Artificial intelligence Optical Coherence Tomography Guided Deep Anterior Lamellar Keratoplasty (AUTO-DALK)

项目摘要 现代眼科手术是由熟练的外科医生通过手术显微镜进行的， 利用徒手技术和手动操作的精密显微仪器，其结果往往是 受限于外科医生的技术水平和经验。为了克服这些人为因素，我们聚集了 一个跨学科的团队，包括临床科学家和眼科医生，光学设备科学家和医疗 机器人工程师将我们实验室中现有的和正在开发的技术转化为精确的，“深度， 学习”人工智能（AI）引导的机器人眼科手术设备，用于精确的自动化深前 板层角膜移植术（AUTO-DALK）。 DALK是内皮功能正常的角膜疾病的一种非常有吸引力的治疗方法。然而，在这方面， 从技术角度来看，这一程序非常具有挑战性，而且耗时，限制了其接受程度 在角膜外科医生中。该手术最具挑战性的方面与基质分层有关 Descemet’s membrane（DM）通常称为“大气泡”的程序用于将基质与 DM采用深层基质内气压注射。然而，即使是经验丰富的外科医生也很难准确地 进行注射。DALK最常见的并发症是针插入深度过大 导致后弹力层穿孔，需要转换为全层穿透性角膜移植术 其恢复期长得多，并且由于排斥而导致移植失败的风险更高。报告的死亡率 后弹力膜穿孔率初学者为31.8%，有经验者为11.7%。 此外，供体和受体角膜之间的界面混浊是由角膜移植引起的常见问题。 针插入深度不足和未能去除宿主间质组织，这导致 术后视力这些问题直接关系到目前的外科实践不能 实时精确评估角膜层内工具提示的深度。 在这里，我们将建立在我们以前和正在进行的工作，在强大的光纤共路光学 基于卷积神经网络（CNN）机器人的相干断层扫描（CP-OCT）和AI引导系统 显微手术工具，使临床医生能够在微米级精确引导手术工具。拟议的预算- DALK手术工具系统能够进行一维实时深度跟踪、运动补偿， 检测仪器与组织的早期接触，使临床医生能够精确、安全地执行DALK。 该工具将建立在一个手持平台上，该平台将由CP-OCT探头、环钻和显微注射器组成， 允许精确和安全地切除角膜前部，直至DM 我们假设AI-OCT提供智能可视化和深度控制的最佳角膜 切割和组织跟踪将以比手动更好的准确性和效率执行DALK任务， 进行环钻切割和“大气泡”肺剥离。

PROJECT SUMMARY Contemporary ocular surgeries are performed by skilled surgeons through operating microscopes, utilizing freehand techniques and manually operated precision micro-instruments, where the outcomes are often limited by the surgeon's skill levels and experiences. To overcome these human factors, we have assembled an interdisciplinary team including a clinician-scientist and eye surgeon, an optical device scientist and medical robotic engineers to translate existing and developing technologies in our laboratories into precision, “deep- learning” artificial intelligence (AI) guided robotic ocular surgical devices for precise automated Deep Anterior Lamellar Keratoplasty (AUTO-DALK). DALK is a highly attractive treatment of corneal disease with normally functioning endothelium. However, the procedure is unusually challenging from a technical perspective and time-consuming, limiting its acceptance among corneal surgeons. The most challenging aspect of the procedure is related to the delamination of stroma from Descemet's membrane (DM). A procedure, commonly called “Big Bubble” is used to separate stroma from DM using deep intrastromal pneumatic injection. However, even experienced surgeons have difficulty precisely placing the injection. The most common complication of DALK is the excessive depth of the needle insertion resulting in Descemet's membrane perforation requiring conversion to full-thickness penetrating keratoplasty with its much longer recovery period and a higher risk of graft failure from rejection. The reported rates of Descemet's membrane perforation for beginner and experienced surgeons are 31.8% and 11.7% respectively. In addition, interface haze between the donor and recipient cornea is a common problem caused by the insufficient depth of needle insertion and failure to remove the host stromal tissue, which results in loss of postoperative visual acuity. These problems relate directly to the inability of the current surgical practice to precisely assess the depth of the tooltips inside the cornea layer in real-time. Here we will build upon our previous and ongoing work in robust fiber optic common-path optical coherence tomography (CP-OCT) and AI-guide system based on convolutional neural network (CNN) robotic microsurgical tools that enable clinicians to precisely guide surgical tools at micron scale. The proposed AUTO- DALK surgical tool system is capable of one-dimensional real-time depth tracking, motion compensation, and detection of early instrument contact with tissue, which enables clinicians to perform DALK precisely and safely. The tool will be built on a handheld platform that will consist of CP-OCT probe, trephine and microinjector that allows precise and safe removal of the anterior section of cornea down to DM We hypothesize that AI-OCT providing intelligent visualization and depth controlled optimal cornea cutting and tissue tracking will perform the task of DALK with better accuracy and efficiency over the manually performed trephine cutting and “Big Bubble” pneumodissection.