Segmentation metric misinterpretations in bioimage analysis

Nature Methods, Год журнала: 2023, Номер 21(2), С. 213 - 216

Опубликована: Июль 27, 2023

Abstract Quantitative evaluation of image segmentation algorithms is crucial in the field bioimage analysis. The most common assessment scores, however, are often misinterpreted and multiple definitions coexist with same name. Here we present ambiguities metrics for show how these misinterpretations can alter leaderboards influential competitions. We also propose guidelines currently existing problems could be tackled.

Язык: Английский

---

Detection and Counting of Maize Leaves Based on Two-Stage Deep Learning with UAV-Based RGB Image

Remote Sensing, Год журнала: 2022, Номер 14(21), С. 5388 - 5388

Опубликована: Окт. 27, 2022

Leaf age is an important trait in the process of maize (Zea mays L.) growth. It significant to estimate seed activity and yield by counting leaves. Detection leaves field are very difficult due complexity scenes cross-covering adjacent seedling A method was proposed this study for detecting based on deep learning with RGB images collected unmanned aerial vehicles (UAVs). The Mask R-CNN used separate complete seedlings from complex background reduce impact weeds leaf counting. We a new loss function SmoothLR improve segmentation performance model. Then, YOLOv5 detect count individual after segmentation. 1005 were randomly divided into training, validation, test set ratio 7:2:1. results showed that Resnet50 better than LI Loss. average precision bounding box (Bbox) mask (Mask) 96.9% 95.2%, respectively. inference time single image detection 0.05 s 0.07 s, performed compared Faster SSD. YOLOv5x largest parameter had best performance. fully unfolded newly appeared 92.0% 68.8%, recall rates 84.4% 50.0%, (AP) 89.6% 54.0%, accuracy 75.3% 72.9%, experimental possibility current research exploring field-grown crops UAV images.

Язык: Английский

---

Opportunities and challenges for deep learning in cell dynamics research

Trends in Cell Biology, Год журнала: 2023, Номер unknown

Опубликована: Ноя. 1, 2023

The growth of artificial intelligence (AI) has led to an increase in the adoption computer vision and deep learning (DL) techniques for evaluation microscopy images movies. This not only addressed hurdles quantitative analysis dynamic cell biological processes but also started support advances drug development, precision medicine, genome–phenome mapping. We survey existing AI-based tools, as well open-source datasets, with a specific focus on computational tasks segmentation, classification, tracking cellular subcellular structures dynamics. summarise long-standing challenges video from perspective review emerging research frontiers innovative applications DL-guided automation dynamics research.

Язык: Английский

---

AnyStar: Domain randomized universal star-convex 3D instance segmentation

2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Год журнала: 2024, Номер unknown, С. 7578 - 7588

Опубликована: Янв. 3, 2024

Star-convex shapes arise across bio-microscopy and radiology in the form of nuclei, nodules, metastases, other units. Existing instance segmentation networks for such structures train on densely labeled instances each dataset, which requires substantial often impractical manual annotation effort. Further, significant reengineering or finetuning is needed when presented with new datasets imaging modalities due to changes contrast, shape, orientation, resolution, density. We present AnyStar, a domain-randomized generative model that simulates synthetic training data blob-like objects randomized appearance, environments, physics general-purpose star-convex networks. As result, trained using our do not require annotated images from un-seen datasets. A single network synthesized accurately 3D segments C. elegans P. dumerilii nuclei fluorescence microscopy, mouse cortical μCT, zebrafish brain EM, placental cotyledons human fetal MRI, all without any retraining, finetuning, transfer learning, domain adaptation. Code available at https://github.com/neel-dey/AnyStar.

Язык: Английский

---

MMV_Im2Im: an open-source microscopy machine vision toolbox for image-to-image transformation

GigaScience, Год журнала: 2024, Номер 13

Опубликована: Янв. 1, 2024

Over the past decade, deep learning (DL) research in computer vision has been growing rapidly, with many advances DL-based image analysis methods for biomedical problems. In this work, we introduce MMV_Im2Im, a new open-source Python package image-to-image transformation bioimaging applications. MMV_Im2Im is designed generic framework that can be used wide range of tasks, including semantic segmentation, instance restoration, generation, and so on. Our implementation takes advantage state-of-the-art machine engineering techniques, allowing researchers to focus on their without worrying about details. We demonstrate effectiveness more than 10 different problems, showcasing its general potentials applicabilities. For computational researchers, provides starting point developing or algorithms, where they either reuse code fork extend facilitate development methods. Experimental benefit from work by gaining comprehensive view concept through diversified examples use cases. hope give community inspirations how integrated into assay process, enabling studies cannot done only traditional experimental assays. To help get started, have provided source code, documentation, tutorials at [https://github.com/MMV-Lab/mmv_im2im] under MIT license.

