OLIVES

眼睛的临床诊断是通过多种数据方式进行的，包括标量临床标签，矢量化生物标志物，二维眼底图像和三维光学相干断层扫描 (OCT) 扫描。虽然临床标签，眼底图像和OCT扫描是仪器测量，但矢量化的生物标志物是其他测量的解释属性。临床医生使用所有这些数据模式来诊断和治疗眼部疾病，如糖尿病视网膜病变 (DR) 或糖尿病性黄斑水肿 (DME)。要在眼科医学领域中使用机器学习算法，就需要研究这些相关数据模式之间的关系和相互作用。现有数据集的局限性在于 :( $ i $) 他们将问题视为疾病预测而不评估生物标志物，并且 ($ ii $) 他们不考虑治疗期间所有四种数据模式之间的明确关系。在本文中，我们介绍了用于研究视觉眼语义 (OLIVES) 数据集的眼科标签，该标签解决了上述限制。这是第一个OCT和眼底数据集，其中包括临床标签，生物标志物标签和来自相关临床试验的时间序列患者治疗信息。该数据集由眼底图像组成，每个眼底图像均带有OCT扫描和生物标志物，以及临床标签和DR或DME的疾病诊断。总的来说，每只眼睛的平均数据在至少两年的时间内平均治疗了几周和注射。OLIVES数据集在机器学习研究的其他领域 (包括自我监督学习) 中具有优势，因为它提供了基于医学的替代增强方案。

Clinical diagnosis of eye diseases is performed using multiple data modalities, including scalar clinical labels, vectorized biomarkers, two-dimensional fundus images, and three-dimensional optical coherence tomography (OCT) scans. While clinical labels, fundus images, and OCT scans are instrument-measured data, vectorized biomarkers are interpretative attributes derived from other measurements. Clinicians utilize all these data modalities to diagnose and treat ocular diseases such as diabetic retinopathy (DR) or diabetic macular edema (DME). To deploy machine learning algorithms in the field of ophthalmology, it is essential to explore the relationships and interactions between these relevant data modalities.

The limitations of existing datasets are two-fold: (i) they frame the problem solely as disease prediction without evaluating biomarkers, and (ii) they fail to account for the explicit interrelationships across all four data modalities throughout the course of treatment.

In this paper, we present the Ophthalmic Labeling for Investigating Visual Ocular Semantics (OLIVES) dataset, which addresses the abovementioned limitations. This is the first joint OCT and fundus image dataset that encompasses clinical labels, biomarker labels, and time-series patient treatment information sourced from relevant clinical trials. The dataset comprises fundus images, each paired with corresponding OCT scans and biomarkers, along with clinical labels and disease diagnoses for either DR or DME. On average, each eye’s dataset includes average treatment weeks and injections over a period of at least two years.

The OLIVES dataset also presents advantages in other machine learning research domains, including self-supervised learning, as it provides medically grounded alternative augmentation strategies.

搜集汇总

背景与挑战

背景概述

OLIVES数据集是一个综合性的眼科医学数据集，包含临床标签、生物标志物、眼底图像和OCT扫描等多种数据模式，专注于糖尿病视网膜病变和糖尿病性黄斑水肿的研究。该数据集由佐治亚理工学院和Retina Consultants of America于2022年发布，特别强调了其在机器学习研究中的应用价值，尤其是在自我监督学习领域。