Detection of Lung Cancer using Deep Learning based Convolution Neural Network Models

One of the leading cause for cancer based deaths worldwide is because of Lung cancer. Successful treatment and survival can be significantly improved by early detection of lung cancer. The proposed model has the possibility to aid radiologists by detecting lung cancer in early stages, which can lead to more timely and effective treatment. This thesis focuses in developing a CNN-based model to detect the possibility of lung cancer in medical images. The proposed model uses CNN models, which are trained in advance, such as ResNet50, Inception and Xception, as a feature extractor, and then trains a fully connected neural network on top of the extracted features which classifies if the image is either cancerous or non-cancerous. The model is trained and evaluated based upon a dataset of computed tomography (CT) imaging of the lungs. Experimental results concludes that the model proposed accomplish a high level of accuracy in classifying lung cancer, with an average accuracy of 93.75% for ResNet50, 94.56% for inceptionNetv3 and 97.85% for Xception. The study also compares the accomplishment of the model proposed with other highly developed models, and the results illustrates that the proposed model outperforms other models in terms of accuracy, precision, and recall. The proposed model has the capability to assist radiologists in detecting lung cancer in early stages, which can lead to more timely and effective treatment. The evaluation also gives insight in the usage of deep learning approaches for analysing medical image, particularly in the context of lung cancer detection.