ABSTRACT

Objectives

To assess a recently described CNN (convolutional neural network) DARC (Detection of Apoptosing Retinal Cells) algorithm in predicting new Subretinal Fluid (SRF) formation in Age-related-Macular-Degeneration (AMD).

Methods

Anonymized DARC, baseline and serial OCT images (n = 427) from 29 AMD eyes of Phase 2 clinical trial (ISRCTN10751859) were assessed with CNN algorithms, enabling the location of each DARC spot on corresponding OCT slices (n = 20,629). Assessment of DARC in a rabbit model of angiogenesis was performed in parallel.

Results

A CNN DARC count >5 at baseline was significantly (p = 0.0156) related to development of new SRF throughout 36 months. Prediction rate of eyes using unique DARC spots overlying new SRF had positive predictive values, sensitivities and specificities >70%, with DARC count significantly (p < 0.005) related to the magnitude of SRF accumulation at all time points. DARC identified earliest stages of angiogenesis in-vivo.

Conclusions

DARC was able to predict new wet-AMD activity. Using only an OCT-CNN definition of new SRF, we demonstrate that DARC can identify early endothelial neovascular activity, as confirmed by rabbit studies. Although larger validation studies are required, this shows the potential of DARC as a biomarker of wet AMD, and potentially saving vision-loss.

中文翻译：

---

使用DARC（检测凋亡视网膜细胞）AI（人工智能）技术预测湿性年龄相关性黄斑变性（AMD）

 抽象的

 目标

评估最近描述的 CNN（卷积神经网络）DARC（视网膜细胞凋亡检测）算法在预测年龄相关性黄斑变性 (AMD) 中新的视网膜下液 (SRF) 形成方面的作用。

 方法

使用 CNN 算法评估来自 2 期临床试验 (ISRCTN10751859) 29 只 AMD 眼睛的匿名 DARC、基线和连续 OCT 图像 (n = 427)，从而能够在相应的 OCT 切片上定位每个 DARC 点 (n = 20,629)。同时对兔血管生成模型中的 DARC 进行评估。

 结果

基线时 CNN DARC 计数 >5 与 36 个月内新 SRF 的开发显着相关 (p = 0.0156)。使用覆盖新 SRF 的独特 DARC 点的眼睛预测率具有阳性预测值、敏感性和特异性 >70%，DARC 计数与所有时间点的 SRF 积累量显着相关 (p < 0.005)。 DARC 确定了体内血管生成的最早阶段。

 结论

DARC 能够预测新的湿性 AMD 活动。仅使用新 SRF 的 OCT-CNN 定义，我们证明 DARC 可以识别早期内皮新生血管活动，正如兔子研究所证实的那样。尽管还需要更大规模的验证研究，但这显示了 DARC 作为湿性 AMD 生物标志物的潜力，并有可能挽救视力丧失。