A holistic approach to identifying functional units of tongue motion during speech

点击翻译按钮获取中文摘要

PROJECT SUMMARY Oral cancers have the seventh highest incidence, with roughly 51,540 new cases and 10,030 cancer- related deaths expected to occur in 2018. Although a variety of treatment methods are available, the death rate is higher than that for most cancers with five-year rates of about 50 percent. The most frequently used treatment method, glossectomy surgery, involves the surgical removal of tumors and surrounding tissues, and the addition of grafted tissues, often followed by radiotherapy. Although tongue cancer and its treatment have debilitating effects on speech, the impact of varying degrees of resection and reconstruction on the formation of functional units in speech has remained poorly understood. In order to produce intelligible speech, a variety of local muscle groupings of the tongue—i.e., functional units—emerge and recede rapidly and nimbly in a highly coordinated fashion. Therefore, understanding the formation of functional units that are critical for speech production can provide substantial insights into normal, pathological, and adapted motor control strategies in controls and patients with tongue cancer for novel therapeutic, surgical, and rehabilitative strategies. One of the critical challenges in pre-operative surgical and treatment planning, as well as in post- operative evaluation for tongue cancer is the difficulty in developing objective and quantitative measures and in evaluating their functional outcome predictability. To address this, in this proposal, three integrated approaches will be used in in vivo tongue motion during speech to seamlessly identify the functional units and associated quantitative measures: multimodal MRI methods, multimodal deep learning, and biomechanical simulations. This will provide a convergent approach, thereby allowing us to (1) test hypotheses about the spatiotemporal basis of muscle coordination in a consilient way, and (2) develop objective quantitative measures that are required for understanding the complex biomechanical system as well as for predicting the functional outcomes after various reconstruction methods. The first proof of concept study published by the PI and the team identified the functional units of speech tasks using the sparse non-negative matrix factorization framework, in which the magnitude and angle of displacements from tagged MRI were used as our input quantities. With these advances in place, we will further incorporate muscle fiber anatomy from diffusion MRI and motion tracking from tagged MRI into our framework to yield physiologically and anatomically meaningful functional units. In addition, we will create a completely novel and integrated way of directly relating the functional units to tongue muscle anatomy, learning joint representation via a multimodal deep learning technique, and linking them to biomechanical simulations. Furthermore, 3D and 4D atlases will be utilized to identify objective and quantitative measures based on our functional units analysis. Taken together, the successful implementation of our integrated framework will identify functional units that can be used for research on tongue motion, for surgical planning, and for diagnosis, prognosis, and rehabilitation in a range of speech-related disorders.