基于BP神经网络的CT图像骨皮质分割

目的在CT图像中通过对骨皮质的分割与测量,测定骨量、骨骼的几何形状以及骨强度,并计算相应的组织形态计量学参数。方法通过DCMTK解读CT图像,提取相应的图像信息。利用OpenCV对图像进行预处理,在感兴趣的区域(ROI)设置的基础上,提取图像的纹理特征作为输入向量;以对训练样本手工分割的结果作为导师信号,对BP神经网络进行训练;用训练好的网络对CT图像序列中的骨皮质进行分割,并对分割后的结果进行后处理及显示。结果骨皮质CT图像的神经网络分割效率符合实际应用的需求,分割结果形状明显,与周围组织区分度高,满足临床诊断需求。结论纹理特征明显的情况下,可以达到较为满意的分割效果。分割结果轮廓平滑,分割精度高、成功率高、适应性强;而且图像分割过程人工介入少,可以用于整套CT图像骨皮质的批量分割。不足之处在于此方法神经网络训练时间相对较长。

Bone cortex segmentation of CT images based on BP neural network

Objective To measure the bone mass, the shape of bones and the bone strength through segmentation of the bone cortex in CT images, and to calculate the corresponding parameters in histomorphometry. Methods CT images were first interpreted through the DCMTK to draw information of the corresponding images, then the OpenCV are used for preprocessing on the basis of ROI (range of interest), and the texture features of the image were extracted as the input vector. Results of the manual segmentation were used as the mentor signal to train BP neural network, which were then used for segmenting the bone cortex in a sequence of CT images. Results of the segmentation were further processed and displayed. Results The segmentation efficiency of the bone cortex in CT images through neural network met the needs of the practical application. The separation results showed an obvious shape of the bone cortex with easy distinguishing from the surrounding tissues, which could satisfy the demand of the clinical diagnosis. Conclusions When the texture features of the bone cortex are evident, this method can achieve a more satisfying segmentation effect with smooth contours, high segmentation accuracy and strong adaptability. With less artificial intervention in the process of the image segmentation, this method can be used for batch CT image segmentation of a complete set of the bone cortex. The inadequacy of the method lies in relatively longer training time demanded for the neural network training.