基于BP神经网络的医学图像分割新方法

发布时间：2018-01-05 02:19

本文关键词：基于BP神经网络的医学图像分割新方法　出处：《计算机科学》2017年S1期 　论文类型：期刊论文

【摘要】：对于医学图像而言,其分割结果的准确性对医生诊断病情并给出正确的治疗方案至关重要。应用传统的BP神经网络对医学图像进行分割,存在对初始权重值敏感、学习速率固定、收敛速度慢和易陷入局部极小值等问题。因此,提出了一种基于改进的粒子群优化算法的BP神经网络的医学图像分割方法。首先,应用粒子群优化算法与BP神经网络的映射关系,通过粒子群强大的搜索功能找到最佳适应函数,使对应的BP神经网络的均方误差达到最小值,克服了BP神经网络产生多个局部最小值的可能;其次,确定粒子的最佳位置后,在BP神经网络学习中获得最合理的权值和偏置值,以提高网络的收敛速度;最后,BP神经网络经反复训练后,获得最佳输出值,并计算阈值,通过阈值来分割图像区域。实验结果表明,利用改进的算法能够得到更清晰的图像分割效果,提高了图像的分割精度,对临床的诊断也具有重要参考意义。
[Abstract]:For the medical images, the accuracy of the segmentation results for doctors to diagnose the disease and treatment are crucial for correct. BP neural network is applied to the traditional medical image segmentation, is sensitive to the initial value of weight, the learning rate is fixed, the problem of slow convergence and easy to fall into local minimum. Therefore, a medical image segmentation method of improved BP neural network based on particle swarm optimization algorithm. Firstly, the mapping relationship between optimization algorithm and BP neural network using particle swarm particle swarm, through the search function to find the best powerful fitness function, the mean square error of BP neural network corresponding to the minimum, has overcome the BP nerve network multiple local minimum possible; secondly, to determine the best position of the particle, to obtain the most reasonable weights and bias values in BP neural network learning, to improve the network convergence Speed; finally, the BP neural network after repeated training, get the best output value, and calculate the threshold to image segmentation by threshold. Experimental results show that the algorithm can get the image segmentation effect more clearly, improve the accuracy of image segmentation, the clinical diagnosis also has important reference significance.

【作者单位】：包头医学院计算机科学与技术系;
【基金】：内蒙古自治区自然科学基金项目(2016MS0601) 包头医学院科学研究基金项目(BYJJ-QM 201637)资助
【分类号】：TP183;TP391.41
【正文快照】： 本文受内蒙古自治区自然科学基金项目(2016MS0601),包头医学院科学研究基金项目(BYJJ-QM 201637)资助。1引言医学图像对疾病的诊断、选择治疗手段和方法起到决定性的作用,能够最大限度地为医生提供患者的信息。医学图像的分割是按照纹理、颜色等特征,把图像分解成不同特性的区

【相似文献】