Downloads: 128 | Views: 285
Research Paper | Electronics & Communication Engineering | India | Volume 3 Issue 10, October 2014
Lung Segmentation and Tumor Identification from CT Scan Images Using SVM
Ganesh Sable [3] | Harsha Bodhey
Abstract: Segmentation of images has become important and effective tool for many technological applications like lungs segmentation from CT scan images, medical imaging and many other post-processing techniques. Lung cancer is the primary cause of deaths for both sexes in most countries. Lung nodule, an abnormality which leads to lung cancer is detected by various medical imaging techniques like X-ray, Computerized Tomography (CT), etc. Detection of lung nodules is a challenging task since the nodules are commonly attached to the blood vessels. Many studies have shown that early diagnosis is the most efficient way to cure this disease. This paper presents an adaptive segmentation of the lungs and the lobes and also an efficient algorithm for tumor classification from CT scan images. In pre-processing, first the RGB image is converted into gray scale image and then the noise is removed from this image using wiener filter. After that we have segmented the lungs from the original CT image using thresholding. The lobes of the lung will be segmented using marker based watershed transformation. Through the project we have developed an algorithm for identifying the tumor from segmented lung images. For identification we have used GLCM features and SVM classifier together. At last the tumor present in the CT scan image is isolated using FCM approach. The results indicate a potential for developing an automatic algorithm to segment lung lobes and tumor classification for surgical planning of treating lung cancer. For this, we have collected online database of 51 patients from Lola 11. The proposed system is implemented in MATLAB software.
Keywords: Lobe Segmentation, Watershed Transformation, Feature Extraction, GLCM Gray Level Co-occurrence Matrix, FCM Fuzzy C-means, SVM Support Vector Machine
Edition: Volume 3 Issue 10, October 2014,
Pages: 64 - 69