Article Preview
Top1. Introduction
Lung tumor is a progressive disease containing abnormal cells leading to cancer. The abnormality present devastates the proper regular performance and functioning of lungs. Automatic diagnosis system in medical imagining has increased the survival rate of lung patients at early stage from 20 percent to 70 percent based on 5 years survey since it provides the appropriate results at the right time. (Gajdhane & Deshpande, 2014). The survival rate prediction is alarming and the necessary factor which proved to help in proper treatment and diagnosis of lung cancer patient (Hawkins et al., 2014). There are two major types of lung carcinoma broadly subdivided into nonsmall cell lung cancer and small cell lung cancer. The nonsmall cell lung carcinomas subtypes are squamous cell carcinoma, adeno carcinoma and large cell carcinoma (Patil & Jain, 2014). The prognosis of lung Tumor is the most challenging task as the cells are assembled on each other therefore it is essential to determine the features and structure of diagnosed image (Tiwari, 2016). The automated lung nodule detection on medical images involves image enhancement, image segmentation and feature extraction to classify the stages of tumor so that proper planning of treatment could be accomplished on lung cancer patient (Tariq et al., 2013). The research in the field of medical imaging is rising especially in magnetic resonance imaging of lung tumor so that effective technique could be developed for evaluation of tremendous data. On the medical lung MRI (Magnetic Resonance Imaging) zoomed image, the foremost method implemented is image filtering to remove distortion and improve the quality of image. There are numerous ways to perform filtering on image by using methods such as thresholding, fast Fourier transform, morphological operation, median filter, Gaussian filter (Dimililer et al., 2017). The median filter is found to be effective technique in this paper to remove noise and distortion (Tun & Khaing, 2014). There is various image segmentation algorithm used in medical image research for analysis of cancer detection and finding the measurement of lung nodule detection (Norouzi et al., 2014). Based on the previous research the segmentation technique on MRI image is broadly classified into (a) Thresholding (b) Clustering (c) Region-Growing (d) Edge and Line oriented (e) Region splitting (Sharma et al., 2018). Although various distinct techniques are proposed by the researchers in literature still the various methodologies are proposed to meet the challenges of segmentation and providing better outcomes. This paper includes the Marker controlled watershed transform (Kanitkar et al., 2015) algorithm which provides better results than adaptive Otsu algorithm for detection of abnormalities (Prasad, 2013). The Marker controlled watershed algorithm (Abdillah et al., 2017) is improved technique for observing the structure of lungs whereas the adaptive thresholding is dynamic method to analyse the image. Watershed algorithm is efficient image processing tool based on mathematical morphology to recognize and envision cancer presence in lungs of patient as it relocates the cancer on pixels with high contrast (Janardhanan & Satishkumar, 2014). The feature extraction on region of interest is performed after image segmentation to obtain the geometrical and intensity based mathematical characteristic on Magnetic Resonance imaging using masking and binarization. The features such as Area, Perimeter, Standard deviation and Centroid provide the location and other analysis features of tumor (Tun & Soe, 2014). The classifier system in medical image research diagnosis provides the evaluation of data of patient disease and gradually providing better results to experts. SVM has become progressively prevailing supervised learning algorithm including classification, novelty detection and regression (Sweilam et al., 2010). The dataset used for training and testing in Support vector machine is selected from cancer institute for research. A finite training set is formulated using previously known decision parameters. It is supervised learning approach which enables the input and output mapping functions from labelled and categorized training dataset. On the basis of training, SVM classifies the categories (Low, Medium, high) of unknown data intelligently for testing purpose (Kaucha et al., 2017). The presented paper has developed the computer aided diagnosis and established the prediction model of SVM (Support vector machine) (Xiuhua et al., 2010) by evaluation of parameters such as accuracy, sensitivity and specification using true and false prediction. This proposed system is an integrated diagnostic system for lung cancer detection and classifies the tumor patient stages by constructing optimal classifier using SVM (Support vector machine classifier).