基于概率神经网络的放疗加速器多叶准直器系统故障识别诊断研究  

Research on fault identification and diagnosis of multileaf collimator system of radiotherapy accelerator based on probabilistic neural network

作　　者：杨皓文 罗吉[1] 高大地 邸元帅 YANG Haowen;LUO Ji;GAO Dadi;DI Yuanshuai(Department of Radiation Oncology,Fudan University Shanghai Cancer Center,Shanghai 200032)

机构地区：[1]复旦大学附属肿瘤医院放疗射治疗中心,上海200032

出　　处：《北京生物医学工程》2024年第1期78-82,87,共6页Beijing Biomedical Engineering

摘　　要：目的随着当前医院肿瘤患者放疗数量的日益增长,对放疗设备持续稳定治疗的要求越来越高。放疗直线加速器是放射治疗的主要设备,多叶准直器(multileaf collimator,MLC)是调强放疗最为频繁的系统之一,但其故障发生率较高,一旦发生故障宕机,不仅影响患者治疗效果,还会给医院带来经济损失。因此,快速准确识别并排除故障,对保障MLC系统的正常运行具有重要意义。本文提出一种基于概率神经网络(probabitistic neural network,PNN)的MLC系统故障识别诊断方法,为MLC系统的不同故障现象和类型提供维修依据。方法结合复旦大学附属肿瘤医院医科达放疗加速器故障维修经验及日常报错记录,整理分析MLC系统构成及常见故障现象共140例,统计研究常见故障下设备状态的各项参数数据。选取能够表征故障特征的信息作为输入向量和故障分类输出向量,用不同特征输入向量的组合代表不同的故障类型。数据归一化乱序处理后,创建PNN神经网络模型并进行训练。最后对比分析故障的实际分类和预测分类结果。结果通过分类结果对比和混淆矩阵可知,训练集样本一共98个,预测对比精确度为100%;测试集样本一共42个,预测对比精确度为97.619%,训练总时间为4.626 s。结论基于PNN概率神经网络的MLC系统故障识别诊断模型具有训练速度快、容错性好、识别诊断精准度高等优势。Objective With the increasing number of radiotherapy for cancer patients in hospitals,the requirements for sustained and stable treatment of radiotherapy equipment are getting higher and higher.Radiotherapy linear accelerator is the main equipment for radiotherapy,and multileaf collimator(MLC)is one of the most frequent systems for intensity modulated radiotherapy.However,its failure rate is high.Once a failure occurs,it will not only affect the treatment effect of patients,but also bring economic losses to hospitals.Therefore,it is of great significance to identify and eliminate faults quickly and accurately to ensure the normal operation of MLC system.In this paper,a fault identification and diagnosis method of MLC system based on probabitistic neural network(PNN)is proposed,which provides maintenance basis for different fault phenomena and types of MLC system.Methods Combined with the maintenance experience and daily fault records of Medical University Radiotherapy Accelerator in the Cancer Hospital Affiliated to Fudan University,140 cases of MLC system structure and common fault phenomena were sorted and analyzed,and the parameter data of equipment state under common faults were statistically studied.The information that can characterize the fault characteristics was selected as input vector and fault classification output vector,and the combination of different characteristic input vectors represented different fault types.After the data were normalized and disordered,PNN neural network model was established and trained.Finally,the actual fault classification and prediction classification results were compared and analyzed.Results Through the comparison of classification results and confusion matrix,we found that there were 98 samples in the training set,and the accuracy of prediction comparison was 100%.There were 42 samples in the test set,the accuracy of prediction and comparison was 97.619%,and the total training time was 4.626 s.Conclusions The fault identification and diagnosis model of MLC system base

关 键 词：放疗加速器 多叶准直器 概率神经网络 故障识别 精确度 

分 类 号：R318[医药卫生—生物医学工程]