Язык: Английский

---

Morphodynamics of human early brain organoid development

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Авг. 22, 2023

Abstract Brain organoids enable the mechanistic study of human brain development, and provide opportunities to explore self-organization in unconstrained developmental systems. Here, we establish long-term, live light sheet microscopy on unguided generated from fluorescently labeled induced pluripotent stem cells, which enables tracking tissue morphology, cell behaviors, subcellular features over weeks organoid development. We a novel dual-channel, multi-mosaic multi-protein labeling strategy combined with computational demultiplexing approach simultaneous quantification distinct during track Actin, Tubulin, plasma membrane, nucleus, nuclear envelope dynamics, quantify morphometric alignment changes state transitions including neuroepithelial induction, maturation, lumenization, regionalization. Based imaging single-cell transcriptome modalities, find that lumenal expansion morphotype composition within developing neuroepithelium are associated modulation gene expression programs involving extracellular matrix (ECM) pathway regulators mechanosensing. show an extrinsically provided enhances lumen as well telencephalon formation, grown absence extrinsic have altered morphologies increased neural crest caudalized identity. Matrixinduced regional guidance morphogenesis linked WNT Hippo (YAP1) signaling pathways, spatially restricted induction Wnt Ligand Secretion Mediator (WLS) marks earliest emergence nontelencephalic regions. Altogether, our work provides new inroad into studying morphodynamics, supports view matrix-linked mechanosensing dynamics play central role

Язык: Английский

---

Nuclear instance segmentation and tracking for preimplantation mouse embryos

Development, Год журнала: 2024, Номер 151(21)

Опубликована: Окт. 7, 2024

ABSTRACT For investigations into fate specification and morphogenesis in time-lapse images of preimplantation embryos, automated 3D instance segmentation tracking nuclei are invaluable. Low signal-to-noise ratio, high voxel anisotropy, nuclear density, variable shapes can limit the performance methods, while is complicated by cell divisions, low frame rates, sample movements. Supervised machine learning approaches radically improve accuracy enable easier tracking, but they often require large amounts annotated data. Here, we first report a previously unreported mouse line expressing near-infrared reporter H2B-miRFP720. We then generate dataset (termed BlastoSPIM) H2B-miRFP720-expressing embryos with ground truth for instances. Using BlastoSPIM, benchmark seven convolutional neural networks identify Stardist-3D as most accurate method. With our BlastoSPIM-trained models, construct complete pipeline lineage from eight-cell stage to end development (>100 nuclei). Finally, demonstrate usefulness BlastoSPIM pre-train data related problems, both different imaging modality model systems.

Язык: Английский

---

Depth-enhanced high-throughput microscopy by compact PSF engineering

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Июнь 7, 2024

Abstract High-throughput microscopy is vital for screening applications, where three-dimensional (3D) cellular models play a key role. However, due to defocus susceptibility, current 3D high-throughput microscopes require axial scanning, which lowers throughput and increases photobleaching photodamage. Point spread function (PSF) engineering an optical method that enables various imaging capabilities, yet it has not been implemented in the cumbersome extension typically requires. Here we demonstrate compact PSF objective lens, allows us enhance depth of field and, combined with deep learning, recover information using single snapshots. Beyond applications shown here, this work showcases usefulness obtaining training data learning-based algorithms, applicable variety modalities.

Язык: Английский

---

SegElegans: Instance segmentation using dual convolutional recurrent neural network decoder in Caenorhabditis elegans microscopic images

Computers in Biology and Medicine, Год журнала: 2025, Номер 190, С. 110012 - 110012

Опубликована: Март 22, 2025

Язык: Английский

---

Advanced optical imaging for the rational design of nanomedicines

Advanced Drug Delivery Reviews, Год журнала: 2023, Номер 204, С. 115138 - 115138

Опубликована: Ноя. 18, 2023

Despite the enormous potential of nanomedicines to shape future medicine, their clinical translation remains suboptimal. Translational challenges are present in every step development pipeline, from a lack understanding patient heterogeneity insufficient insights on nanoparticle properties and impact material-cell interactions. Here, we discuss how adoption advanced optical microscopy techniques, such as super-resolution microscopies, correlative high-content modalities, could aid rational design nanocarriers, by characterizing cell, nanomaterial, interaction with unprecedented spatial and/or temporal detail. In this nanomedicine arena, will implementation these versatility specificity, can yield high volumes multi-parametric data; machine learning rapid advances microscopy: image acquisition data interpretation.

Язык: Английский

---