骨X射线图像纹理的分形特征分析

Fractal Feature Analysis on Texture of Bone X-Ray Images

将骨X射线图像的投影功率谱视为分数布朗运动 ,由此提出一种新的分形维数的定义 ,即沿不同方向 ,对功率谱密度空间的投影功率谱运用线性递归方法计算方向分形维数 ,并将最大和最小方向分形维数的比值定义为一项各向异性特征 ,该比值反映了骨小梁模式各向异性的程度 .实验结果表明 ,这项各向异性指数对两类骨质疏松X射线图像有较好的分辨力 ,可用于模式分类 .结合骨小梁模式的谱特征 ,定义的方向分形维数以及由其得到的各向异性指数所形成的特征集应用于骨质疏松的定量分类 。

The projection power spectrum of bone X ray images along each orientation can be considered as fractional Brownian motion. A new fractal dimension is developed. For different orientations, the directional fractal dimension is calculated by applying a regression method to the projection power spectrum in power spectrum density space. The ratio of maximum and minimum directional fractal dimensions in polar coordinates defines an anisotropic texture feature. The ratio reflects the degree of anisotropy of trabecular patterns, and is successfully used in our pattern classification. Numerical results show that the anisotropic index has good discrimination on two different grades of osteoporosis. The anisotropic index and the spectrum features of bone compose a feature sets, and are further applied to quantitative classification of osteoporosis bone. This indicates the potential applicability of our method to the objective assessment of osteoporosis.