Lung cancer is one of the main reasons for death globally, with an impressive rate of about five million deadly cases per year. Detection of lung cancer at an early stage is necessary to prevent deaths and increase the survival rate. However, most of the current works on lung cancer detection stop at the algorithm level, leading to a lack of a platform that allows patients to get diagnosis results by simply uploading their CT scans. To slightly fill up the vacancy, this paper designs an auxiliary lung cancer diagnosis system based on deep learning techniques to detect lung cancer, provide associated knowledge and send out personal reports. Specifically, our system consists of two main parts, the lung cancer diagnosis assistant and the lung cancer diagnosis center. The lung cancer diagnosis assistant allows patients to upload CT scans and get their reports in the mailbox. Meanwhile, it encourages patients to learn medical knowledge about lung cancer. When detecting lung cancer, we first resample and resize the uploaded images to avoid the impact of picture quality differences and reduce memory consumption brought by 3D CT scans. Then, we construct our 3D convolutional neural network (3D CNN) model based on processed images. We verify the effectiveness of our model compared with a baseline model on the Data Science Bowl 2017 (DSB17) dataset, which is a lung cancer dataset consisting of over a thousand low-dose CT images from high-risk patients. The baseline model is a relatively simple model with only 2 convolutional layers. Our improved edition gets an accuracy of around 64%, 5% higher than the baseline model, which illustrates the effectiveness of our 3DCNN model. The experiment results and analyses demonstrate that our system is of practical value to detect lung cancer and provide extra services for patients.
|