[4C1-OS-27a] [Organized Session] OS-27

[4C1-OS-27a-02] Opacity Annotation of Diffuse Lung Diseases Using Convolutional Neural Network and SVM

〇Takuro Higuchi¹, Shingo Mabu¹, Noriaki Hashimoto¹, Shoji Kido¹, Yasushi Hirano¹, Kenji Kondo^2,3, Jun Ozawa³ (1. Yamaguchi University, 2. Advanced Industrial Science and Technology, 3. Panasonic Corporation)

Keywords:CNN, SVM

Research on Computer Aided Diagnosis (CAD) has been conducted to give second opinions to medical doctors, where deep learning has been actively applied to building CAD systems because of its automatic feature extraction ability. However, deep learning requires a large number of annotated data to achieve high performance. This paper deals with an opacity classification problem of diffuse lung diseases using machine learning and aims to achieve high classification accuracy using smaller number of training data with annotation. The proposed method first extract regions of interest (ROI) images from lung CT images using Selective Search. Then, the classification is executed by combining VGG16 and SVM. VGG 16 is one of the convolutional neural network structures and used as a feature extraction method, then the opacity classification is executed by SVM. From the experimental results, the proposed method showed higher classification accuracy than VGG16 alone.