A Novel Multilevel Hybrid Segmentation and Refinement Method for Automatic Heterogeneous True NSCLC Nodules Extraction

The experts uses chest Computed Tomography (CT) images to manually analyze the presence of cancerous nodule during cancer screening process. Due to heterogeneous and low intensity nature of CT image, manual image analyzing becomes difficult which leads to different problems like false positive detection, consumption of huge analyzing time, observer error, etc. Developing an efficient automatic Computer Aided Detection (CAD) system is essential to reduce the frequency of missed lung cancer nodules, make diagnosis simpler and time saving. The CAD system improves the accuracy of lung tumor detection and survival rate of the patient. In this paper, a fully automated model is presented for NSCLC nodule(s) segmentation from CT scan image. The proposed method follows four steps: (1) Preprocessing, (2) Automatic Lung Parenchyma Extraction and Border Repair (ALPE&BR), (3) Automatic lung nodules segmentation using Connected Component Analysis (CCA) and Threshold BasedMathematical Nodule (TBMN) refinement algorithm and (4) Nodules filtering using Hounsfield Unit (HU) value and true cancerous nodule extraction. The ALPE&BR method consists of Automatic Single Seeded Region Growing (ASSRG) algorithm for automatic lung parenchyma extraction and novel hybrid border concavity closing algorithm to get clear lung boundary. The proposed method successfully segments the true cancerous nodules by filtering out false region such as vessels, bone, fat, soft tissues, etc. The proposed method can provide the SN of 99.41%, SP of 99.97%, FPR of 0.019%, DSC of 0.98, and accuracy of 99.97%. These results are used to demonstrate that the proposed method outperforms the existing lung nodule segmentation method.