Cell tracking in low-frame-rate video based on displacement prediction

基于位移预测的低帧率视频中的细胞跟踪

• 批准号: 10648570
• 负责人: Federico Iuricich
• 金额: $ 20.41万
• 依托单位: CLEMSON UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-01 至 2025-08-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10648570
• 关键词: A549; Address; Affect; Air; Algorithms; Area; Biochemistry; Bioinformatics; Biological; Biology; Cell Nucleus; Cells; Cellular biology; Cessation of life; Classification; Clover; Consumption; Data; Data Set; Developmental Biology; Drops; Environment; Equipment; Evaluation; Event; Formulation; Future; Genetic; HL60; Health; Hour; Image; Learning; Link; Locomotion; MCF10A cells; MDA MB 231; Machine Learning; Mammalian Cell; Manuals; Measures; Medicine; Membrane; Methods; Microscope; Morphologic artifacts; Morphology; Movement; Performance; Photobleaching; Phototoxicity; Preparation; Regenerative Medicine; Research; Research Personnel; Resources; Science; Speed; Techniques; Temperature; Testing; Time; Training; Visualization; Work; cell analyzer; cell motility; cell type; cellular imaging; cost; cytate; cytotoxicity; deep learning; deep neural network; design; drug discovery; improved; innovation; interest; light transmission; live cell imaging; neural network; novel strategies; recurrent neural network; tool

Project Summary Tracking living cells in video sequences is a fundamental task in many fields of science, including biochemistry, bioinformatics, cell biology, and genetics. Manually linking cells is extremely time-consuming and not feasible in large-scale analysis. Automatic approaches can compute cell links by measuring how close two instances of a cell are, or how similar they look. These techniques work well with video acquired at a relatively high frame rate, but, unfortunately, acquiring images at high frame rates affects cells negatively. Too frequent imaging not only causes phototoxicity, leading to experimental artifacts, but also photobleaching, leading to the inability to measure quantities of interest over time. In addition, during image acquisition, the environment temperature and air quality are typically less controlled, which could also contribute to cytotoxicity. Moreover, when performing high- throughput live-cell imaging, the lower the acquisition rate, the more cells/plates can be imaged, and, consequently, the more experimental treatments can be applied and studied. If reducing the acquisition rate is beneficial for all these reasons, it severely affects the accuracy of cell tracking algorithms. To this end, we propose a new class of cell tracking approaches based on cell movement predictions. Instead of comparing cells based on their similarity, we propose to predict where every cell will move in the next frame. This will allow for searching the occurrence of such cells, even if the next frame was acquired after an extended period. The new approach will be investigated using a newly generated dataset for low frame rate cell tracking (Aim 1). Cell displacement will be predicted by using a new Recurrent Neural Network designed for the task (Aim 2). Cell tracking algorithms will be defined re-evaluating existing approaches under low-frame rate constraints when using cell displacement information (Aim 3). While current approaches require image acquisition to occur at least every 5-15 minutes, we will investigate the feasibility of cell tracking on images acquired at intervals of up to 2 hours. If successful, our research will allow to accurately track cells in low frame rate video sequences without the need for specialized tools or equipment.

项目摘要 在视频序列中跟踪活细胞是许多科学领域的基本任务， 包括生物化学，生物信息学，细胞生物学和遗传学。手动连接单元是 在大规模分析中非常耗时，并且不可行。自动方法可以 通过测量细胞的两个实例有多接近或它们的外观来计算细胞链接。 这些技术与以相对较高的帧速率获取的视频很好地运行，但是 不幸的是，以高帧速率获取图像会对细胞产生负面影响。太常见了 成像不仅会引起光毒性，从而导致实验性伪影，还会引起实验性毒性 光漂白，导致无法随着时间的推移衡量兴趣数量。此外， 在图像获取过程中，环境温度和空气质量通常较低 受控，这也可能有助于细胞毒性。而且，在执行高级时 吞吐量活细胞成像，采集速率越低，可以成像的单元/板越多， 因此，可以应用和研究更多的实验治疗方法。 如果由于所有这些原因降低采集率是有益的，则严重影响准确性 细胞跟踪算法。为此，我们提出了基于细胞跟踪方法的新类别 关于细胞运动预测。我们提出的不是基于细胞的相似性比较细胞的相似性 预测每个单元将在下一个帧中移动的位置。这将允许搜索 即使在延长后获得下一个框架，也会发生这种细胞。这 将使用新生成的低框架率单元的数据集研究新方法 跟踪（目标1）。通过使用新的复发神经网络预测细胞位移 为任务设计（AIM 2）。细胞跟踪算法将被定义为重新评估现有 使用细胞位移信息时，在低框架速率约束下的方法（AIM 3）。 虽然当前的方法需要至少每5-15分钟一次进行图像获取，但我们 将研究细胞跟踪对图像的可行性，最多2小时。如果 成功，我们的研究将允许在低帧速率视频序列中准确跟踪细胞 无需专门的工具或设备。