分野別セミナー

【講演要旨】Deep Learning has achieved great successes in medical image analysis. However, the features extracted by convolutional neural networks (CNNs) are typically tailored to pixel intensity patterns and overlook essential anatomical structures like connected components and loops. In this paper, we propose a cubical persistent homology approach that utilities topological features of objects for CT image classification. Besides, we also proposed a method that combined the topological features of CT images with their original feature information. For an input 3D-CT image, we first compute its cubical persistence diagram and vectorized topological features into different machine learning classifiers. We evaluate our methods on a public CT image dataset.