracting the lung regions based on 3D CT volumes have been proposed [263]. As visualized on CT images,

CT images of (a) normal chest and (b) right pleural effusion. Red triangles point to interlobar fissure

Fig. 3.56 CT images of (a) normal chest and (b) right pleural effusion. Red triangles point to interlobar fissure

Fig. 3.57 Image of lungs viewed from outside. RUL right upper lobe, RML right middle lobe, RLL right lower lobe, LUL left upper lobe, and LLL left lower lobe. (a) Right lung, (b) left lung

the lung parenchyma, which is mostly air, is dark. This contrast between lung and surrounding tissues is the basis of most segmentation methods [11, 45, 122, 147, 329]. In lung segmentation methods, air is extracted by gray-level thresholding. Therefore, the lungs are identified by imposing restrictions on size and location. Alternatively, the lung volumes can be determined by region-growing segmentation from the trachea. The trachea is recognized as 2D circular air regions in the first slices of the scan or as a 3D tubular air region located centrally in the upper part of the scan. The main lung volume is separated into left and right lungs. The trachea and main stem bronchi are removed. Morphological processes work out filling holes and smooth borders of segmented lungs.

Such methods are known to be simple and effective for normal cases. However, they often fail to extract lungs affected by pathologies, especially when the pathologies involve the pleura. Pleural effusion, like that illustrated in Fig. 3.56b, has higher density than normal lung tissue, and therefore segmentation methods using contrast fail.

For pathological cases, segmentation methods often use the shape of the lung. Sluimer et al. proposed a registration-based approach in which a shape template is registered to an input CT volume [262]. They achieved significant improvements in the segmentation of lungs with pathologies. Kido and Tsunomori proposed another registration-based method using a template obtained from normal cases [153]. Two- step matching improved the performance in a case with severe pleural effusion. Hua et al. proposed a method that combines the classification process with a graph- search algorithm [123]. The method has been shown to be effective in cases with pathology. Nakagomi et al. proposed a graph cut-based method that can take into account the multiple shapes generated from an SSM of the lungs. The method used in energy term introduces neighboring structures of the lungs, e.g., the aorta and the body cavity [210].

Interlobar Fissure Extraction The interlobar fissures are very thin and low contrast on CT images. Simple methods such as gray-level threshold cannot extract them. Many extraction methods have been proposed. Kubo et al. proposed a method that extracts sheet shapes by morphology operation from the emphasis of 2D linear shadow images [166]. Saita et al. suggested a method that uses blood vessel information [241]. This method classifies lobar blood vessels, whereas the region containing the interlobar fissure is identified on the basis of the 3D distance from the lobar blood vessel. The interlobar fissures are extracted by the emphasized sheet shadow from the identified region. Zhang’s [328] method extracts the major fissures using an anatomic pulmonary atlas. A ridgeness measure is applied to the original CT image to enhance the fissure. A fuzzy reasoning system is used in the fissure search to analyze information from three sources: the image intensity, an anatomic smoothness constraint, and the atlas-based search initialization. Van Rikxoort’s method uses supervised enhancement filters [297]. These filters, which enhance interlobar fissures and suppress others, are constructed using training data. Ukil and Reinhardt proposed a method based on the information provided by the segmentation and analysis of the airway and vascular trees [295]. An ROI is generated using this information, and the interlobar fissures are extracted by enhancement using a ridgeness measure in the ROI. Pu et al. pointed out existing methods that were problematic in the presence of pathologies and susceptible to interindividual differences; they proposed a method using the membranous properties of the fissures. The method creates polygons by the marching cube algorithm. Surface shapes are extracted using multiple Laplacian smoothing from polygons [231]. The interlobar fissures are extracted by normal vectors of surfaces that are made by extended Gaussian image. This interlobar fissure is filled blank spaces by average of normal vector of the surface [232]. Agarwala et al. proposed an atlas-based segmentation approach [2]. An atlas is a set of two images: the intensity image and its segmentation. Because segmentation obtained using an atlas-based approach may have local errors because of local failures of the image registration algorithm, they applied a local version of selective and iterative methods for performance level estimation that uses local weights for fusion of the input segmentations. Lassen et al. employ a method that uses information on blood vessels and sheet shapes extraction using a Hessian matrix [169]. Matsuhiro proposed a method that uses features of the interlobar fissures’ film shape that can extract interlobar fissures from pathological cases, e.g., lung cancer, interstitial pneumonia, and severe emphysema [189]. This method contains three extraction phases: coarse extraction, fine extraction, and correction. Coarse extraction enhances images by 4D curvature. Film shapes are extracted from contrast-enhanced images. The interlobar fissures’ shape extraction parameters, e.g., angles and sizes, are trained by training case data. In these data, interlobar fissures are predetermined manually. Coarse interlobar fissures are extracted from film shapes by these parameters. Fine extraction is implemented in iterative enhancement around fissures. Correction is implemented in interpolation of interlobar fissures by normal vectors. The extraction results are illustrated in Fig. 3.58.

This section describes lung segmentation and interlobar fissure extraction methods. Developing the performance of those methods is expected. Methods that segment pulmonary segment and secondary pulmonary lobule are expected to be developed, too.

Extraction results of interlobar fissures. (a) Interstitial pneumonia, (b) lung cancer, (c) hypersegmentation

Fig. 3.58 Extraction results of interlobar fissures. (a) Interstitial pneumonia, (b) lung cancer, (c) hypersegmentation

< Prev   CONTENTS   Source   Next >

Related topics