Detection and Localization of Pulmonary Carcinoma Using Deep Learning Approach in Computed Tomography Images

摘要

The detection and localization of cancerous pulmonary nodules from CT scans at an early stage can reduce the fatality rate of lung Carcinoma. The ability of Convolutional Neural Networks (CNN) to identify suspicious tumors in medical images with high precision and accuracy inspired the researchers to seek its performance in the area of lung malignancy detection. Moreover, the deep learning network reduces the workload of radiologists by processing and extracting information at a faster rate. This paper proposes the detection of nodules using pre-trained ResNet-50 model and its localization by sliding patch extraction and normalized squared difference method. The efficiency of the system was analyzed on the dataset released by the LUNA16 challenge. The performance measures of deep model computed have achieved 99.08% accuracy, 98% sensitivity rate, 98.98% specificity rate, 99.1 % AUC, and F1 score of 98.09%.