deep learning techniques to detect lung cancer and annotation of the lung images

Abstract

newline Lung cancer detection has been one of the most serious challenges in the medical sciences in recent years. In this work, a total of 8000 lung images were collected out of which 1080 CT lung images were used, with four classes: adenocarcinoma, small-cell-carcinoma, Squamous cell carcinoma and normal lung. Machine learning models that are implemented for detecting lung cancer are compared with the three classifier algorithms using SKlearn (Scikit-learn) tool: support vector machine (SVM), logistic regression (LR) and random forest (RF). The result of the proposed three classifiers has been analysed. The accuracy of the SVM is 85% Logistic regression is 80% and random forest is 84% are obtained. When comparing LR, and RF algorithms with SVM, it achieved a better accuracy of 85% for detecting cancer in CT lung images. newlineA convolutional neural network (CNN)-based model was implemented by applying the ReLU and Tanh activation functions. Build a basic CNN model and analyse it using regularisation and augmentation techniques to get better accuracy. Explore the use of pre-trained models, namely VGG-16, Resnet-50, and InceptionV3, for lung cancer detection. Feature extraction and fine-tuning techniques were employed to utilise the pre-trained models effectively. newlineVGG16 was used as a feature extractor, achieving training and validation accuracies of 92% and 88%, respectively. Inception V3 and ResNet50 were also evaluated, with Inception V3 achieving lower accuracies (74% and 87%) compared to VGG16 and ResNet50 obtaining accuracies of 76% and 82%, respectively. CNN model with ten-layer sequential for lung cancer detection. The lung cancer detection application was deployed using the Heroku platform, providing accessibility through the web browser. This application was built using Flask, a cloud-based web server and Python-based newlinev newlinemicro-web framework. This acts as a user-friendly application programming interface (API) between the cloud server and the proposed application model. newlineAnnotation of lung images was perfor