A novel generic dictionary-based denoising method for improving noisy and densely packed nuclei segmentation in 3D time-lapse fluorescence microscopy images

Abstract

Time-lapse fluorescence microscopy is an essential technique for quantifying various characteristics of cellular processes, i.e. cell survival, migration, and differentiation. To perform high-throughput quantification of cellular processes, nuclei segmentation and tracking should be performed in an automated manner. Nevertheless, nuclei segmentation and tracking are challenging tasks due to embedded noise, intensity inhomogeneity, shape variation as well as a weak boundary of nuclei. Several nuclei segmentation approaches have been reported in the literature. However, most of them require fine-tuning of many parameters. Furthermore, they can only be applied to particular models or acquisition systems. As a result, there is a crucial need for a generic method to segment noisy and densely packed nuclei in microscopy images. In this paper, we present a sparse representation method for denoising of microscopy images. The proposed method works properly in the context of most challenging cases. In addition, it provides a denoised image and simultaneously the potential locations of nuclei. To evaluate the performance of the proposed method, we tested our method on two datasets from cell tracking challenge. Across all datasets, the proposed method achieved satisfactory results with 96.96% recall for C.elegans dataset. Besides, in Drosophila dataset, our method achieved very high recall (99.3%).

Detection and Spatiotemporal analysis of in-vitro 3D migratory Triple-Negative Breast cancer cells

“…was assessed using manual segmentation, by drawing the approximate borders of the nuclei, and compared to the performance of four alternative methods of Nasser et al 20 , a CellProfiler pipeline 8 , and two pretrained models (cyto and nuclei) of the Cellpose deep-learning segmentation algorithm. 30 The segmentation performance was calculated using the accuracy, recall, precision, F1 score, Jaccard index (supplementary section 2.3).…”

“…Novel machine learning based methods exhibit improved performance, however their applicability may be limited to specific datasets, and their performance may be decreased in datasets with high cell shape and volume heterogeneity, as well as high cell density. 20 Recent studies have mainly focused on the preprocessing of 3D image stacks, to improve the segmentation results of simpler segmentation algorithms, and are applicable to large datasets. 20 Concluding, the advances made in both experimental and image processing methods provide us with the opportunity to further investigate the spatiotemporal organization and progression of cells in greater detail using 3D cell cultures.…”

“…20 Recent studies have mainly focused on the preprocessing of 3D image stacks, to improve the segmentation results of simpler segmentation algorithms, and are applicable to large datasets. 20 Concluding, the advances made in both experimental and image processing methods provide us with the opportunity to further investigate the spatiotemporal organization and progression of cells in greater detail using 3D cell cultures.…”

“…The performance of the proposed pipeline was evaluated against manual segmentation, and four alternative pipelines including established 8 , and novel machine learning algorithms. 20,30 The segmented nuclei were subsequently used for the investigation of their spatiotemporal progression using point-pattern analysis, and density analysis methods. The novel combination of these methods enabled us to detect the position of the cells in the 3D space with higher accuracy, as well as to examine the organization and progression of cancer growth.…”

“…Segmentation performance of the proposed pipeline (FluoDeSeg), the method developed by Nasser et al20 , the CellProfiler pipeline 8 , and the two Cellpose models (cyto and nuclei)30 as compared to manual segmentation. Results are reported as Mean ± Standard Deviation.…”