Detection of Candidate Nodules in Lung Tomography by Image Processing Techniques

Abstract

In this project, the data set generated from the data of the National Cancer Institute (NCI) derived from DICOMformatted lung cancer data has been converted to PNG format to make it suitable for the operation of image processing algorithms. The converted data is filtered by a second-order median filter to prevent noise. A laplacian filter was applied to clarify the boundaries of the candidate nodules. The threshold must be determined according to the histogram value of each view. To select a different threshold for each image was subjected to the determination threshold Otsu method. In Otsu threshold selection method determines the point at which the minimum-class variance and between-class variance is a maximum, and determines the threshold value. In the availability of candidate nodules, nodule diameter is intended for detection of the large nodules than 2 mm. To meet this requirement, the Otsu thresholding method was applied on the MATLAB platform to sieve nodules smaller than 2 mm in diameter and candidate nodules were detected. By determining the morphological characteristics of the identified candidate nodules in MATLAB platform, candidate nodule; Major features such as area, major axis length, minor axis length, and perimeters are extracted. The boundaries of the identified candidate nodules are bounded by the edge detection algorithm and numerically ordered.