Technology development for multiparametric and multimodality image guidance

拟议的研究目标是开发基于“信号和图像处理”和“机器学习”的技术工具，以在多模式和数据驱动的框架中推进基于MRI和超声的癌症检测组织分型。MRI模态揭示了表征灌注和扩散的生理特征，而基于超声的参数与诸如弹性、粘度、幂律、弹性各向异性、散热和声速的物理性质有关。磁共振和超声波可以一起提供组织的物理和生理学特征。该研究计划有两个主要的工程组成部分。一个组件处理用于组合MRI和超声的准确配准的问题。MR超声配准的当前方法是基于表面的并且是主观的，这是由于依赖于根据MR和超声图像对解剖器官（例如前列腺）的轮廓勾画。我计划开发一种新的图像相似性度量的基础上最大的信息系数，使基于强度的注册。第二个组成部分建立了一个数据驱动的方法，以增强MRI中的组织分型过程，并在合并的MR-超声放射学轮廓。这种方法依赖于从数据中学习癌症阶段。它消除了物理或生理建模和推导分析解决方案来描述癌症进展的必要性。该方法将应用于计算动态增强MRI的组织分型特征，并建立基于计算机的诊断系统。所提出的技术存在许多临床应用。在前列腺癌中，迫切需要一种基于图像的非侵入性预后方法。在乳腺癌中，用于疾病亚型的基于图像的解决方案可以实现患者特异性治疗。两个正在进行的研究生职位和一个本科生暑期合作职位将在我的研究计划中创建。一个研究生职位将专注于基于图像的计算机辅助诊断，重点是统计机器学习，信号和图像处理。第二个职位将侧重于计算机辅助干预，重点是图像配准。

The proposed research targets developing technological tools based on "signal and image processing" and "machine learning" to advance MRI and ultrasound-based tissue typing for cancer detection, in a multimodal and data-driven framework. MRI modalities reveal physiologic features characterizing perfusion and diffusion, while ultrasound-based parameters are related to physical properties such as elasticity, viscosity, power law, elastic anisotropy, heat dissipation and speed of sound. Together, MR and ultrasound could provide a profile of physics and physiology of the tissue. The research program has two major engineering components. One component deals with the problem of accurate registration for combining MRI and ultrasound. Current methods of MR-ultrasound registration are surface-based and subjective due to dependence on contouring of an anatomical organ, for example prostate, from MR and ultrasound images. I plan to develop a new image similarity measure based on maximal information coefficient to enable intensity-based registration. The second component builds a data-driven method to enhance the process of tissue typing in MRI, and in the combined MR-ultrasound radiologic profile. This approach relies on learning the cancer stage from data. It removes the necessity for physical or physiological modeling and derivation of analytic solutions to describe cancer progress. The method will be applied both in calculation of tissue typing features from dynamic contrast enhanced MRI, and in building a computer-based diagnosis system. Many clinical applications exist for the proposed technics. In prostate cancer, a non-invasive image-based method for prognosis is desperately needed. In breast cancer, an image-based solution for subtyping of the disease can enable patient-specific therapy. Two ongoing graduate student positions and one undergraduate summer co-op position will be created in my research program. One graduate position will focus on image-based computer-aided diagnosis with emphasis on statistical machine learning, signal and image processing. The second position will focus on computer-aided interventions with focus on image registration.