Artificial Neural Network-Based Classification System for Lung Nodules on Computed Tomography Scans Emre Dandﾕl 1,3 , Murat Çakﾕro÷lu 2 , Ziya Ekúi 3 , Murat Özkan 3,4 , Özlem Kar Kurt 5 , Arzu Canan 6 1 Bilecik Vocational High School, Bilecik ùeyh Edebali University, Bilecik, Turkey, emre.dandil@bilecik.edu.tr 2 Faculty of Technology, Mechatronics Engineering, Sakarya University, Sakarya, Turkey, muratc@sakarya.edu.tr 3 Faculty of Technology, Department of Comp. Eng., Sakarya University, Sakarya, Turkey, ziyae@sakarya.edu.tr 4 Bolu Vocational High School, Abant Izzet Baysal University, Bolu, Turkey, muratozkan@ibu.edu.tr 5 Faculty of Medicine, Department of Chest Diseases, Abant øzzet Baysal University, Bolu, Turkey, aghhozlem@yahoo.com 6 Faculty of Medicine, Department of Radiology, Abant øzzet Baysal University, Bolu, Turkey, arzuolcun@gmail.com Abstract—Lung cancer is the most common type of cancer among various cancers with the highest mortality rate. The fact that nodules that form on the lungs are in different shapes such as round or spiral in some cases makes their detection difficult. Early diagnosis facilitates identification of treatment phases and increases success rates in treatment. In this study, a holistic Computer Aided Diagnosis (CAD) system has been developed by using Computed-Tomography (CT) images to ensure early diagnosis of lung cancer and differentiation between benign and malignant tumors. The designed CAD system provides segmentation of nodules on the lobes with neural networks model of Self-Organizing Maps (SOM) and ensures classification between benign and malignant nodules with the help of ANN (Artificial Neural Network). Performance values of 90.63% accuracy, 92.30% sensitivity and 89.47% specificity were acquired in the CAD system which utilized a total of 128 CT images obtained from 47 patients. Keywords-lung cancer, lung nodule, CAD, CT images, ANN classification I. INTRODUCTION Nowadays, lung cancer is one of the most deadly types of cancer. [1]. Various treatment options are used for lung cancer patients such as surgery, radiotherapy and chemotherapy. Despite these methods, 5 year survival rate for lung cancer patients is as low as 14 %. However, as in other cancer cases, survival rate may go up to 49 % if identified at an early stage [2]. Computerized tomography (CT) is the most frequently used imaging technique in the diagnosis of lung cancer [3]. Nodules and pathological residues with varied diameter can be comfortably viewed by CT [3]. Nodules on the lung are classified as benign or malignant. During diagnosis, malignant nodules that are solid and atypical can be assessed as benign in some cases. However, in most cases, a solid nodule is usually classified as malignant [4]. It is crucial to diagnose nodules at early stages in order to accelerate the treatment process. CAD systems designed for the medical application provide various benefits for successful detection of pulmonary nodules. It is possible to start treatment process early with the help of these systems and they facilitate decision making process of physicians. In the literature, there are some studies regarding early diagnosis of lung cancer and identification of nodules. Okumura et al. [5] detected lung cancer with filtering techniques by using X- Ray CT images. Campadelli et al. [6] used image processing techniques to segment the lung on X-ray images for nodule identification. Lee et al. [7] developed a new approach regarding automatic detection of benign nodules. They used genetic algorithm based template matching technique on CT images. Kanazawa et al. [8] suggested a fuzzy cluster based CAD system for the identification of pulmonary nodules. Biradar and Patil [9] designed a CAD system to detect benign lung nodules by using CT images. They used the extraction of regions of interest and basic image processing techniques. Choi and Choi [10] proposed a CAD system to automatically classify lung nodules. Furthermore, there are various ANN-based CAD in literature. Suzuki et al. [11] proposed a pattern-recognition technique based on ANN using low- dose CT images for reduction of false positives in computerized detection of lung nodules. In another paper, Coppini et al. [12] presented a neural-network-based system for the computer aided detection of lung nodules in chest radiogram. Kuruvilla and Gunavathi [13] described a computer-aided classification method in CT images of lungs developed using ANN. However, in these studies, the true positive and false positive rates are not enough to meet the requirements of clinical use. Moreover, since these studies don’t focus on early-detection of lung nodule. They don’t include any suggestion for the detection of small size nodules. This study proposes an ANN based CAD system for automatic classification of benign/malign pulmonary nodules at early stages. In this paper, Self-Organizing Maps (SOM) [14] has been used for nodule segmentation to enable the smallest nodules in the lungs. GLCM (gray- level co-occurrence matrix) [15] method has been utilized for the feature extraction of benign or malignant nodules. ANN, which is an effective classification technique, has been employed for classification. Rest of the paper is organized as follows: Section 2 provides details of the designed CAD system. Section 3 includes results of the experimental processes and analysis. Performance evaluation of the proposed CAD and Discussions are explained in the last section. II. MATERIAL AND METHOD A. CT Image Dataset An image database was created for the designed CAD system by collecting a total of 128 CT images from 47 different patients. There are 128 benign/malignant nodule in dataset. Based on pathological results, 52 of these nodules were malignant and 76 were benign. Images in International Conference of Soft Computing and Pattern Recognition 978-1-4799-5934-1/14/$31.00 ©2014 IEEE 382