Optimal Convolutional Neural Network Model for Early Detection of Lung Cancer on CT Images

Abstract

Lung cancer diagnosis and prediction in the earliest stage of the disease can be extremely helpful to improve the survival rate of patients, but diagnosing cancer is one of the most challenging tasks for a radiologist. In this paper, we propose that convolutional neural networks (CNNs) can effectively detect and predict lung cancer. The design of model has 8 layers in which one is input layer and two conv layers and a ReLU, max-pooling to extract features and a fully connected layer which has sigmoid activation function which in terms connected to 1 neuron which is an output layer. This ouput layer is responsible for classifying the image in to a benign or malignant nodules. We have acquired the dataset from Kaggle having 1097 cases which can be passed as input data for input layer. A 2D set of resampled images with 64 * 64 pixels of size which are rotated, scaled, and randomly translated images is generated as input samples.