Zhou Mengxi, Zhang Yue, Kirkendall Eli, Karimi Monsefi Amin, Wolfe Matthew, Chitkara Kiran A, Choi Stacey S, Doble Nathan, Parthasarathy Srinivasan, Ramnath Rajiv
The Ohio State University, Department of Computer Science and Engineering, 2015 Neil Ave., Columbus, OH 43210, USA.
The Ohio State University, College of Optometry, 338 W 10th Ave., Columbus, OH 43210, USA.
Biomed Opt Express. 2025 Jul 17;16(8):3237-3254. doi: 10.1364/BOE.563128. eCollection 2025 Aug 1.
Adaptive optics-optical coherence tomography (AO-OCT) enables cellular-level in vivo visualization of cone photoreceptors in the human retina. Cone biomarkers, such as density, inner segment (IS), and outer segment (OS) lengths, are potentially important for the early detection of many outer retinal conditions. However, their dense spatial packing necessitates automated analytical methods, and most existing approaches focus primarily on cone detection without addressing their detailed structural characteristics. To address this limitation, a unified neural network, termed ISOSNet, is introduced for simultaneous cone detection and IS/OS length measurement. Labeled AO-OCT B-scan datasets, encompassing healthy individuals across multiple retinal locations, were collected for model training and evaluation. Experimental results demonstrate an F1 score of 0.886 for cone detection and relative error rates of 6% and 11% for IS and OS length measurement, respectively. Validation on images from diseased retinas-despite the model being trained only on healthy retina data-highlights the generalizability of the proposed framework.
自适应光学光学相干断层扫描（AO-OCT）能够在细胞水平对人类视网膜中的视锥光感受器进行体内可视化。视锥生物标志物，如密度、内节（IS）和外节（OS）长度，对于许多视网膜外层疾病的早期检测可能具有重要意义。然而，它们密集的空间排列需要自动化分析方法，并且大多数现有方法主要侧重于视锥检测，而没有考虑其详细的结构特征。为了解决这一局限性，本文引入了一种统一的神经网络ISOSNet，用于同时进行视锥检测和IS/OS长度测量。收集了涵盖多个视网膜位置的健康个体的标记AO-OCT B扫描数据集，用于模型训练和评估。实验结果表明，视锥检测的F1分数为0.886，IS和OS长度测量的相对误差率分别为6%和11%。尽管该模型仅在健康视网膜数据上进行训练，但对患病视网膜图像的验证突出了所提出框架的通用性。
