Lung Cancer Diagnosis Using Deep Convolutional Neural Network

Machine Learning;Deep Learning;Computer Vision;Lung Cancer Diagnosis;Nodule Detection;620.5

융합과학기술대학원 융합과학부 ;

University:서울대학교 대학원

关键词: Machine Learning;    Deep Learning;    Computer Vision;    Lung Cancer Diagnosis;    Nodule Detection;    620.5;   

Others  :  http://s-space.snu.ac.kr/bitstream/10371/142270/1/000000150478.pdf

美国|英语

来源: Seoul National University Open Repository

【 摘 要 】

In this paper, a method of diagnosing lung cancer from low-dose CT (computed tomography) scan images of high-risk patients using deep 3d convolutional neural network is described. Previously, diagnosis of lung cancer was performed by pulmonary radiologist based on malignant lung nodules from the CT images. However, with great success of neural network in various fields, the diagnosis potentially can be aided by deep learning. In Kaggle Data Science Bowl (DSB) 2017 challenge, which motivated this paper, encouraged participants to solve lung cancer detection problem by offering CT scan data of potential lung cancer patients. On this challenge, a whole pipeline was designed to predict whether the patients have lung cancer within 1 years of scanning. The proposed system is composed of three major stages, preprocessing of raw CT scan images, nodule detection and lung cancer prediction. Finally, a binary cross entropy loss score of 0.51092 was achieved on the test set which ranked 35 out of 1972 teams.