Ultrasound-MRI Registration for Minimally Invasive, Radiation-Free Spine Surgery

医生们寻求微创和无辐射解决方案的技术创新。典型的脊柱外科手术依赖于两个主要的成像组件；术前成像：使用MRI或CT的高质量诊断技术；术中成像：使用透视、c臂或超声进行实时成像和可视化，用于指导手术，通常因其成本低或/和实时反馈而被选择。目前这些程序的问题是，它们严重依赖于人为因素，如医生的专业知识和经验，以执行两个图像集的认知注册。仅考虑辐射安全性和成本，超声是最佳选择，但超声成像很难解释，除非由非常有经验的专业人员执行。由于操作的徒手性质和噪声水平大大高于CT/MRI，因此很难识别图像中的目标位置。当使用透视或c型臂进行实时成像时，图像比超声更清晰，更容易发现病变，但在手术过程中患者和外科医生的辐射暴露是一个严重的问题。我们的建议是将实时二维超声成像与静态三维MRI成像相匹配，用于微创脊柱手术，同时仍然确保实时高质量的成像可视化。主要思路是主动引导超声探头扫描目标病变，并在MRI显示的点上进行配准，然后使用各种深度学习方法对超声图像进行特征检测/跟踪，进行详细和更高程度的配准。商业注册硬件/软件是可用的，但没有提供无辐射和微创的解决方案。微创手术将帮助患者更快地恢复，同时在手术过程中消除c型臂的使用将降低手术成本，但最大的好处来自于该过程的无辐射特性。该研究成果将促进超声成像在脊柱外科手术中的应用，为患者和外科医生提供一个更安全的环境。

Medical doctors look for technological innovations in minimally invasive and radiation-free solutions. Typical spine surgery relies on two main imaging components; (i) Preoperative imaging : A high quality diagnostic technique using MRI or CT and (ii) Intraoperative imaging: Real time imaging and visualization using either fluoroscopy, C-Arm or ultrasound, used for guidance of the procedure and typically selected for its low cost or/and real time feedback. Current problems with these procedures are that they rely heavily on human factors such as the physician's expertise and experience to perform a cognitive registration of the two image sets. Considering only radiation safety and cost, ultrasound is the best choice, but ultrasound imaging is difficult to interpret unless performed by very experienced professionals. It is difficult to identify the target location in the images due to freehand nature of the manipulation and noise levels which are substantially higher than CT/MRI. When fluoroscopy or a C-Arm are used for real time imaging, the images are clearer and it is easier to find lesions than with ultrasound, but radiation exposure of both patient and surgeon during surgical procedures is a serious problem. Our proposal is real-time 2D ultrasound imaging registered to static 3D MRI imaging for minimally invasive spine surgery while still ensuring real-time high quality imaging visualization. The main idea is to actively guide an ultrasound probe to scan the target lesion and register it at spots shown in the MRI and then perform feature detection/tracking on ultrasound images for detailed and higher degree registration using various deep learning methods. Commercial registration hardware/software is available, but nothing that provides radiation-free and minimally invasive solutions. Minimally invasive surgery will help patients to recover faster while operation costs will be lowered by eliminating C-arm use during surgery, but the greatest benefits derived from the radiation-free nature of the process. The research outcome will promote the use of ultrasound imaging in spine surgery, providing a safer environment for both patient and surgeon.