Lung Cancer Classification Optimization using CNN and Edge Detector from CT Scan Images

This paper presents a lung cancer cell classification method using a 3-layer Convolution Neural Network (CNN) from edge images of CT scan images. It compares the performance with the gray images-based classification method with the same model of CNN. This system uses 602 CT scan images of normal lung, adenocarcinoma, squamous cell carcinoma, and large cell carcinoma classes which were annotated and stored as gray images. Experimental results show that the edge images-based CNN classifier performs better than the gray images-based CNN classifier. As batch size and maximum number of epochs have several impacts including memory utilization, computation time, and accuracy, to determine the optimal batch size and maximum value of epochs this research experiment with several batch sizes and different values of maximum epochs. Experiment results also show that batch size significantly impacts classification performance. For a batch size of 24 and a max epoch value of 20, this system resulted in the best classification accuracy of $91 \%$ with the edge images-based CNN classifier and $85 \%$ with the gray images-based CNN classifier.