CF-OCT-segmentation
收藏资源简介:
该数据集是首个针对慢性植入大鼠模型中人工耳蜗诱导的耳蜗内纤维化的光学相干断层扫描(OCT)图像数据集,由俄勒冈健康与科学大学等机构创建。数据集包含173张来自5个样本的标注图像,每张图像以高分辨率标注了纤维化、鼓阶/自由空间和CI/轨道三个语义类别。数据通过OCT B-scans获取并重切片为En Face图像,由经验丰富的分割者使用Hasty.ai工具手动标注。该数据集旨在应用计算机视觉方法客观量化耳蜗纤维化负担,从而助力研究残余听力损失的机制,为混合人工耳蜗患者的疗效提升提供关键支撑。
This dataset is the first optical coherence tomography (OCT) image dataset focused on intra-cochlear fibrosis induced by cochlear implants in chronic cochlear implant rat models, developed by institutions including Oregon Health & Science University. The dataset comprises 173 annotated images from 5 samples, with each image annotated at high resolution for three semantic categories: fibrosis, scala tympani/free space, and CI/electrode track. The data were acquired as OCT B-scans and then reformatted into En Face images, and manually annotated by experienced segmenters using the Hasty.ai tool. This dataset aims to employ computer vision methods to objectively quantify the burden of cochlear fibrosis, thereby advancing research into the mechanisms of residual hearing loss and providing critical support for improving the therapeutic outcomes of patients with hybrid cochlear implants.
Cochlear Fibrosis OCT数据集详情
数据集概述
- 数据集名称:Cochlear Fibrosis OCT(CF-OCT)数据集
- 数据来源:动物模型(cochlear fibrosis,耳蜗纤维化)
- 数据模态:光学相干断层扫描(OCT)图像
- 下载地址:可通过Open Science Framework(OSF)获取(项目链接:https://osf.io/cghn7,注册DOI:https://doi.org/10.17605/OSF.IO/WB5FS)
- 标注类别:共4类,包括CI/Track(红色)、Fibrosis(绿色)、ST/Free Space(蓝色)及背景类
- 标注卷数:包含5个已标注的OCT体积(OCTV1L、OCTV7L、OCTV9L、OCTV10L、OCTV11L)
数据集用途
- 主要目标:用于耳蜗纤维化的语义分割与量化分析
- 应用场景:
- 探究耳蜗纤维化与残余听力损失的关系
- 辅助人工耳蜗(CI)植入的研发
- 支持EAS(电声刺激)患者的治疗研究
基准模型
该数据集提供了多种语义分割模型的基准测试结果,包括:
- 2D-OCT-UNET(提出的模型):非常深的2D U-Net架构,包含7个编码器-解码器块和跳跃连接,使用Group Normalization(ng=2)和Dropout(0.1),输入分辨率1024×1024像素
- VGG16-UNET
- UEfficientNet
- SegFormer
- MST-DeepLabv3
- SAM(Segment Anything Model)的微调参考见:https://github.com/mazurowski-lab/finetune-SAM
量化功能
代码内置纤维化量化功能,可自动计算原始体积(OCTV1L、OCTV7L、OCTV9L、OCTV10L、OCTV11L)中的纤维化量。
实现环境
- 语义分割模型(除SAM外):Python 3.9.12、Tensorflow 2.9.1、Keras 2.9.0
- SAM模型:Python 3.9.12、PyTorch 2.0.1 (cu117)
- 实验平台:Ubuntu 18.04.3 LTS,Intel Core i9-9900K CPU,Nvidia GeForce RTX 2080 Ti GPU,62GB RAM
引用信息
Dietlmeier, J., Greenberg, B., He, W., Wilson, T., Xing, R., Hill, J., Fettig, A., Otto, M., Rounsavill, T., Reiss, L.A.J., Yi, J. OConnor, N.E., Burwood, G.W.S. (2025). Towards Investigating Residual Hearing Loss: Quantification of Fibrosis in a Novel Cochlear OCT Dataset. IEEE Transactions on Biomedical Engineering。




