Automated Lung Cancer Screening Using Deep Transfer learning
Lung cancer is found to be the prevalent reason for cancer-related death all over the world. There are many procedures for finding lung cancer, like Chest Radiography (x-ray), Computed Tomography (CT), Magnetic Resonance Imaging (MRI), etc. But, most of these procedures are costly, time-consuming and are detecting in its advanced stages. Accordingly, there is a strong desire for a peculiar technology for the initial recognition of lung cancer. Cancer detection in its beginning can raise the survival rate. Recently sputum cytology proves to be an affordable method for early detection. This paper proposes a low cost and noninvasive computer-aided diagnosis (CAD) technique for primary recognition of lung carcinoma by analyzing the supplementary changes in sputum samples. The problems connected with sputum analysis include uneven stain, overlying cells, non-focused cells and inappropriate illumination in microscopic images. Hence, in this paper, the contrast of sputum images is improved by the decorrelation stretch enhancement. Moreover, one of the deep learning architectures such as ResNet101 is utilized for the detection of supplementary changes with the transfer learning method. The outcomes show better accuracy than traditional methods.