[1P4-J-10] Vision, speech: organisms and medicine

[1P4-J-10-01] Construction of a Classification Model for Nodule Detection in Lung CT Images

〇Taku Ri¹, Tatsuya Yamazaki¹ (1. Niigata University)

Keywords:CT image, Convolutional Neural Network, nodule

This paper proposes a method to detect nodules with cancer possibility from lung field CT images. In the proposed method, an image trimmed from the lung field is input to a classification model constructed by CNN (Convolutional Neural Network), and if it is judged to be a nodule, a mark is labelled on the lung field CT image at trimmed position. In this paper, in order to confirm the effectiveness of the classification model in the proposed method, we construct a model to classify into three classes of solid nodules, ground-glass nodules and non-nodules, and verify classification accuracy. As a result of the verification, for ground-glass nodule images and non-nodule images, the classification accuracy is 95% more. However, the classification accuracy of solid nodule images is 81.87%, and most of misclassified them have been classified as non-nodule.