Lung Cancer Detection Using Integration of Hybrid Segmentation Approach and DL Techniques

Abstract

Cancer is a frequent illness with a rising death rate in recent years. Lung Cancer (LC) is a deadly illness that has a high patient death rate. Patients' lives can be saved by correctly determining the LC stage and receiving an early diagnosis of this illness. LC can be identified using a variety of image processing, biomarker-based, and machine automation techniques, however, medical professionals have difficulties in accurately and promptly diagnosing the disease. These automated detection systems currently use a diversity of Machine Learning (ML) methods to identify LC in its early stages. However, the processing of LC detection is time-consuming, and these systems do not offer reliable detection. This work proposes a hybrid segmentation approach that combines the Enhanced Kernal Fuzzy Clustering (EKFC) algorithm with the Global Particle Swarm Optimizer (GPSO) to carry out segmentation. CNN architecture is used to classify and extract features. The supplied image's categorization layer is responsible for identifying whether the tumor is abnormal or normal. In this work, the CT scan pictures are extracted using the lung imaging data that were acquired from the Kaggle website. The suggested segmentation methodology outperforms the other two segmentation approaches in the market with a Dice Index of 0.93. Furthermore, the Convolutional Neural Network (CNN) from the suggested segmented technique obtains 97.8% classification accuracy compared to the LSTM model.