PULMONARY NODULE DETECTION USING LOCAL CONTRAST THRESHOLDING WITH CIRCULAR WINDOW (CWLCT)

Abstract

According to WHO, lung cancer is one of the major causes of cancer deaths worldwide. Lung nodule detection has substantially improved by using image processing techniques and Computer Aided Diagnosis (CAD) systems. A novel method is presented for processing Postero Anterior chest radiographs for extracting a candidate nodule based on their local properties using circular window-based local contrast thresholding (CWLCT). The algorithm is tested on different subtlety levels of chest radiographs and compared with square window-based Bernsen local contrast thresholding. For the first test 50 chest radiograph images with different subtlety levels are used. The experimental results of candidate nodule extraction using local contrast thresholding with a circular window show that for 90% of images true positive nodules are extracted accurately with minute error in the size and 4 false positive nodules/image. In the second test, the proposed method is compared with square window-based Bernsen local contrast thresholding and showed a reduction of 3.8 % in nodule size error. Utilizing CWLCT not only lowers the number of false positive nodules per picture but also assists in precisely segmenting nodules from chest radiograph images.