BASED DEEP LEARNING MODEL AN EFFICIENT OPTIMIZATION FOR NODULE SEGMENTATION IN LUNG

Abstract

Abstract Automatic and accurate segmentation of lung nodules is necessary for lung cancer analysis and it is a basic process in CAD (Computer-aided diagnosis). However, several kinds of the nodule and visual similarities in the chest make it complex to design segmenting the lung nodule. Further, the major challenge to detect lung cancer is accuracy which is affected by several factors. Hence, in this work, an efficient optimization based deep learning model for lung nodule segmentation. Initially, the input CT image is pre-processed by the RBF (rapid bilateral filtering). Then, this image is subjected to a segmentation process and this process is carried out by 2D otsu thresholding. Then, to select the optimal threshold value the metaheuristic optimization IHS (Improved harmony search) is used. Finally, the segmented nodules are extracted and classified by the ResNet-152 for identifying the presence of cancer. The experimental analysis is carried out on the benchmark dataset LIDC-IDRI. The experimental results showed that the proposed model achieves better results on the basis of JSI (Jaccard similarity index), Dice and accuracy respectively. The maximum dice value achieved by 2D otsu thresholding- IHS- ResNet-152 model is 0.997 respectively.