分割
人工智能
计算机科学
图像分割
尺度空间分割
计算机视觉
模式识别(心理学)
跟踪(教育)
基于分割的对象分类
特征提取
心理学
教育学
作者
Keliang Zhao,Jovial Niyogisubizo,Li Xiao,Yi Pan,Dong‐Qing Wei,Didi Rosiyadi,Yanjie Wei
标识
DOI:10.1109/bibm58861.2023.10385935
摘要
The precise segmentation and tracking of cells in microscopy image sequences play a pivotal role in biomedical research, facilitating the study of tissue, organ, and organism development. However, manual segmentation and tracking of cells is time-consuming and often require professional experiences. Besides, segmenting cells in the images with a low signal-to-noise ratio remains difficult. While deep learning (DL) has become a common method for cell segmentation, few DL- based methods address concurrent cell segmentation and tracking. In this paper, we propose a novel DL approach featuring graph-based tracking for cell segmentation and tracking in microscopy images. We combine Deeplabv3+ for semantic segmentation and ResNet50 for enhanced feature extraction, enabling comprehensive cell detection and instance segmentation. Post-processing, involving non-maxima suppression and outlier detection, refines predictions and produces final segmentation. The tracking method is based on the relative position of graph nodes to track segmented cells, encompassing cell division and apoptosis. We conduct our experiments on the induced pluripotent stem (iPS) cell datasets, and the results show that the segmentation and tracking performance of our method yields superior performance compared to the benchmark models. More specifically, our approach achieved DET values of 0.955 and 0.913, TRA values of 0.951 and 0.906, and SEG values of 0.690 and 0.665 on two iPS dataset videos, respectively. Additionally, the performance assessment encompasses four real microscopy datasets from the Cell Tracking Challenge (CTC). Our method greatly reduces the cost of manual labeling and labor-intensive costs.
科研通智能强力驱动
Strongly Powered by AbleSci AI