Design of Automatic Lung Nodule Detection System Based on Multi-Scene Deep Learning Framework

Abstract

Nowadays, the efficient identification of the lung nodule greatly leads to the chance of lung cancer risk assessment. Hence, the exact locations of lung nodules are a critical and complicated task. Researchers in this area have been working widely for almost two years. However, previous computer-aided detection (CAD) modules, such as transforming CT, segmenting the lung nodule and extracting the features are mostly complex and time-consuming, because more modules will require the creation of a complete image processing system. In addition, certain state-of-the-art deep learning systems are specified in the database standard. For this purpose, this paper suggests an efficient identification system for lung nodules based on Multi-Scene Deep Learning Framework (MSDLF) by the vesselness filter. A four-channel CNN model is designed to enhance the radiologist's knowledge in the detection of four-stage nodules by integrating two image Scenes. This model can be applied in two different classes. The results show that the Multi-Scene Deep Learning Framework (MSDLF) is efficient for increasing the accuracy and significantly reducing false positives in an enormous amount of image data in the detection of lung nodules.