Development of methods for structure analysis based on generative learning and its application to understanding of medical images

在本研究项目中，我们开发了基于生成学习方法的医学图像中器官和疾病识别方法，该方法对人体器官等方差较大的数据具有鲁棒性。在气道和血管树的识别中，我们研究了生成学习的有效图像特征，并从医学图像中准确地提取它们。在器官的识别中，我们构造了一个ATLAS，它以数学方式表示器官的变化，并准确地提取它们。在淋巴结检测中，我们开发了一种新的滤波器，该滤波器对淋巴结有特异性的响应，并且检测出的淋巴结具有很高的准确性。我们证实了生成学习方法对于识别人体器官和疾病是有效的。

In this research project, we developed methods for recognizing organs and diseases in medical images based on the generative learning approach, which is robust to the data with large variance such as human organs. In the recognition of the airway and blood vessel trees, we investigated effective image features for the generative learning and extracted them from medical images accurately. In the recognition of organs, we constructed an ATLAS, which represents variations of organs mathematically, and extracted them accurately. In the lymph nodes detection, we developed a novel filter which responses to the lymph nodes specifically, and detected lymph nodes with high accuracy. We confirmed that the generative learning approach was effective for recognition of human organs and diseases.