3-D medical image recognition by model-based generative learning

在本研究计画中，我们提出一种利用产生式学习从有限的学习样本中建构模式分类器的方法，并将其应用于医学影像的分割与病灶侦测。一般来说，需要大量的学习样本才能实现有效的分类。然而，很难像癌症一样收集足够的样本。为了克服这一点，我们开发了生成适当的学习图像和适当的特征向量的方法。利用该方法实现了一种新的三维医学图像的快速刚性配准方法。

In this research project, we developed a method to construct pattern classifier from limited learning sample by generative learning and applied them to the medical image segmentation and detection of lesion. In general, huge learning sample is required to make an efficient classifier. However, it is difficult to collect adequate sample like cancer. To overcome this, we developed methods to generate both appropriate learning images and appropriate feature vectors. By using this method, we realized a novel quick rigid registration method of three dimensional medical images.