Tracking Data I/II of the publication "A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction"

DATA belonging to the paper "A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction" Katharina Löffler, Tim Scherr, Ralf Mikut doi: https://doi.org/10.1101/2021.03.16.435631 ----------------------------- To investigate the influence of different segmentation errors on the tracking performance we simulate errorneous segmentation data: - under-segmentation (referred to as "merge" in the folders), over-segmentation("split"), False Negatives ("remove"), combination of the aforementioned errors ("mixed") - percentages: 1,2,5,10,20 of errorneous masks per dataset - runs: 5 randomly initialized runs per combination - datasets: Fluo-N2DH-SIM+ and Fluo-N3DH-SIM+ each with two image sequences ---> in total 4 (error types) * 5 (percentage) * 5 (runs) * 2 (data sets) * 2 (image sequences) = 400 datasets The datasets can be recreated by running our code https://git.scc.kit.edu/KIT-Sch-GE/2021-cell-tracking ---------------------------- RESULTS We evuated the four tracking algorithms KIT-Sch-GE(1), KTH-SE, MU-Lux-CZ and our proposed algorithm on the aforementioned datasets and compare their performance using the CTC metrics DET, SEG and TRA. This repository contains all metrics as xls files and all tracking results as image sequences. xls files ----------- compare_all_trackers_on_synt_bm.csv Comparing the tracking algorithms MU-Lux-CZ, KTH-SE, KIT-Sch-GE(1) and the proposed tracking algorithm on synthetically degraded segmentation data Fluo-N2DH-SIM+ and Fluo-N3DH-SIM+ (Cell Tracking Challenge datasets). Reported scores are DET, SEG and TRA from the Cell Tracking Challenge (Fig8 and Fig9 and Supplementary Figures 3 and 4 are created from this data) compare_postprocessing_on_synth_bm.csv Comparing the different post-processing strategies of the proposed tracking algorithm algorithm on synthetically degraded segmentation data Fluo-N2DH-SIM+ and Fluo-N3DH-SIM+ (Cell Tracking Challenge datasets). Reported scores are DET, SEG and TRA from the Cell Tracking Challenge (Fig7 and Fig8 and Supplementary Figures 1 and 2 are created from this data) PLEASE NOTE: the folder compare_postprocessing_synth_bm is provided in the repository 10.5281/zenodo.5227610 due to size restrictions. folders (decompressed approximately 90GB of data!) ----------- tracking_data compare_all_synth_bm Contains all tracking results for each tracking algorithm on the synthetically degraded datasets () compare_all_synth_bm_no_error Contains the tracking results for each tracking algorithm provided with the perfect ground truth segmentation data compare_postprocessing_synth_bm [will be stored in 10.5281/zenodo.5227610 due to size restrictions] Contains all tracking resuls for each postprocessing configuration of the proposed cell tracking algorithm the leaf folders are names run_xPOSTPROCESSING where x is the run number and POSTPROCESSING the postprocessing key Postprocessing keys: ("no untangle" or "no masks" is indicated by an overline in the paper) all ("untangle + masks" in the paper) nd ("no untangle + masks") nd_ns-l ("no untangle + no masks") ns-l ("untangle + no masks")本数据集来自论文《基于图的细胞追踪算法：少量手动可调参数与自动化分割错误校正》（A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction），作者为Katharina Löffler、Tim Scherr、Ralf Mikut，DOI：https://doi.org/10.1101/2021.03.16.435631 为探究不同分割错误对细胞追踪性能的影响，我们模拟了带错误的分割数据集： - 欠分割（文件夹中记为"merge"）、过分割（记为"split"）、假阴性（记为"remove"），以及上述错误的组合（记为"mixed"）； - 错误掩码占比：每个数据集的1%、2%、5%、10%、20%； - 每组组合设置5次随机初始化的运行实验； - 数据集包含Fluo-N2DH-SIM+和Fluo-N3DH-SIM+，二者各包含2个图像序列。 总数据集数量为：4（错误类型）×5（错误占比）×5（运行次数）×2（数据集类型）×2（图像序列）=400个数据集。 本数据集可通过运行作者提供的代码复现，代码地址：https://git.scc.kit.edu/KIT-Sch-GE/2021-cell-tracking ## 实验结果 我们在上述数据集上评估了四种细胞追踪算法：KIT-Sch-GE(1)、KTH-SE、MU-Lux-CZ以及本文提出的算法，并采用细胞追踪挑战（Cell Tracking Challenge, CTC）的DET、SEG和TRA指标对比其性能。本仓库包含所有以XLS格式存储的指标数据，以及所有以图像序列形式存储的追踪结果。 ### XLS文件说明 1. `compare_all_trackers_on_synt_bm.csv`：对比了MU-Lux-CZ、KTH-SE、KIT-Sch-GE(1)以及本文提出的追踪算法在经过合成退化处理的分割数据集Fluo-N2DH-SIM+和Fluo-N3DH-SIM+（均为细胞追踪挑战数据集）上的表现。报告指标为细胞追踪挑战的DET、SEG和TRA得分（本文图8、图9及补充材料图3、图4均由该数据生成）。 2. `compare_postprocessing_on_synth_bm.csv`：对比了本文提出的追踪算法的不同后处理策略在经过合成退化处理的分割数据集Fluo-N2DH-SIM+和Fluo-N3DH-SIM+（均为细胞追踪挑战数据集）上的表现。报告指标为细胞追踪挑战的DET、SEG和TRA得分（本文图7、图8及补充材料图1、图2均由该数据生成）。 ### 注意事项 请注意：由于文件大小限制，`compare_postprocessing_synth_bm`文件夹已上传至10.5281/zenodo.5227610。 ### 文件夹说明（解压后约占用90GB存储空间！） - `tracking_data/compare_all_synth_bm`：包含所有追踪算法在合成退化数据集上的全部追踪结果。 - `tracking_data/compare_all_synth_bm_no_error`：包含所有追踪算法在完美真值分割数据集上的追踪结果。 - `tracking_data/compare_postprocessing_synth_bm`【因文件大小限制，将存储于10.5281/zenodo.5227610】：包含本文提出的细胞追踪算法的每种后处理配置的全部追踪结果。子文件夹命名格式为`run_xPOSTPROCESSING`，其中`x`为运行编号，`POSTPROCESSING`为后处理关键字。 后处理关键字说明： - `all`：对应论文中的"untangle + masks" - `nd`：对应"no untangle + masks" - `nd_ns-l`：对应"no untangle + no masks" - `ns-l`：对应"untangle + no masks" 注：论文中使用上划线表示"no untangle"或"no masks"。