Abstract– Lung Cancer is the most deadly cancer. Early detection of the disease can improve survival rate. Automation oflung nodules detection aid radiologists in quickly and accurately diagnosingthe disease.
Developing computer aided diagnosis (CADx) systems for lung canceris a challenging task. Several components make up CADx. One of the mostsignificant components is lung segmentation, an essential prerequisite to efficientlydetect and classify lung nodules. Lung segmentation is the process ofsegregating lungs fields from other tissues in the CT image.
Conventionalmethods for lung segmentation either do not accurately segments normal andabnormal lungs or rely heavily on user generated features for the lungs. Deeplearning has outperformed other methods in image processing tasks. Recently anew architecture has been proposed and implemented exclusively for medicalimages to solve this problem namely U-Net convolutional network. In this study u-nethas been implemented on lungs dataset consisting of 267 CT images of lungs andtheir corresponding segmentation maps. The accuracy and loss achieved is .Considering the computational constraints the results obtained are state of theart.Keywords – U-Net convolutionalnetwork; Lung Parenchyma; Segmentation methods; Thoracic CT Scans; I. INTRODUCTIONGlobally, lungcancer is the leading cause of cancer mortalities.
There were 1.69 milliondeaths in 2015 due to lung cancer 1. Early detection of lung cancer increasesthe survival rate, but is like searching for needle in the haystack. Abnormalsmall, round or oval shaped growth in the lung called lung nodules may be thefirst sign of lung cancer and detection of those is very exhaustive given thecomplex structure of lungs. Computed Tomography (CT) is an important diagnosticmodality to detect lung nodules. The automation of detection and diagnosis oflung nodules benefits both the radiologists and patients.
Accurate detection oflung nodule at an early stage leads to proper treatment and saves patientslife. There are several computer aided diagnosis (CAD) systems developed overthe years to assist in the diagnosis of lung cancer. CAD system componentsinclude lung segmentation, nodule detection and segmentation, false nodulereduction, nodule classification. Each component is complicated in its own way.The significance of segmentation ofthe lungs from chest CT scans has been elaborated in 2. Lung segmentation isa prerequisite for the subsequent automated analysis of lung nodules since itallows for the estimation of lung volumes and detection and quantification ofabnormalities within the lungs.
In case of erroneous lung segmentation, findingsmight be missed or findings outside the lungs might be included in the analysis.The importance of accurate lung segmentation for the automated detection ofnodules is illustrated in 5. Their experiments have showed that accuracy ofnodule detection has increased when lung segmentation has been applied.
A naivelung segmentation algorithm was applied to 60 scan, 17% of nodules were notdetected as a consequence of improper lung segmentation. Another lungsegmentation algorithm improved the results and only 5% of nodules were notdetected.The task of lungsegmentation is challenging because of the complexity in the lung region and theexistence of similar density structures, such as arteries, veins, bronchi andbronchioles, and the use of different scanning devices with different scanning protocols3.
In general, the existing techniques for lung segmentation can be classifiedinto different categories based on: a) Intensity. Eg. Thresholding. b) Region.
Eg. Region Growing. c) Shape. Eg. Sobel. d) Edge. Eg. Wavelet Transform.
e) Machine Learning. Eg. SVM. Theobjective of this work is to apply unetsegmentation network to extract lung fields from the CT scans. II. RELATED WORKAutomatic lung segmentation has always been achallenging task. Several algorithms have been proposed that address theproblem of accurately extracting lung region from the CT images.
Conventionalmethods mostly rely on the attenuation values of different areas on the CT. Thefact that similar regions have same intensity helped the development of apopular intensity based segmenation method namely thresholding. Tremendousnumber of research papers has developed thresholding methods with manyvariations. Traditional algorithms are two-dimensional and process each axialsection of the scan separately. 3D image processing of CT scans has beendeveloped that take into account height, width and depth of an image. Most of the methods for lung segmentation begin bydetermining the lung fields using optimal gray-level thresholding and connectedcomponents or region growing method. The detected lung region contains tracheaand bronchi which has to be removed from the image. When the lungs are joined in the anterior orposterior junctions, they have to be separated to obtain the left and rightlung regions only.
Post processing include applying of the morphologicaloperations on the segmented image to smoothen the borders and fill the gaps.The literature of lung segmentation include numerous papers which are describedbelow: Hu et al (2001) were the first to publish a threshold-based lungsegmentation method based on the method described above. Ukil and Reinhardt (2005) improvedupon the method proposed by Hu et al (2001) by introducing a smoothing atthe mediastinal area based on the airway tree to guarantee consistency amongsegmentations of different subjects and intra-subject over time. Sluimer et al (2005) and vanRikxoort et al (2009a) describe 3D threshold-based methods largely based onthe method of Hu et al (2001).
Sun et al (2006) presented a 3D method for thesegmentation of the lungs from thick-slice CT images. First, a preprocessingwas applied in which the signal-to-noise ratio was improved by applying ananisotropic filter, followed by a wavelet transform-based interpolation methodto construct 3D volume data. In these 3D volume data,the lungs were obtained by regiongrowing using gray-value, homogeneity and gradientmagnitude as input. Cavities inside theresulting lung region were filled using morphologicalclosing.