LOW-DOSE CT-BASED AI MODEL FOR EARLY LUNG CANCER DETECTION AND FALSE-POSITIVE REDUCTION IN POPULATION SCREENING

Abstract

In this research, a low dose artificial intelligence framework for lung cancer detection and minimization of false-negative in mass screening programs is put forward. The aim is to increase the sensitivity of the diagnosis and yet reduce the incidence of clinically unnecessary intervention in the case of a false positive. A full dataset of low dose chest computed tomography (CT) scans of the screening participants were obtained and preprocessed with image normalization, noise reduction and lung region extraction. One novel deep learning system was developed to detect pulmonary nodules and classify the malignancy risk, using convolutional neural networks (CNNs), attention mechanism and multi-layer fusion.A novel deep learning system was designed to detect pulmonary nodules and classify the severity of malignancy, which combines convolutional neural networks (CNNs), attention mechanisms and multi-layer fusion methods. Stratified data sets were used to train and validate the model and accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC-ROC), precision, recall, and false-positive reduction were used to assess the model. A comparison with a conventional computer-aided detection system and radiologist evaluation was performed.