Deep Learning Based Automated Lung Cancer Detection from CT scan Leveraging Transfer Learning
Anuradha Rani Choudhury, Jyotirmayee Rautray, Pranati Mishra, Meenakshi Kandpal, Sasanka Sekhar Dalai
Abstract
Lung cancer has become the prime cause of cancer mortality throughout the board. As preliminary diagnosis can help to detect the ailment with sophisticated medical facilities. Early identification of such a lethal disease is mandatory as it can further help in combating it in the long run. Several deep learning and transfer learning models have been applied to identify the biomedical images in the form CT scans whether affected or not. In this work, we applied deep learning methods to the original 1097 CT scan images of patients from the IQ-OTH NCCD lung cancer dataset which consisted of malignant images, normal images and benign images. After intensive steps of image preprocessing, it is realized that images containing normal, malignant, and benign cases were classified as high-risk individuals for lung cancer detection and thus prediction of malignancy is carried out. In this work, a performance comparison has been presented in detail with respect to the classifiers: custom CNN, VGG16, Resnet50, and InceptionV3. Study echoes the potential of deep learning to be merged with transfer learning in the classification of lung cancer. The dataset IQ-OTH NCCD, greatly enhances the diagnostic accuracy of lung cancer, directly impacting better clinical outcomes in both the detection and treatment of lung cancer. Here, the results obtained a validation accuracy of 98.7% with reduced complexity for the model. The recommended method resulted in a higher F1 score value of 97.7%, thus verifying our recommended methodology for the comparison.