【作者】 王鑫鑫；

【导师】 叶志前；

【作者基本信息】 浙江大学 ， 生物医学工程， 2011， 硕士

【摘要】 当前,我国医院信息化建设已经进入数字化阶段,医疗信息系统广泛应用于现代医院各科室,其主要目的是为了更加充分利用患者临床诊断信息,提高检查信息传递的速度,进而提高医院资源利用率；其中,医学影像诊断信息是当前医学临床诊断主要信息依据,医学影像诊断报告作为放射科医生与相关临床医师相互沟通的主要媒介,能够及时、快速、准确地生成对于优化医疗服务流程、提高医疗服务质量具有重要作用。然而,当前放射科医师在诊断过程中,注意力需要集中在对医学影像的观察、分析等方面,而无暇顾及诊断报告的录入,不能够实现实时输入诊断报告。在特殊情况下(比如急诊、患者数量多),要实现对患者进行及时、准确的诊断治疗,医学影像诊断报告必须在限定时间内完成,因此,为了高效率地生成医学诊断报告,必须应用现代计算机信息技术突破传统医疗报告输入或记录模式的限制。当前,语音识别技术日益成熟,极有可能成为下一代操作系统的主要人机交互方式,将语音识别技术应用于放射科影像诊断报告的录入过程,将大大减少医生做出诊断的时间,从总体上提高医院收容处理病人的能力,同时大大缩短病人的无效等待时间,增加病人的满意度。本文以隐马尔科夫原理为基础,设计与实现专门应用于放射科影像诊断报告录入环节的分布式语音识别系统。首先,本文详细讨论了分布式语音识别相关的关键技术,主要包括HMM的核心思想与基本算法,并从系统构建的角度,进一步介绍了分布式语音识别系统客户端所涉及的语音端点检测与参数提取技术、系统客户端与服务器端之间语音数据传输协议以及服务器端主要的语音识别搜索技术。其次,详细阐述放射科信息系统中分布式语音识别系统的设计,从系统需求出发,构建系统整体框架,设计客户端语音数据处理模块、语音特征数据传输模块与服务器端语音识别模块,并针对系统整体性能需求,提出保证措施。最后,实现了放射科信息系统中分布式语音识别系统,并对实验结果进行了分析。实验结果表明,在实验室环境条件下,针对特定人的专门用于放射科影像报告录入的语音识别系统识别效果较为理想,能够与IBM Viavoice9.1软件的识别效果相媲美。更多还原

【Abstract】 At present, the hospital information construction in China has entered into the digital stage. The medical information system is extensively applied, the main purpose of which is to better take advantage of patient diagnostic information and to improve the transmission speed of inspection information, thereby increasing the utilization of hospital resources. Apparently, the medical image diagnostic information is one of the most important sources of medical information.As the key communication medium between Radiologist and related clinical radiologist, the medical image diagnostic reports must be generated timely and accurately. However, in the diagnostic process, the radiologist needs to focus on the medical image analysis, which leads to the diagnostic report cannot be made in real time. In order to provide timely and accurate diagnosis and treatment for some special occasion, such as the emergency, the medical diagnostic report must be completed within a limited time. Therefore, the modern computer information technology must be applied to break through the limitations of traditional medical report or record input mode. Currently, speech recognition technology has become more sophisticated, and it is highly likely to become the main human-computer interaction method of the next generation operating system. Introducing speech recognition technology into the entry process of the radiology diagnostic imaging report, will greatly reduce the time doctors make a diagnosis. Overall, it will improve hospitals’abilities of receiving and treating patients, while greatly reducing the valid waiting time for patients and increasing their satisfactions. This thesis designs and implements a specialized distributed speech recognition system used in generating radiology diagnostic imaging reports. Firsty, the paper discusses related key technologies of the distributed speech recognition, including the core idea of HMM and basic algorithms. The author illustrates further information on the front-end distributed speech recognition system involved in speech endpoint detection, parameter extraction technology, voice data transmission protocol between the client and server-side as well as the major server search and identify technology. Secondly, the distributed speech recognition system designed in radiology information systems is given in detail, including the system requirements analysis, setting up the system overall framework and designing front-end voice and data processing module, voice characteristics of data transmission module and server-side speech recognition module. Finally, a distributed speech recognition system in radiology information system is accomplished, and the experimental results are analyzed. The results show that in the laboratory environmental conditions, the recognition effect of speech recognition system dedicated to a specific person’s entry about Radiology Imaging Reporting is relatively ideal, and it’s comparable with the effect of IBM Viavoice 9.1 system.更多还原