Model weights and processed data - Cluster-based human-in-the-loop strategy for improving CTC detection and classification

The DINO model weights and the processed data accompany the following GitHub repository: CTC-HiL Detecting and differentiating circulating tumor cells (CTCs) in blood samples from cancer patients is challenging. Among the challenges is the reliance on manual annotation and evaluation. This study introduces a human-in-the-loop (HiL) approach to enhance ML-based CTC detection by combining self-supervised deep learning with a traditional classifier, using iterative sampling and human experts for efficient labeling based on classification performance in latent space clusters. Overview of what you can download from here: The DINO model ctc_dino.pt was trained on unlabeled single cell fluorescence images with three-channels (DAPI, CK, CD45) from 60 patients with metastatic breast cancer. The provided features and dataframes can be used to reproduce the simulation findings and the cluster plot. For more information, visit CTC-HiL.本数据集配套的DINO模型权重与预处理数据可从以下GitHub仓库CTC-HiL获取。 癌症患者血液样本中的循环肿瘤细胞（circulating tumor cells, CTCs）检测与分型极具挑战性，其中一大难题便是依赖人工标注与评估。 本研究提出一种人机协同（human-in-the-loop, HiL）方案，通过将自监督深度学习与传统分类器相结合，并基于潜在空间聚类的分类表现，借助迭代采样与人类专家完成高效标注，以优化基于机器学习（machine learning, ML）的CTCs检测任务。 您可在此下载的内容概述如下： 本研究所用的DINO模型权重文件ctc_dino.pt，基于60例转移性乳腺癌患者的三通道（DAPI、CK、CD45）未标注单细胞荧光图像训练得到。 本数据集提供的特征与数据帧可用于复现仿真结果与聚类可视化图。如需了解更多信息，请访问CTC-HiL。