Co-OCTDL
收藏资源简介:
OCTDL数据集包含2064张高分辨率光学相干断层扫描(OCT)图像,代表七种不同的疾病,并具有显著的类别不平衡。该数据集由医学专家进行标注,以确保诊断的可靠性。为了解决数据集不平衡的问题,研究人员构建了一个完美平衡版本的Co-OCTDL数据集,其中每个扫描都作为3x1布局的复合图像呈现。新的数据集Co-OCTDL通过将多个同类别图像融合成单个图像,提高了训练样本的信息密度,并增加了模型区分细微疾病模式的能力。
The OCTDL dataset contains 2064 high-resolution optical coherence tomography (OCT) images corresponding to seven distinct diseases, and exhibits significant class imbalance. This dataset was annotated by medical experts to guarantee the reliability of the diagnostic labels. To address the class imbalance issue of the original dataset, researchers constructed a perfectly balanced variant named Co-OCTDL, where each original OCT scan is presented as a composite image with a 3x1 layout. The newly developed Co-OCTDL dataset enhances the information density of training samples by fusing multiple images of the same class into a single composite image, and improves the model's ability to distinguish subtle disease patterns.

- 1通过摩洛哥费斯西迪穆罕默德本阿卜杜拉大学(USMBA)多学科工程学院 · 2025年



