In Vitro Monitoring: The Next Revolution in IVF Technology

由于推迟生育，生育治疗继续大幅增加。不幸的是，女性的生殖能力从第二年和第三年开始下降，因此到40岁时，生育力显著下降，流产的可能性很高。辅助生殖技术包括体外处理卵子和胚胎的生育治疗。最著名的一种是体外受精，通过超刺激卵巢产生多个卵子进行体外受精。然后对受精卵(胚胎)进行孵化，直到它们达到囊胚期(受精后约5天)，并选择植入。在孵化过程中，根据胚泡的内部形态结构对胚泡进行分级，从而确定最佳候选胚泡。这一评估阶段是体外受精治疗中的咬合限制，通过使用工程科学的新技术可以进行革命性的研究。拟议的研究计划将研究信号处理、计算机视觉和机器学习策略，以挖掘与植入潜力相关的形态质量的延时显微胚胎图像。一个新的计算框架将基于分层体系结构开发--从单帧到帧序列。在帧级别，形态特征和结构成分将通过使用模型和训练数据集的手动和深度学习算法自动识别。胚胎的结构成分是高度不同的，从一个阶段到另一个阶段表现出戏剧性的转变，算法将相应地设计。在序列级别，统计模型和跟踪过滤器将从帧级别链接观测和计算的度量，以创建累积分数。这将识别和量化胚胎发育的动态和形态动力学特征。通过这项拟议的研究计划，将首次有可能创建在整个孵化过程中依赖于形态特征的临床适用的计算工具。这将使新的科学假设能够测试各种形态和形态运动胚胎特征与妊娠结局之间的联系。

Fertility treatment has continued to increase dramatically as a result of delayed childbearing. Unfortunately, female reproductive capacity declines from the 2nd and 3rd decades of life so that by the age of 40 there is a significant reduction in fertility with a high chance of miscarriage. Assisted Reproductive Technology includes fertility treatments that involve handling eggs and embryos outside the body. The best known one is In Vitro Fertilization where the ovaries are hyper-stimulated to produce multiple eggs for external fertilization. The fertilized eggs (embryos) are then incubated until they reach the blastocyst stage (about 5 days after fertilization) and selected for implantation. Optimal candidates are identified during the incubation process by grading the blastocysts based on their internal morphological structures. This evaluation stage is the biting constraint in IVF treatment and can be revolutionized though the use of new technology from engineering science.The proposed research programme will investigate signal processing, computer vision, and machine learning strategies to mine time-lapsed microscopic embryo images for morphological qualities related to implantation potential. A new computational framework will be developed based on a hierarchical architecture - from single frames to frame sequences. At the frame level, morphological features and structural components will be identified automatically through hand-crafted and deep learning algorithms using models and training datasets. The structural components of an embryo are highly variant and exhibit dramatic transitions from one phase to another, and the algorithms will be designed accordingly. At the sequence level, statistical models and tracking filters will link the observations and computed metrics from the frame level to create cumulative scores. This will identify and quantify the dynamic and morpho-kinetic characteristics of the embryonic developments. Creating clinically applicable computational tools that rely on morphological features throughout the incubation will be made possible for the first time through this proposed research programme. This will enable new scientific hypotheses to be tested about the link between various morphologic and morpho-kinetic embryo characteristics and pregnancy outcomes.