【医学+深度论文:F08】2018 Performance of Deep Learning Architectures and Transfer Learning for Detecting
08
Performance of Deep Learning Architectures and Transfer Learning for Detecting Glaucomatous Optic Neuropathy in Fundus Photographs
2018
Method : ONH 二分类问题
Dataset :14822(stereoscopic)
Architecture : model - VGG16/Inceptionv3/ResNet50;transfer - pretraining on ImageNet database
Results : Best model AUC=0.91,moderate-to-severe 0.97,mild AUC 0.89
评估了几种深度学习架构和迁移学习
不同的CNN结构最适合于不同的问题,而且事先知道什么结构适合于给定的任务是很困难的,因此经验测试通常是确定这个问题的最佳方法。
Methods
Data
- Nidek Stereo Camera Model 3-DX (Nidek Inc, Palo Alto, California)
simultaneous stereoscopic ONH photographs
- VF
Humphrey Field Analyzer II (Carl Zeiss Meditec, Dubin, CA)
standard 24-2 testing pattern / the Swedish Interactive Tresholding Algorithm
Preprocessing and Data Augmentation
在训练数据的子集上建立了一个单独的深度学习模型来估计disc中心的位置
同一患者的图片分在同一个分区中
数据增强
ONH图像降采样到一个共同的大小为224 × 224像素。
评价健康组、轻度功能丧失组、中度功能丧失组和重度功能丧失组AUC的均值和标准差。此外,还计算了该模型在不同特异性水平下识别任何轻度、中度和重度多边形的敏感度,以表征模型性能。
其他人实验标记为GON要求: cup-to-disc 0.70,局限性切口,边缘宽度≤0.1 disc直径,可见神经fber层缺损,或disc出血。
Results
- 使用 transfer 更好
- transfer Resnet50最好
- 检测重度青光眼正确率更高
Discussion
不同种族群体
比较不同的深度学习架构(VGG16 vs. Inception vs. ResNet50)和策略(原生学习vs.迁移学习)的结果,可能有助于为将来识别GON的模型开发提供指导。
未来的工作应该研究ONH区域大小、分辨率以及立体信息对模型性能的影响。合并更多的周边和立体信息可能有助于改善青光眼损伤的识别。
- 【医学+深度论文:F11】2018 A deep learning model for the detection of both advanced and early glaucoma using
- 【医学+深度论文:F09】2018 Deep Convolution Neural Network for Accurate Diagnosis of Glaucoma Using Digital
- Deep Learning for Design and Retr of Nano-photonic Structures 论文学习
- 论文阅读:End-to-End Learning of Deformable Mixture of Parts and Deep Convolutional Neural Networks for H
- 论文笔记 A Large Contextual Dataset for Classification,Detection and Counting of Cars with Deep Learning
- 【医学+深度论文:F14】2018 Joint Optic Disc and Cup Segmentation Based on Multi Label Deep Network
- 【深度学习】论文导读:ELU激活函数的提出(FAST AND ACCURATE DEEP NETWORK LEARNING BY EXPONENTIAL LINEAR UNITS (ELUS))
- 【医学+深度论文:F15】2018 Disc-Aware Ensemble Network for Glaucoma Screening From Fundus Image
- 论文阅读理解 - Deep Learning of Binary Hash Codes for Fast Image Retrieval
- 论文笔记:Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- [深度学习论文笔记][Weight Initialization] Exact solutions to the nonlinear dynamics of learning in deep lin
- 实时SLAM的未来及与深度学习的比较The Future of Real-Time SLAM and “Deep Learning vs SLAM”
- 1607.CVPR-Joint Learning of Single-image and Cross-image Representations for Person ReID 论文笔记
- 中文译文:Minerva-一种可扩展的高效的深度学习训练平台(Minerva - A Scalable and Highly Efficient Training Platform for Deep Learning)
- The Future of Real-Time SLAM and "Deep Learning vs SLAM" 深度学习与slam
- 论文学习-深度学习目标检测2014至201901综述-Deep Learning for Generic Object Detection A Survey
- End-to-End Learning of Geometry and Context for Deep Stereo Regression
- [深度学习论文笔记][Image Classification] Deep Residual Learning for Image Recognition
- 深度学习国外课程资料(Deep Learning for Self-Driving Cars)+(Deep Reinforcement Learning and Control )
- 论文笔记《A Survey of Model Compression and Acceleration for Deep Neural Networks》