International Journal of Scientific & Engineering Research Volume 8, Issue 10, October-2017 1632 ISSN 2229-5518 IJSER © 2017 http://www.ijser.org Detection of Lung Nodules in CT Images Using Features fusion and Genetic Algorithm Hanan M. Amer,Hamdi A. A. Elmikati,Fatma E.Z. Abou-Chadi,Sherif S. Kishk, and Marwa I. Obayya Abstract— The aim of this study is to increase the accuracy of early detection of pulmonary nodules through the development of a Computer-Aided Detection (CAD) system. A comparative study of performance for the most commonly used techniques in feature extraction and classification was per- formed to identify the technique that gives the highest accuracy and lowest false positives. This study was conducted on the pulmonary nodule candi- dates directly without removing the blood vessels to design a precise automatic detection system to detect the pulmonary nodules in early stage. The main significant features of the pulmonary nodule candidates are extracted by using four feature extraction technique: Histogram of Oriented Gradients (HOG) features, Computerized Tomography (CT) Value Histogram (VH) features, texture features of Gray Level CO-Occurrence Matrix (GLCM) based on wavelet coefficients, and statistical features of first and second order. To make use of the extracted features, a feature fusion technique was used to concatenate the extracted features together and select features in a new hybrid feature vector. A Genetic Algorithm (GA) search based on the classifica- tion accuracy rate (CAR)of the utilized classifier was also applied to the hybrid feature vector. To get the highest classification accuracy, three classifiers were selected and their performance was compared. These are: Artificial Neural Network (ANN), Radial Basis Function Neural Network (RBF-NN) and Support Vector Machine (SVM). Three parameters were used to compare the classifier performance: the classification accuracy rate (CAR) , the sensi- tivity (S) and the Specificity (SP). The results have shown that using the selected hybrid features vector and the SVM classifier gives the highest CAR of 99.6% and a 0.008 false positive per scan. Index Terms— Computer-aided detection, computed-tomography, discrete wavelet transform, principal component analysis, CT value histogram, histogram of oriented gradients, gray level co-occurrence matrix. ——————————  —————————— 1 INTRODUCTION S a result of the high rates of air pollution and the spread of smoking in recent years, lung cancer has be- come one of the most important diseases that pose a great threat to humanity because of the high rates of infection and the difficulty of treatment and rapid spread. Developing new techniques in the treatment and early detection of this disease has become the concern of scientists in medical fields [1], [2]. As early detection of this disease increases the chance of sur- vival of the patient for a period of up to 5 years by up to a per- centage of 70%, as well as it increases the chance of success of treatment whenever diagnosed in the early stages, this led to the increasing importance of work on the development of early detection techniques [2], [3]. One of the most important techniques used in the diagnosis of lung cancer is Computerized Tomography (CT) of the pa- tient's chest. It is one of the most accurate methods, because it is a lung imaging on many sections, which results in this examina- tion number large images, enabling radiologists and physicians to examine all parts of the lung [4].But as a result of the increase in the number of images resulting from this examination in ad- dition to the use of low radiation doses during the examination to protect the patient from the risk of exposure to large amounts of radiation, all made the examination of these images by a ra- diologist difficult and onerous task [5].This motivated scientists to design computerized systems that process and analyze these images and allow automatic determination of the presence of pulmonary nodules. These systems are known as Computer- Aided Detection (CAD) systems [6], [7]. In general, any CAD system for the automatic detection of pulmonary nodules is composed of four main stages: a prepro- cessing step for image contrast enhancement and noise reduc- tion. Then the automatic segmentation stage that aims to extract the human's lung area followed by a feature extraction proce- dure of the pulmonary nodule candidates in the digital CT im- ages and the final stage is the classification[8].Fig. 1 demon- strates the basic stages of any CAD system. The present paper presents a detailed description of the third and fourth stages: feature extraction and classification to detect accurately the pulmonary nodules. It is a continuation of our work which aims the development of a complete automatic CAD system for the detection of lung cancer in its early stages. The first and second stages of the system which are image pre- processing step and automatic segmentation system have been completed and published previously in [9]. The performance of the complete system is evaluated through the computation of A ———————————————— • Hanan M. Amer is assistant teacher in department of electronics and com- munications, faculty of engineering, Mansoura University, Egypt. E-mail: hanan.amer@yahoo.com • Hamdi A. A. Elmikati is professor in department of electronics and com- munications, faculty of engineering, Mansoura University, Egypt. • Fatma E.Z. Abou-Chadi is professor in department electrical engineering, faculty of engineering, the British University in Egypt. • Sherif S. Kishk is Assistant Professor in department of electronics and communications, faculty of engineering, Mansoura University, Egypt. • Marwa I. Obayya is Assistant Professor in department of electronics and communications, faculty of engineering, Mansoura University, Egypt. IJSER