文章摘要

文章信息/Info

Title:: Recognition of Osteoporosis for Mouse Shin in Low Power Microscope based on Multi-feature Fusion

作者:: 蔡洁1; 周珂1; 何文广1; 吴天秀2△; 王龙1; 1. 广东医科大学信息工程学院，湛江 524023；2. 广东医科大学基础医学院，湛江 524023

Author(s):: CAI Jie1; ZHOU Ke1; HE Wenguang1; WU Tianxiu2; WANG Long1; 1.School of Information Engineering Guangdong Medical University；2.School of Basic Medical Science,Zhanjiang, 524023，China

Keywords:: Classification and recognition; Coefficient of variation method; Texture analysis; shape analysis; Osteoporosis

摘要:: 低倍镜下纹理特征不清晰，并且相较于高倍镜更易受到方向和距离的影响。为此我们提出改进的纹理特征计算方法，结合形状分析方法，进行特征级融合以提高骨质疏松识别准确率。通过对纹理参数在方向和距离上的分析，发现相关和短游程矩对方向非常敏感，并且距离在等于3时各参数变化趋于稳定。基于变异系数法利用相关和短游程矩计算各个方向权重系数，使用距离等于3时的纹理数据得到最终纹理参数结果，融合形状参数，用线性支持向量机、K-最近邻分类算法和线性判别分析方法进行分类识别。采用加权纹理参数比常规未加权的纹理参数识别准确率高，同时融合了形状参数后识别准确率比仅用纹理参数高。线性判别分析识别率最高达到了92.3%。采用加权纹理融合形状参数的方法识别准确率显著提高，具有诊断应用价值。

Abstract:: For low resolution image, the texture features are more sensitive to direction and distance. An improved method to compute texture features was explored, and combined with shape analysis to raise the accuracy of recognition finally. By analyzing the influence of directions and distances on texture features, we found that correlation and short run emphasis were very sensitive to directions, meanwhile the differences of texture features (based on gray-level co-occurrence matrix) between distances were gradually stable when the distance equalsed to 3. Based on coefficient of variation method, the weighted coefficients were calculated with correlation and short run emphasis. The final weighted texture features were gotten when distance was 3. Combining with shape features, the classifiers of LSVM, KNN, LDA were used to assess Osteoporosis. The weighted texture features showed a higher accuracy than traditional texture features with the use of LSVM, KNN, LDA classifiers, respectively; and the fusion of texture and shape features had a significant improvement in classification with LSVM, KNN, LDA. Meanwhile, the highest recognition accuracy achieved 92.3%. The proposed weighted texture features combining with shape features provide a higher accuracy for recognition of Osteoporosis, and it is helpful to clinical diagnosis.