Classification of pulmonary nodules using Novel Z with Tilted ZLocal Binary Pattern (Z?TZLBP) Abstract Inthis paper , a novel feature vector named Z with Tilted Z Local Binary Pattern (Z?TZLBP)is proposed for extracting pulmonary nodule image features powerfully. The goalis to reduce LBP’s complexity by reducing the size of the feature vector. Inthe proposed work by dividing the vicinity pixels in to non overlapped groupsof Z and TZ(Tilted Z) further texture features of pulmonary images areextracted for feature extraction. The Single Classifier KNN has been used withdifferent distance metrices for classification purpose.
Metric Accuracy andF-measure is used to evaluate the performance of the proposed system. Keywords Classification , Pulmonary Image , FeatureExtraction , Active Contour , KNN 1 Introduction Pulmonary images are very muchimportant to detect lung diseases using CT imaging modality for the assessmentof pulmonary nodules. The feature types of the pulmonary nodule in CT imagesare important cues for the malignancy prediction 1-2, diagnosis and advancemanagement 3-4.The texture features of nodule solidity and semanticmorphology feature of speculation are critical to differentiate of pulmonarynodules and other subtypes. Meanwhile other semantic features calcificationpattern, roundedness, margin clearness are shown to be helpful for theevaluation of nodule classification.
The nodule may be found in bronchial tubesor outside of bronchial tube. If the nodule<=3mm the detection of malignancyis difficult. The determination of clinical characteristics may differ frompatient to patient and also depends on experience of the observer. Computer aided diagnosis is anassistive software package to provide computational diagnostic references for the clinical image reading anddecision making support. The histogram features for the high level textureanalysis helped to extract nodule feature. Ciompi et al5 developed the bag offrequencies descriptor that can successfully distinguished 51 spiculatednodules from the other 204 non-spiculated nodules.
However the mapping from thelowlevel image features toward the high level semantic features in the domainof clinical terms is not straight forward task. This semantic featureassessment maybe useful for clinical analysis. The Lung Image DatabaseConsortium (LIDC) dataset for its rich annotation database supports thetraining and testing CAD scheme6-7. The nodules diameters larger than 3mm are further rated byradiologist referred with semantic features of “subtlety”, “calcification”, “spericity”, “margin”,”speculation”, “texture”, “lobulation”, “internalstructure”and “malignancy”8. Absorption and scattering oflight rays are the two major issues that cause reduced quality of images.Several methods have been proposed to enhance the quality of the pulmonaryimages.
Histogram equalization technique , Contrast stretching methods capableto enhance the image quality. Contrast Limited Adaptive Histogram Equlization(CLAHE) has been applied to improve the image contrast. Otsu’s adaptivethresholding method form image segmentation has been effective for manyapplications. This provides brightbackgrounds for images. Various thresholding techniques such as Local, Globaland Multilevel thesholding have been applied for the segmentation of pulmonary nodules images.
The texture featuredescriptor that has been widely used for image classification is Local BinaryPattern. Pican et al. have used GLCM’s twenty four types features for extraction and for each image suitablefeatures have to be chosen for extraction.
Hence there is a need of efficient feature descriptor forclassification process. In past years Neural Network is performed for classification results time consuming.K-Nearest Neighbor as classifier with Euclidean distance used to classifynodules. Padmavathi et al15 have classified images using probalisitic neuralnetwork which gives better results than SIFT algorithm with three classes ofdataset.
Eduardo et al have classifiedimages using nine machine learning algorithms such as: Decision Trees, RandomForest, Extremely Randomised Trees ,Boosting, Gradient Boosted Trees, Normal Bayes Classifier,ExpectationMaximization, NN and SVM. Bhuvaneswari.P et al16 have classified coral andtextures using KNN by considering K=1 and the accuracy reported as 90.35%.