Abstract |
According to the Korea Consumer Agency, out of 645 applications for medical damage related to misdiagnosis received from 2012 to 2016, 374 cases (58% of all applications) were misdiagnoses of cancer. Most of them were cancer, but 81.4 percent misdiagnosed as non-cancer, and 8.6 percent misdiagnosed as cancer. The most common causes were the image reading error and negligence of additional examination. Among the cancer types, lung cancer is most often misdiagnosed. Also, as the diagnosis is delayed, so cancer already advances and usually occurs at stage 3 or stage 4. According to the Disease Policy Division of the Ministry of Health and Welfare, the incidence of lung cancer (people per 100,000 population) increased every year from 28 in 1999 to 55.8 in 2018, doubling. The death rate also increased every year from 22.1 in 1999 to 34.8 in 2018. In 2019, lung cancer was the number one cause of death from cancer. In this paper, we design and implement a lung cancer diagnosis system using medical image data of individuals. The proposed system obtains a dataset with medical images acquired from The Cancer Imaging Archive (TCIA) of the National Cancer Institute (NCI). Our system trains the constructed dataset using a CNN (Convolutional Neural Network) model widely used in deep learning. Our system receives medical images from users for cancer diagnosis and then applies the CNN model to learn the obtained dataset. It then uses the trained CNN model to analyze new medical images and reasons out their results. Our system is developed for medical image readers and doctors. With the learning results of medical images, our system can reduce the false positive by comparing the other conventional methods. |