Increasing the Efficiency of Lung Cancer Detection by Improving Local Magnification Operations of the FPR Network

Abstract

Lately, lung cancer has become a terminal disease increasing the mortality rate due to the late diagnosis of the ail-ment. Early diagnosis can help reduce the death rate abundantly. The prediction of abnormalities from the given input images is a crucial factor. Deep learning has played an important role in early cancer detection by training networks to detect abnormali-ties via the given image. Convolution Neural Network (CNN) are most commonly used for cancer detection. In this paper, we pro-pose a CNN with the concept of down-sample in the Region of Interest (RoI) of the Computed Tomography (CT) images where the RoI will be subjected to magnification. Here, the magnifica-tion operation will first identify a spot from the upper region and then travel downwards towards the end of the CT image. Howev-er, every RoI will undergo local magnification process before the network could detect the next lesion. Detecting lesion are more effective as the lesions are disrupted structures in the human tissues that projects anomalies in the section viewed. Therefore, these anomalies can be useful in detecting lung cancer efficient-ly.