A Versatile Segmentation Approach For Diagnosis of Lung Cancer

Abstract

The necessary and critical step is to evaluate the development of lung cancer and nodule segmentation. Immobile challenge in the field of segmentation and classification of pulmonary lung nodule is particularly used to identify the small size nodule. To improve and sustain the diagnosis analysis, this paper puts forward and widens a new approach to segmentation and classification method for lung nodule size less than 3mm. In this paper, we examined and proposed a new method based on transition region based P-Tile thresholding and followed by Watershed processing for segmentation. First we reap the ROI from the input CT image and enhance the region of nodule by median filtering algorithm. Second Object contours are obtained by transition region based analysis. Third to extract multiple objects ROI from object contours employ M-Type morphological operation. Fourth prepare the images for segmentation by reducing noise and smoothing operations like weighted average filter. Kuwahara filter is used to smoothen the images and to preserve the edge position. Then we make use of crack code analysis to renovate lung boundaries. Finally the result is obtained by overlap the extracted image with the restored lung mask. To evaluate the novel segmentation method examines 90 lung nodules with 3mm to 9mm diameter from LIDC database. The presented novel approach attain ground truth rate of 86.93% ±0.09 with false positive rate of 15.09% ±0.06.