Analysis of Lungs CT Image Using Deep Learning Techniques

Lung cancer claimed the lives of more than 1.5 million people globally in 2017, making up 26% of all cancer-related deaths. In response to this difficult circumstance, a number of device detection techniques have been created to identify lung cancer at an early stage and increase the odds of survival for the patients. The proposed method aims to examine deep learning's efficacy in lung nodule malignancy diagnosis. This objective is motivated by deep learning's accomplishments in challenges requiring the categorization of natural and medical images. The Lung Picture Data Consortium and Set Of images Resource Initiative (LIDC/IDRI) dataset were processed using CNN like VGG16, VGG19, MobileNet, Xception, Recurrent neural networks, InceptionV3, Inception-ResNet-V2, DenseNet169, DenseNet201, NASNet Mobile & NASNet Large as based feature extraction methods. The specified deep characteristics were classified using the Naive Bayes classifier, MLP, SVM, KNN & Random Forests (RF). The effectiveness of classifiers was also evaluated using the metrics Efficiency (ACC), Areas Under the Curve (AUC), Positive Cases Rates (True positive rate), Precision (PPV), and F1-measure to compare it to those of other classifiers in the literature as well as to other classifiers. The highest deep extraction and classification combinations at 88.41% of ACC & 93.19% of AUC was CNN - ResNet50 with SVM - RBF. Results obtained by using a CNN trained with images other than medical ones are comparable to those of other studies in the area. For the time being, chest computed tomography (CT) pictures may be used to classify lung nodule malignancy using deep transfer learning to identify representative imaging biomarkers.