Planning and optimization algorithms for image-guided medical procedures.
详细信息
- 作者：Alterovitz ; Ron.
- 学历：Doctor
- 年：2006
- 导师：Goldberg, Ken
- 毕业院校：University of California
- 专业：Engineering, Biomedical.;Engineering, Industrial.;Computer Science.
- CBH：3253748
- Country：USA
- 语种：English
- FileSize：11757246
- Pages：151
文摘
Exciting advances in medical imaging are enabling clinicians to noninvasively examine anatomy and metabolic processes deep below the skin's surface. To effectively utilize this new wealth of digital information, new computational methods are needed to plan and optimize medical procedures on a patient-specific basis. In this dissertation, we combine ideas from robotics, physically-based modeling, and operations research to develop new planning and optimization algorithms for image-guided medical procedures.;We focus on four planning and optimization problems, each of which introduces new computational challenges and is subject to unique constraints imposed by the physician's treatment requirements, the patient's anatomy, and the physical limitations of medical equipment and devices. First, we develop an image registration method that explicitly considers tissue deformation when mapping targets between images acquired at different times. Results using prostate medical images indicate a statistically significant improvement in registration accuracy compared to previous methods. Second, we develop a motion planning algorithm for traditional needle insertion procedures to correct for tissue deformation caused by forces exerted by the needle. The method, which combines a finite element model of soft tissue with numerical optimization, is applicable to a variety of minimally invasive procedures, from biopsies to cancer treatments such as cryotherapy and brachytherapy. Third, we develop a nonholonomic motion planning algorithm that explicitly considers uncertainty in motion, and we apply it to a new class of highly flexible bevel-tip needles that can be steered to targets in soft tissue previously inaccessible to stiff needles. The algorithm combines geometric planning with Markov Decision Processes and Dynamic Programming. Results indicate that traditional shortest paths do not maximize the probability of successfully acquiring the target when the needle's response to controls is not known with certainty. Fourth, we consider dose optimization for high-dose-rate brachytherapy cancer treatment, a medical procedure in which physicians place radioactive sources in close proximity to cancer cells. We formulate the close optimization problem as a linear program, enabling the fast computation of mathematically optimal solutions and the statistical validation of internationally used clinical planning software. Overall, these results advance the development of new planning and optimization algorithms for image-guided medical procedures.