Virtual Biopsy with the Eye - Using Machine Learning to Detect and Track Chronic Kidney Disease (CKD)

该项目的目的是使用全方位的视网膜成像技术-超宽场扫描激光检眼镜（UWF-SLO，该项目的工业合作伙伴Optos制造的相机），光学相干断层扫描（OCT）和OCT血管造影（OCTA）-来构建视网膜血管的详细3D图像。然后，使用机器学习，我将从UoE的CVS获取的丰富而全面的临床数据集中分析和分割这些3D图像，以获得血管健康的指标和标记。这种分割和分析的主要目的是通过测量脉络膜厚度和体积来检测和跟踪CKD（以前只测量了一组离散点的厚度）-基于我的主管团队在他们的论文“慢性肾脏疾病中的脉络膜视网膜变薄与炎症和内皮功能障碍的联系”中的工作-但是这种详细的视网膜分割和血管跟踪也将有望使其他这样的研究成为可能，并使我们能够改进现有的视网膜血管健康度量。CKD目前影响着全球6-11%的人口，能够跟踪进展和治疗效果对于改善患者生活质量至关重要。视网膜成像是非侵入性的并且相对快速，因此将是侵入性肾活检和对血液和尿液样本的缓慢实验室测试的优选临床替代方案，特别是如果该过程可以自动化，即通过使用机器学习训练分类器。

The aim of this project is to use the full range of retinal imaging techniques available - Ultra-Wide Field Scanning Laser Opthalmoscopes (UWF-SLO, cameras manufactured by Optos, the industrial partner of this project), Optical Coherence Tomography (OCT), and OCT-Angiography (OCTA) - to build a detailed 3D image of the retinal vasculature. Using machine learning, I will then analyse and segment these 3D images from a rich and thorough clinical dataset acquired by CVS here at UoE for metrics and markers of vascular health. The primary aim of this segmentation and analysis is to detect and track CKD by measuring choroidal thickness and volume (previously only thickness at a discrete set of points was measured) - building on the work of my supervisor team in their paper "Chorioretinal thinning in chronic kidney disease links to inflammation and endothelial dysfunction" - but this detailed retinal segmentation and vascular tracking will also hopefully enable other such investigations, and enable us to improve existing retinal vascular health metrics. CKD currently affects 6-11% of the world's population, and being able to track progression and treatment efficacy is important to improve patient quality of life. Retinal imaging is non-invasive and relatively quick, thus would be a preferable clinical alternative to invasive kidney biopsies and slow lab tests on blood and urine samples, particularly if the process could be automated i.e. by training a classifier using machine learning.