通过使用深度神经网络的光学连贯性断层扫描对视网膜的可解释检测
Murat Seçkin Ayhan1, Jonas Neubauer2, Mehmet Murat Uzel2,3
1Institute for Ophthalmic Research, University of Tübingen, Elfriede Aulhorn Str. 7, 72076, Tübingen, Germany.
深度神经网络 (DNN) 在OCT扫描中准确地检测和测量视网膜 (ERM),即使是小扫描. 突出地图突出了ERM,有助于选和决策支持.
科学领域:
- 眼科医生 眼科 眼科
- 医疗成像医学成像
- 人工智能的人工智能
背景情况:
- 视网膜 (ERM) 是视力损伤的常见原因之一.
- 准确检测和分类ERM大小对于患者管理至关重要.
- 目前用于ERM检测的方法可能耗时且主观.
研究的目的:
- 开发和验证深度神经网络 (DNN) 用于在OCT扫描中自动检测和大小分类ERM.
- 评估DNN在中部和偏中部黄斑区域的ERM识别方面的表现.
- 使用突出地图可视化ERM本地化并提高检测准确性.
主要方法:
- 收集了11 061张OCT图像的数据集,并根据ERM的存在和大小进行分级 (小:100-1000μm,大:>1000μm).
- 一组DNN在75%的数据上接受了培训,10%用于验证,15%用于测试.
- 使用Guided-Backprob生成突出地图,并进行t-SNE分析以减少维度.
主要成果:
- 在检测无ERM (AUC:0.99),小ERM (AUC:0.92) 和大ERM (AUC:0.99) 方面,DNN实现了高性能,整体三向准确率为89%.
- 小ERM是最难检测的,但突出地图有效地突出了它们.
- t-SNE分析显示了以大小为基础的分类,并在组过渡时揭示了分类不确定性.
结论:
- DNN可靠地检测并按尺寸分类ERM,在FOVEAL和PARACENTRALOCT扫描中.
- 拟议的DNN模型,增强了突出性地图,可以准确地识别即使是微妙的,小ERM.
- 这项技术有可能在眼科查计划和临床决策支持系统中未来应用.
更多相关视频
09:37Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
04:48Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
相关概念视频
Self-Evaluation: Self-Enhancement and Self-Verification
General State of Stress
Specifically, consider a tetrahedral element where one face, labeled XYZ, is perpendicular to the line OA, and the remaining faces align with the coordinate axes with point O as the origin. At any point, such as point O, the stress tensor can be used to determine the stress...
Ethnic Identity within a Larger Culture
Introduction to Stress and Lifestyle
Psychological Responses to Stress
Stress and Mental Health
Individuals with depression often experience challenges in both their personal and professional...
