Deep Learning Based Nodule Detection from Pulmonary CT Images

Abstract

In recent years, the morbidity and mortality of lung cancer are rising rapidly, and it has become one of the most malignant tumors with the highest morbidity and mortality. In the early stage of lung cancer, the pulmonary nodules are usually expressed in morphology. With the widespread use of CT technology, scanning can be used to detect malignant nodules in the lesion, which can greatly improve the survival rate of patients with lung cancer. However, the CT image is usually very high in dimensionality, which requires the doctor to spend a lot of time reading, and some tiny nodes are difficult to detect and easily lead to misdiagnosis. Computer aided detection technology can assist radiologists to diagnose, and effectively improve the efficiency and quality of diagnosis. Computer aided diagnosis of pulmonary nodules involves segmentation of lung parenchyma, suspected nodules extraction, and automatic recognition of pulmonary nodules. In this paper, segmentation of lung parenchyma and suspected nodules extraction are similar to traditional methods. The pulmonary parenchyma is extracted from original CT images by the maximum interclass variance method, and the connected regions are extracted in the lung parenchyma, which are the suspected nodules. Suspected nodules are classified by means of convolutional neural networks. Big data-driven artificial intelligence in the early diagnosis of lung cancer, not only can save the lives of countless patients, but also for the alleviation of medical resources and doctors and patients.