How many specimens make a sufficient training set for automated three-dimensional feature extraction?

Royal Society Open Science, Год журнала: 2024, Номер 11(6)

Опубликована: Июнь 1, 2024

Deep learning has emerged as a robust tool for automating feature extraction from three-dimensional images, offering an efficient alternative to labour-intensive and potentially biased manual image segmentation methods. However, there been limited exploration into the optimal training set sizes, including assessing whether artficial expansion by data augmentation can achieve consistent results in less time how these benefits are across different types of traits. In this study, we manually segmented 50 planktonic foraminifera specimens genus Menardella determine minimum number images required produce accurate volumetric shape internal external structures. The reveal unsurprisingly that deep models improve with larger eight being 95% accuracy. Furthermore, enhance network accuracy up 8.0%. Notably, predicting both measurements structure poses greater challenge compared structure, owing low contrast differences between materials increased geometric complexity. These provide novel insight sizes precise diverse traits highlight potential enhancing multivariate images.

Язык: Английский

---

A scoping study of the whole-cell imaging literature: a foundational corpus, potential for data-mining and research synthesis, and a call for standardization of an emerging field

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Фев. 4, 2025

ABSTRACT The level of cellular organization bridging the mesoscale and whole-cell scale is coming into focus as a new frontier in cell biology. Great progress has been made unraveling complex physical functional interconnectivity organelles, but how entire organelle network spatially arranges within cytoplasm only beginning to be explored. Drawing on cross-disciplinary research synthesis methods, we systematically curated volumetric imaging literature through 3 rounds screening involving independent reviewers, resulting 89 top hits 38 “borderline” studies. We describe trajectory current state field (2004-2024). A broad characterization, or “scoping review”, bibliometrics, study design, reporting practices shows accelerating technological development output. find high variability design practices, including modality, model organism, contexts, organelles imaged, analyses. Due laborious, low-throughput nature most trends toward small sample sizes (<30 cells) types. common quantitative analyses across studies, ratios inter-organelle contact Our dataset now enables future aggregate comparative potentially reveal larger patterns generate more generalized hypotheses. This work establishes growing data, motivates call for standardized reporting, data sharing practices. More broadly, showcase potential rigorous secondary methods strengthen biology’s review reproducibility toolkit, create avenues discovery, promote open that support data-reuse integration.

Язык: Английский

---

Subcellular Feature-Based Classification of α and β Cells Using Soft X-ray Tomography

Cells, Год журнала: 2024, Номер 13(10), С. 869 - 869

Опубликована: Май 18, 2024

The dysfunction of α and β cells in pancreatic islets can lead to diabetes. Many questions remain on the subcellular organization islet during progression disease. Existing three-dimensional cellular mapping approaches face challenges such as time-intensive sample sectioning subjective identification. To address these challenges, we have developed a feature-based classification approach, which allows us identify quantify their structural characteristics using soft X-ray tomography (SXT). We observed significant differences whole-cell morphological organelle statistics between two cell types. Additionally, characterize subtle biophysical individual insulin glucagon vesicles by analyzing vesicle size molecular density distributions, were not previously possible other methods. These sub-vesicular parameters enable predict types systematically supervised machine learning. also visualize distinct subtypes Uniform Manifold Approximation Projection (UMAP) embeddings, provides with an innovative approach explore heterogeneity cells. This methodology presents for tracking biologically meaningful that be applied any system.

Язык: Английский

---

Automated segmentation of soft X-ray tomography: native cellular structure with sub-micron resolution at high throughput for whole-cell quantitative imaging in yeast

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 1, 2024

Abstract Soft X-ray tomography (SXT) is an invaluable tool for quantitatively analyzing cellular structures at sub-optical isotropic resolution. However, it has traditionally depended on manual segmentation, limiting its scalability large datasets. Here, we leverage a deep learning-based auto-segmentation pipeline to segment and label in hundreds of cells across three Saccharomyces cerevisiae strains. This task-based employs iterative refinement improve segmentation accuracy key structures, including the cell body, nucleus, vacuole, lipid droplets, enabling high-throughput precise phenotypic analysis. Using this approach, compared 3D whole-cell morphometric characteristics wild-type, VPH1-GFP, vac14 strains, uncovering detailed strain-specific organelle size shape variations. We show utility SXT data curvature analysis entire organelles detection fine morphological features using surface meshes. Our approach facilitates comparative analyses with high spatial precision statistical throughput, subtle single population level. workflow significantly enhances our ability characterize anatomy supports scalable studies mesoscale, applications investigating architecture, biology, genetic research diverse biological contexts. Significance Statement offers many powerful multi-organelle imaging, but, like other resolution volumetric imaging modalities, typically limited by low throughput due laborious segmentation. Auto-segmentation soft overcomes limitation, multiple whole populations. The combination statistically useful represents avenue more thorough characterizations toto opens new mesoscale questions modeling morphology, interactions, responses perturbations.

Язык: Английский

---

Soft X-ray tomography analysis of mitochondria dynamics in Saccharomyces cerevisiae

Biology Direct, Год журнала: 2024, Номер 19(1)

Опубликована: Ноя. 29, 2024

Язык: Английский

---