Automatic Detection of Lung Cancer from Lung CT Images Using 3D Convolution Neural Network

Abstract

There are enough evidences that lung cancer is the most fatal among all the cancer types predicting 22% of total cancer-related deaths in the year 2021 in USA itself for both male and female. Only 21% of lung cancer patients survive for 5 or more years which is the third-worst among all the cancers (only pancreas cancer (10%) and liver cancer (20%) have lower 5-year survival rate than lung cancer). It is highly expected that detecting lung cancer in its initial stages can increase the survival rate. An efficient computer-aided diagnosis (CADe) tool for automatic identification of lung nodules from lung CI images may be very helpful, and many research groups across the world have consistently been working on it. We too have developed a 3D CNN model for auto-detection of lung nodules from CT images which have been described herein. The model consists of 25 layers which include 5 convolution layers, 8 ReLU layers, 3 max-pooling layers, 4 fully connected layers, and 1 softmax layer and is capable of delivering 100 and 60% accuracy at epoch 30 and epoch 1, respectively. ROC area of the developed model is found to be 1, which indicates that the model is a good one and quite satisfactory.