Pulmonary Nodule Detection Based on CT Images Using Convolution Neural Network

Abstract

Pulmonary nodule is a common lung disease, which can be prone to misdiagnosis and missed diagnosis. With the extensive application of CT technology, doctor's diagnostic efficiency has been greatly improved. However, the amount of CT image data is relatively large. Radiologists have to take a lot of time to read these images, and easy to overlook some minor lesions. Computer aided detection technology is an effective way to improve the efficiency and quality of the doctor's diagnosis. This paper put forward a kind of lung segmentation method based on morphology and statistic of size of the image area, while effectively eliminating the influence of the trachea to pulmonary parenchyma image segmentation. We also propose a method of region of interest(ROI) extraction based on morphology and circular filter, reducing the number of false positive and trying to retain the integrity of the ROI form. Finally, we have realized a reliable lung nodules compute aided diagnosis application on CT image, using Convolution neural network.