Ultrax: Real-time tongue tracking for speech therapy using ultrasound

言语声音障碍(SSD)是儿童时期最常见的沟通障碍，影响到6.5%的英国儿童，即每个教室有2名儿童。固态硬盘使人们难以与同龄人交流和融入社会，但对大多数类型的固态硬盘的干预效果很弱。言语、语言和沟通障碍(SLCD)是目前英国政府的一项重要优先事项，2011年被指定为国家言语、语言和沟通年。最近的一份政府报告(伯考，2008年)强调需要制定一项研究方案，以加强SLCD儿童和青年的证据基础。我们的研究计划旨在通过开发有助于评估、诊断和治疗SSD的技术来满足这一需求。目前在言语和语言治疗中，技术支持是稀少的。现有的技术运行昂贵或操作复杂，因此没有在临床实践中采用。该项目将开发技术(Ultrax)，将超声波转化为专门设计的舌头成像设备，以提供舌头运动的实时视觉反馈。对SSD的大多数干预在很大程度上依赖于听觉技能；客户必须听取自己的作品并对其进行修改。然而，有了Ultrax，拥有SSD的人实际上将能够看到自己舌头的运动，并使用这些信息来修改他们的语音。通过将标准的医用超声波探头放在下巴下面，已经可以捕捉舌头的运动。超声波有可能提供有关非典型语音的强大信息，并使说话者能够修改自己的错误发音。然而，图像是颗粒状的，信息(特别是关于舌尖的信息)经常丢失，图像很难理解。我们将通过利用关于可能的舌形和运动范围的先验知识来改进这一图像，以便在跟踪超声图像序列中的舌形轮廓时提供有价值的约束。我们将应用舌头模型来解决这个问题，利用显式的基于序列的优化来动态跟踪和平滑时间。我们将使用这项技术来增强超声图像，将它们转换为动态、实时的舌体运动2D视频，我们假设这些图像将A)更容易被儿童理解B)将可见舌形的范围从仅元音和/r/扩大到包括/t/、/k/、/ch/和其他通常是治疗目标的辅音Ultrax将被用于为SSD患者提供生物反馈治疗，并通过比较治疗前后的舌形来客观评估进展情况。我们将收集12名成年人(超声波和核磁共振)和90名小学生(超声波)的舌部运动的超声波和核磁共振图像的大型数据库，以此作为舌部轮廓模型的基础并测试其性能。同时，我们将90名儿童分成3组，并记录每组对以下3种超声显示的反应：1.原始的、平扫的、超声的；2.添加解剖背景的平扫超声；3.在本项目中开发的完全增强的超声。研究人员将评估孩子们模仿舌头形状和动作的能力，以确定他们是否发现增强后的图像比未增强后的图像更容易理解，从而提高了使用超声波进行生物视觉反馈的能力。我们将对9名患有SSD的儿童(每种显示器3名儿童)进行超声波治疗，使我们能够评估治疗过程中出现的实际问题，并为未来的临床试验铺平道路。在我们的研究项目结束时，我们将开发一种新的视觉反馈工具(Ultrax)的基础，供言语和语言治疗师用于SSD的诊断和治疗。

Speech Sound Disorders (SSDs) are the most common communication impairment in childhood, affecting 6.5% of all UK children, or 2 children in every classroom. SSDs make it difficult for people to communicate with peers and integrate with society, yet the efficacy of interventions for most types of SSDs is weak. Speech, Language and Communication Disorders (SLCD) are a key UK government priority at present, with 2011 designated the national year of speech, language and communication . A recent government report (Bercow, 2008) highlighted the need for a programme of research to enhance the evidence base for children and young people with SLCD. Our programme of research aims to fulfill this need by developing technology which will aid the assessment, diagnosis and treatment of SSDs. Currently in Speech and Language Therapy, technological support is sparse. Technologies that do exist have been expensive to run or complicated to operate and hence not adopted in clinical practice. This project will develop technology (Ultrax) to turn ultrasound into a tongue imaging device specifically designed to provide real-time visual feedback of tongue movements. Most interventions for SSDs rely heavily on auditory skills; clients must listen to their own productions and modify them. However, with Ultrax people with SSDs will actually be able to see the movements of their own tongues and use this information to modify their speech. It is already possible to capture tongue movements by placing a standard medical ultrasound probe under the chin. Ultrasound has the potential to provide powerful information about atypical speech and to enable speakers to modify their own incorrect articulations. However, the image is grainy, information (especially about the tongue tip) is often lost and the image is difficult to interpret. We will improve this image by exploiting prior knowledge about the range of possible tongue shapes and movements in order to provide valuable constraints in tracking tongue contours in sequences of ultrasound images. We will apply a tongue model to this problem, making use of explicit sequence-based optimization for dynamic tracking and smoothing through time. We will use this technology to enhance the ultrasound images, transforming them into a dynamic, real-time 2D video of the tongue's movements which we hypothesize will be A) more easily understood by children B) extend the range of visible tongue shapes from only vowels and /r/ to include /t/,/k/,/ch/ and other consonants which are often targets for therapy Ultrax will be used to provide bio-feedback therapy for people with SSDs and to provide a means for objectively assessing progress by comparing tongue shapes before and after therapy. We will collect a large database of ultrasound and MRI images of tongue movements from 12 adults (ultrasound and MRI) and 90 primary school children (ultrasound) on which to base the model of tongue contours and to test its performance. At the same time, we will split the 90 children into 3 groups and record each group's response to one of 3 types of ultrasound display: 1. Raw, unenhanced, ultrasound 2. Unenhanced ultrasound with added anatomical context (e.g. the position of the teeth and roof of the mouth) 3. Fully enhanced ultrasound, developed in this project. The ability of the children to imitate tongue shapes and movements will be evaluated to determine whether they find the enhanced images easier to interpret than unenhanced images, leading to an improved ability use ultrasound for bio-visual feedback. We will trial ultrasound therapy with 9 children with SSDs (3 children for each type of display) enabling us to evaluate practical issues arising during therapy and pave the way for a future clinical trial. At the conclusion of our research project we will have developed the basis for a new visual-feedback tool (Ultrax) for Speech and Language Therapists to use in the diagnosis and treatment of SSDs.