中文摘要：

针对从高分辨率颅脑CT图像中自动检出病变以实现计算机辅助诊断的需求,提出配准前特征提取的方法。该方法的主要特点是在图谱创建过程中使用散布点内插法获得整数点的特征值,而在病变检出时则采用格子点内插法获得非整数点的特征值。相对于配准后的特征提取,配准前的特征提取能够更加准确地描述图像的灰度特征和纹理特征。通过实验验证,基于配准前特征提取的颅脑病变检出方法能够提高病变检出率,提高检测精度,但同时也增加了假阳性率。为了减少假阳性率、进一步提高检测精度,在下一步的工作中要研究基于三维体数据的病变检出方法,同时还需要进一步研究非刚性配准的可逆性。

英文摘要：

This study is purposed to improve the quality of lesion detection from CT images aiming at the requirement of constructing a computer-aided diagnosis system.A pre-registration feature extraction method is proposed in this paper.Scattered point interpolation is used in atlas creation to obtain the features of integer points and lattice interpolation（bilinear interpolation） in lesion detection to obtain the features of non-integer points.The method is more accurate in describing gray features and texture features of images than that of afterregistration feature extraction.Experiments verified that the proposed method improved the detection rate and the detection accuracy but with the increase of the false positive rate.Further studies,the lesion detection methods based on 3D volume data and the reversibility of non-rigid registration,are needed to decrease the false positive rate and make more improvement in higher detection accuracy.