Authors: Lin Li, Jingyi Liu, Shuo Wang, Xunkun Wang, and Tian-Zhu Xiang.
- In this work, we collect the first large-scale Microscopic Image dataset of Trichomonas Vaginalis, called TVMI3K, which consists of 3,158 images covering Trichomonas of various appearances in diverse backgrounds, with high-quality annotations including object-level mask labels, object boundaries, and challenging attributes.
- Besides, we propose a simple yet effective baseline, named TVNet, to automatically segment Trichomonas from microscopic images.
Figure 1: Examples of our proposed TVMI3K. We provide different annotations, including object-level masks, object edges and challenging attributes. We use red boxes to mark Trichomonas and green boxes to mark leukocyte on images for better visualization. Leukocyte shows high similarity with Trichomonas.
TVMI3K datasets: Download
Figure 2: Overview of our proposed TVNet, which consists of high-resolution fusion (HRF) module and foreground-background attention (FBA) module. Refer to our paper for details.
Table 1: Quantitative comparison on our TVMI3K dataset. The best scores are highlighted in bold.
Figure 3: Visual comparison of different methods. Our method provides more accurate predictions than other competitors in various challenging scenarios.
-
Results of our TVNet: Baidu Pan (1let) | Google Drive
-
Performance of competing methods: Baidu Pan (0qv5) | Google Drive
-
The models and source code are coming soon.
-
For evaluation, we adopt PySODMetrics and ② Dice-IoU. They can also be found in
Evaluation
folder.
Please cite our paper if you find the work useful:
@inproceedings{li2022TVS,
title={Trichomonas Vaginalis Segmentation in Microscope Images},
author={Li, Lin and Liu, Jingyi and Wang, Shuo and Wang, Xunkun and Xiang, Tian-Zhu},
booktitle={International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI},
year={2022}
}
If you have any questions about our paper, feel free to contact me via e-mail (ll198196@163.com).