基于多模态特征优化的大鼠胫骨骨质疏松识别

Multimodal feature optimization for recognition of shin osteoporosis in rats

目的利用形状和纹理特征优化结果对低倍镜下大鼠胫骨切片进行模型识别。方法 20只雌性SD大鼠随机分为假手术组和去卵巢组,胫骨上段制成不脱钙骨切片,于2.5倍显微镜下拍摄骨小梁区域照片。利用骨小梁区域二值图像提取差异显著的13个形状特征、灰度图像差异显著的15个纹理特征,通过信息增益方法进行特征筛选,并用筛选后特征对假手术组和去卵巢组图像进行模式识别。结果利用形状和纹理特征共同识别的准确率、特异度明显高于仅用纹理特征识别。结论在图像质量不理想情况下,基于多模态特征优化方法可提供骨质疏松诊断依据。

Objective To apply the morphological and textural feature optimization for model recognition of rat shin sections at low magnification. Methods Twenty female SD rats were randomized to sham and ovariectomized groups. The photographs of bone trabeculaes were taken from the non-decalcification sections of the upper shin. The features were screened by information gain method from 13 morphological features of binary images and 15 textural featrues of grayscale images with statistical differences between Sham and ovariectomized groups. The model recognition was performed between two groups using the screened features. Results The accuracy and specificity were better using morphological and textural features than using textural features. Conclusion Multimodal feature optimization can provide the diagnostic evidence for osteoporosis under the unsatisfactory image quality.