深層学習による胸部CT 画像からの結節状陰影の検出

抄録

In recent years, the number of death due to lung cancer is increasing year by year worldwide. In Japan, the number of deaths by cancer site in 2018 was 1st for men and 2nd for women. Early detection and treatment of lung cancer are important, so it is necessary to detect the nodular shadow in visual screenig. A CT apparatus is used to detect lung cancer. Due to the high performance of the CT apparatus, the number of CT images increases, the burden on the interpreting doctor is heavy, and the diagnostic accuracy may decrease. From these reason, development of a CAD system is required. In this paper, we propose an image analysis method to detect abnormal shadows from chest CT images automatically. As a method, the initial lesion candidate areas are extracted by using temporal subtraction technique that emphasizes temporal change by subtracting from a current image to previous one which is obtained same subject. The image of the area is given as input and classification is performed by CNN (Convolutional Neural Network). In the classification experiment based on our proposed method, 92.33 % of true positive rates and 10.53 % of false positive rates are obtained from the 49 clinical cases.