The Effect of Hyper Parameters on the Classification of Lung Cancer Images Using Deep Learning Methods

Abstract

Cancer is a fatal disease arised from the formation of abnormal cells as a result of random growth in the human body. Lung cancer is the frequently encountered cancer type and causes abnormal growth of lung cells. Diagnosis at an early stage substantially enhances the chance of survivability of the patient, as well as prolongs the survival time. There may even be a complete recovery. For this reason, it is of vital importance to support the diagnosis and detection of doctors and enables them to diagnose more easily and quickly. In this paper, it is aimed to detect lung cancer disease with the help of Alexnet and Resnet50 architectures, which are deep learning architectures, by using computed tomography images. In addition, the performances of the hyper-parameters of maximum epoch and batch size, which are of great importance in training the models, have been compared. According to the results obtained, the highest overall accuracy in automatic detection of lung cancer has been achieved with the AlexNet architecture. The highest overall accuracy value obtained as a result of the simulations is found to be 98.58% with the highest cycle value and the batch size are 200 and 64, respectively.