Quantitative evaluation for ultrasonographic internal structures in mass lesions of the maxillofacial region

本研究的目的是建立一种新的基于定量数据的纹理分析方法，用于评价眼-脑系统特征的均匀性和粗糙度。构造了两种数值模型，研究了基本像素（基元）或其簇（面积）的大小和值对视觉均匀性和粗糙度的影响。在CRT上显示幻象图像，并提供给11名观察者。基元的大小和面积对均匀度和粗糙度的视觉评价结果有很大影响。灰度值的变化对图像质量的影响不大。均匀度的观察者内变异显著大于粗糙度。在此基础上，提出了一种新的基于眼-脑系统特征的纹理分析方法（游程分析）。获得了两个定量的质地指数（非均匀性指数和粗糙度指数）。利用该方法对数字体模进行了分析，并将所得指标与体模的视觉评价指标进行了比较。游程面积分析法能正确区分图像的均匀性和粗糙度。此外，从该方法获得的视觉评价结果和定量数据之间的相关性很高。最后，应用该方法对临床声像图进行分析。定量纹理指数恰当地显示了肿块病变的内部结构。这一事实表明，该方法可以为纹理的视觉评价提供定量数据。

The purpose of the project was to construct a new texture analysis method based on the quantitative data for the eye-brain system characteristics in evaluating homogeneity and coarseness. Two kinds of numerical phantoms were constructed to evaluate the effects of sizes and values of the elementary pixels (primitives) or their clusters (areas) on visual homogeneity and coarseness. Images of phantoms were displayd on CRT and presented to 11 observers. The size of primitives and areas greatly effected the results of visual evaluation for homogeneity and coarseness. The effects of the changes of gray value were only a little. Intraobserver variation for homogeneity was significantly larger than that for coarseness. Then, new texture analysis method (run area analysis) based on the eye-brain system characteristics has been developed. Two quantitative textural indices (heterogeneity index and coarseness index) were obtained. Numerical phantom was analyzed using this method, and the indices obtained were compared those of visual evaluation of the phantoms. Run area analysis method properly differentiated homogeneity and coarseness of the image. In addition, correlation between the results of visual evaluation and quantitative data obtained from this method was high. Finally, clinical ultrasonograms were analyzed with this method. Quantitative textural indices properly demonstrated the internal structures of the mass lesions. This fact indicates that this method could be an aid for visual evaluation for texture by offering quantitative data for it.