颅内声压干扰下的病变医学图像区域识别

Lesions Area Identification of Medical Images under Disturbance of Intracranial Pressure

研究病变医学图像识别准确率优化问题。病变医学图像受到颅内声压的干扰,在采集过程中很难形成准确的特征属性提取,造成病变区域特征模糊化。传统算法对随机声压干扰下的病变医学图像特征很难形成有效的约束,造成病变区域识别精度下降。为了提高颅内声压干扰下病变医学图像的识别准确率,提出一种灰度共生矩阵和鲶鱼粒子群优化神经网络的病变医学图像识别算法(CFPOS-BP)。首先采用灰度共生矩阵提取病变医学图像特征,然后将特征输入到BP神经网络进行学习,通过粒子群优化算法优化BP神经参数,并引入"鲶鱼"效应克服粒子群算法存在的局部最优缺陷,最后采用具体病变医学图像数据库对算法性能进行仿真测试。仿真结果表明,相对于传统病变医学图像识别算法,CFPOS-BP可以获得更优的病变医学图像识别准确率,提高了病变医学图像识别准确率和识别效率。

This paper studied the optimization problem of recognition accuracy of lesions medical images. In the process of gathering lesions medical images, sound intracranial pressure can cause interference. It is difficult to form an accurate characteristic attribute extraction, resulting in lesion area features blurred. In order to improve the recog- nition accuracy rate of the lesions medical images under intracranial sound pressure interfere, we proposed a feature extraction method for the lesion medical images based on GLCM and catfish particle swarm optimization neural net- work algorithm （CFPOS- BP）. Firstly, the characteristics of lesion medical images were extracted with GLCM and the results were input to the BP neural network learning. Then the particle swarm optimization algorithm was applied to optimize BP neural parameters. Meanwhile, the ＂catfish effect＂ was introduced to overcome the local optimum de- fect of particle swarm optimization. Finally, the simulation tests were made with medical image database of specific lesions. The simulation results show that compared with traditional lesions in medical image recognition algorithm, the CFPOS - BP can get better recognition accuracy of pathological change medical image. It can improve the recog- nition accuracy of pathological change medical image as well as the recognition efficiency.