| 谭青,丁亚军.基于GRNN的HIFU治疗中生物组织变性自动识别方法[J].中国医学影像技术,2020,36(6): |
| 基于GRNN的HIFU治疗中生物组织变性自动识别方法 |
| Automatic recognition of biological tissue degeneration in HIFU treatment based on GRNN |
| 投稿时间：2019-06-18 修订日期：2020-06-14 |
| DOI： |
| 中文关键词: HIFU 广义回归神经网络 特征选择 P值 欧几里得距离 |
| 英文关键词:HIFU Generalized regression neural network Feature selection P-value Euclidean distance |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| 摘要点击次数: 1590 |
| 全文下载次数: 636 |
| 中文摘要: |
| 目的 探讨一种基于广义回归神经网络 (Generalized Regression Neural Network，GRNN)的高强度聚焦超声（High intensity focused ultrasound，HIFU）治疗所致生物组织变性的自动识别方法。方法 提取超声减影图像的灰度-梯度共生矩阵及灰度差分中18个特征参数，分别利用P值假设验证法、欧几里得距离（欧氏距离）判定法对特征参数进行二次选择，最后选取欧氏距离靠前的2个最佳特征参数随机组合输入GRNN进行生物组织变性自动识别。结果 分别结合GRNN的总识别率：一次筛选掉的平均值、对比度分比二次筛选时欧氏距离靠后的小梯度优势、能量低，二次筛选得到的灰度分布不均性、梯度分布不均匀性分别约91.18%、90.20%，2个最佳特征参数组合高达98.04%。结论 经筛选后的特征参数组合输入GRNN的总识别率显著提高。 |
| 英文摘要: |
| Objective To explore an automatic identification method of tissue degeneration induced by high intensity focused ultrasound (HIFU) treatment based on generalized regression neural network (GRNN). Methods 18 characteristic parameters of gray-gradient co-occurrence matrix and gray difference of ultrasound subtraction image were extracted.These characteristic parameters were selected twice by P-value hypothesis verification method and Euclidean distance determination method by orderly,finally, two best ones were randomly combined with GRNN for automatic identification of biological tissue degeneration. Results About the total recognition rate of combined with GRNN：average value and contrast of the first screening out is lower than small gradient advantage and energy which are further-back Euclidean distance during the second screening. After the second screening, the non-uniformity of gray distribution is about 91.18%, the non-uniformity of gradient distribution is about 90.20%，2 best characteristic parameters combined is as high as 98.04%. Conclusion The total recognition rate of the selected characteristic parameters were combinedly input GRNN is significantly improved. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
