Bayesian Classification and Artificial Neural Network ... - IEEE Xplore

3 downloads 0 Views 477KB Size Report
Bayesian Classification and Artificial Neural Network. Methods for Lung Cancer Early Diagnosis. Fatma Taher. Department of Electronic and Computer ...
Bayesian Classification and Artificial Neural Network Methods for Lung Cancer Early Diagnosis Fatma Taher

Naoufel Werghi and Hussain Al-Ahmad

Department of Electronic and Computer Engineering Khalifa University Sharjah, UAE [email protected]

Department of Electronic and Computer Engineering line 1 Khalifa University Sharjah, UAE {naoufel.werghi, alahmad}@kustar.ac.ae

Abstract—Lung cancer is a serious illness which can be cured if it is diagnosed at early stages. One technique which is commonly used for early detection of this type of cancer consists of analyzing sputum images. However, the analysis of sputum images is time consuming and requires highly trained personnel to avoid diagnostic errors. Image processing techniques provide a reliable tool for improving the manual screening of sputum samples. In this paper, we address the problem of extraction and segmentation the sputum cells based on the analysis of sputum color image with the aim to attain a high specificity rate and reduce the time consumed to analyze such sputum samples. In order to form general diagnostic rules, we use a Bayesian classifier to extract the sputum cells followed by using a Hopfield Neural Network (HNN) to segment the extracted cells into nuclei and cytoplasm regions from the background region. The final results will be used for a Computer Aided Diagnosis (CAD) system for early detection of lung cancer. We used some performance criteria such as sensitivity, specificity and accuracy to evaluate the proposed methods. Our methods are validated via a series of experimentation conducted with a data set of 88 images.

I.

INTRODUCTION

Lung cancer is considered to be the leading cause of cancer death throughout the world, and it is difficult to detect in its early stages, because symptoms appear only at advanced stages. Physicians use several techniques to diagnose lung cancer such as chest radiograph and sputum cytological examination where a sputum sample can be analyzed for the presence of cancerous cells. Manual screening for the sputum cells identification involves a labor-intensive task with a high false negative rate [1]. Automatic screening will offer several advantages, such as improving the sensitivity of the test. Recently, some medical researchers have proven that the analysis of sputum cells can assist for a successful diagnosis of lung cancer, for this reason we attempt to come with a CAD system for detecting the lung cancer in its early stages based on the analysis of the sputum color images, therefore the CAD system would be a great support for pathologists, to handle larger amounts of data. Eventually, this system will be useful for handling large sputum image databases and relieving the

978-1-4673-1260-8/12/$31.00 ©2012 IEEE

physician from tedious and routinely tasks. The most recent estimate statistics according to the American Cancer Society indicate that 226,160 new cases will be diagnosed (116,470 in men and 109,690 in women) in U.S , and there will be estimated 160,340 mortalities from lung cancer (87,750 in men and 72,590 among women), in 2012 [2]. Several works have been addressed the segmentation of the sputum cells for lung cancer detections [3]. The analysis of sputum images had been used in [4] for detecting tuberculosis. Nothing has been done in developing CAD systems based on sputum cytology apart of early the attempts of Sammouda [5] and Taher [3]. In this paper our novel contribution lies in the following: 1) Using the Bayesian classification to detect and extract the sputum cells from the sputum color images prepared by the Papanicolaou standard staining method [6]. 2) The segmentation of sputum cells using a HNN to segment the sputum images which is characterized by noisy and cluttered background patterns into three regions, background, cytoplasm and nuclei regions. There are many algorithms which can be used for medical image segmentation, such as histogram analysis, region approach, edge detection and adaptive thresholding. A review of such image segmentation techniques can be found in [7]. Other authors have considered the use of color information as the key discriminating factor for cell segmentation for lung cancer diagnosis [8]. The rest of this paper is organized as follows. In section 2, Bayesian classification is introduced. In section 3, HNN segmentation method is described. In section 4, experiments are presented. Finally, the conclusions are discussed in section 5. II.

BAYESIAN CLASSIFIER

The aim of the detection process is to determine whether or not a pixel in the sputum image belongs to the sputum cell using the pixel color information. The Papanicolaou staining method used a red dye, results in a red sputum cell with darkred nucleus and clear-red cytoplasm regions. This staining allows, to some extent, the sputum cell to have a distinctive chromatic appearance. Therefore, it is possible to separate

773

automatically the sputum cell from the background using its color attributes. In this work, we investigate how the choice of color space affects the sputum cell detection. There exist various color spaces for encoding a pixel color. In this paper, we consider four commonly used color representations in image processing: RGB, YCbCr, HSV and L*a*b*. Bayesian classifier is a probabilistic classification [9], which allows a systematic and methodologist estimation of the threshold parameters rather than using a heuristic rule based on trial and error testing. In Bayesian classifier a pixel x is considered part of the of sputum region If p(bg|x)