基于径向基函数神经网络的膝关节磁共振T2 map数据分类

Knee joint magnetic resonance RBF neural network-based T2 map data classification

目的:设计一种膝关节骨性关节炎(OA)磁共振T2 map数据分类器,用于OA疾病分类。方法:通过磁共振成像(MRI)T2 mapping技术,采集46例膝关节MRI图像共计1380个数据,按膝关节软骨全器官磁共振成像评分(WORMS)分区方法提取10个亚区的T2值数据,以T2值数据为特征量进行数据挖掘,建立径向基函数(RBF)神经网络分类器,结合临床诊断结果实行对采集样本数据分类识别。结果:RBF分类器对于膝关节T2 map数据最终识别准确率为75%,体现了良好的OA数据分类效果。结论:基于直接确定法的RBF神经网络构造的膝关节OA分类器无需任何迭代,通过简单步骤就得到最优权值、合适的中心以及方差,适合作为OA的疾病分类器。

Objective： To design a knee Osteoarthritis classifier of magnetic resonance T2 mapdata, which used for OA disease classification. Methods： Collected 46 cases of knee image witha total of 1380 data by magnetic resonance imaging（MRI） T2 mapping technique, and extractedT2 value data of 10 Asian region based on articular cartilage whole organ magnetic resonanceimaging score（WORMS） partition. Then took the T2 value data as the characteristic quantityby data mining, and structured radial basis function（RBF） neural network classifier, combinedwith the clinical diagnosis to classify and recognize the data of collected sample. Results：The study finally found that RBF classifier reflected 75% of recognition accuracy rate, and itshowed good effect of OA data classification. Conclusion： The knee osteoarthritis RBF neural network classifier based on direct determination method can get the optimal weights, right center and variance by simple steps, without any iteration. We suggest that it is a classifier fit to OA disease.