Open AccessDetection of Pulmonary Nodule using Shape-Based Feature Descriptor and Neural Network
Author(s) -
Nurfarhana Hazwani Jusoh,
Haniza Yazid,
Shafriza Nisha Basah,
Saufiah Abdul Rahim
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1372/1/012066
Subject(s) - thresholding , artificial intelligence , computer science , segmentation , pattern recognition (psychology) , nodule (geology) , feature (linguistics) , computer aided diagnosis , filter (signal processing) , region of interest , feature extraction , computer vision , artificial neural network , cad , image segmentation , image (mathematics) , engineering , paleontology , linguistics , philosophy , engineering drawing , biology
This research aims to detect the pulmonary nodule presented in lung Computed Tomography (CT) scan images. Generally, a Computer-Aided Diagnostic (CAD) system was designed and developed to aid the radiologists in medical imaging department to reduce the time and to obtain faster and better results for lung nodules diagnosis of a patient. Four major stages involve in this paper which are pre-processing, segmentation, features extraction and classification. The images that were utilized were acquired from LIDC-IDRI database that available publicly for CT scan lung images. Initially, the median filter was employed in pre-processing to filter and remove the noises, unwanted distortions and artifacts presented in the images during scanning process. For the second stage, the implementation of Otsu thresholding (thresholding-based method) and watershed algorithm (region-based method) were used to segment the nodules (Region of Interest (ROI)). Manual cropping method was implemented to segment the nodule for further process. The main contribution of this paper is the extraction of the features based on shape descriptor. 10 features were extracted from the segmented nodules. Finally, all extracted features from the segmented nodules were classified into nodule candidates and non-nodule candidates using Back Propagation Neural Network (BPNN). Based on the experiment, it can be observed that the proposed approach works well with CT scan images and segmented the interested nodules with the accuracy of 94%. This semi-automated approach is fast compared with the conventional approach used by the radiologists in current time being.