DEMIST: A Deep-Learning-Based Detection-Task-Specific Denoising Approach for Myocardial Perfusion SPECT

IEEE Transactions on Radiation and Plasma Medical Sciences, Год журнала: 2024, Номер 8(4), С. 439 - 450

Опубликована: Март 25, 2024

There is an important need for methods to process myocardial perfusion imaging (MPI) SPECT images acquired at lower radiation dose and/or acquisition time such that the processed improve observer performance on clinical task of detecting defects compared low-dose images.To address this need, we build upon concepts from modelobserver theory and our understanding human visual system propose a Detection task-specific deep-learning-based approach denoising MPI (DEMIST).The approach, while performing denoising, designed preserve features influence detection tasks.We objectively evaluated DEMIST using retrospective study with anonymized data in patients who underwent studies across two scanners (N = 338).The evaluation was performed levels 6.25%, 12.5% 25% anthropomorphic channelized Hotelling observer.Performance quantified area under receiver operating characteristics curve (AUC).Images denoised yielded significantly higher AUC corresponding commonly used task-agnostic DL-based method.Similar results were observed stratified analysis based patient sex defect type.Additionally, improved fidelity as root mean squared error structural similarity index metric.A mathematical revealed preserved assist tasks improving noise properties, resulting performance.The provide strong evidence further denoise low-count SPECT.

Язык: Английский

---

Nonlinear slow-timescale mechanisms in synaptic plasticity

Current Opinion in Neurobiology, Год журнала: 2023, Номер 82, С. 102778 - 102778

Опубликована: Авг. 30, 2023

Язык: Английский

---

Impact of Traditional and Embedded Image Denoising on CNN-Based Deep Learning

Applied Sciences, Год журнала: 2023, Номер 13(20), С. 11560 - 11560

Опубликована: Окт. 22, 2023

In digital image processing, filtering noise is an important step for reconstructing a high-quality further processing such as object segmentation, detection, and recognition. Various image-denoising approaches, including median, Gaussian, bilateral filters, are available in the literature. Since convolutional neural networks (CNN) able to directly learn complex patterns features from data, they have become popular choice tasks. As result of their ability adapt various denoising scenarios, CNNs powerful tools denoising. Some deep learning techniques CNN incorporate strategies into model layers. A primary limitation these methods necessity resize images consistent size. This resizing can loss vital details, which might compromise CNN’s effectiveness. Because this issue, we utilize traditional method preliminary reduction before applying CNN. To our knowledge, comparative performance study using embedded against baseline approach (without denoising) yet be performed. analyze impact on performance, paper, firstly, filter means use model. Secondly, embed layer validate denoising, performed extensive experiments both traffic sign recognition datasets. decide whether will adopted type used, also present exploiting peak-signal-to-noise-ratio (PSNRs) distribution images. Both accuracy PSNRs used evaluate effectiveness approaches. expected, results vary with filter, impact, dataset However, shows better accuracy, while lower computational time most cases. Overall, gives insights CNN-based analyses, autonomous driving, animal facial

Язык: Английский

---

Compact simultaneous label-free autofluorescence multi-harmonic microscopy for user-friendly photodamage-monitored imaging

Journal of Biomedical Optics, Год журнала: 2024, Номер 29(03)

Опубликована: Март 14, 2024

SignificanceLabel-free nonlinear optical microscopy has become a powerful tool for biomedical research. However, the possible photodamage risk hinders further clinical applications.AimTo reduce these adverse effects, we constructed new platform of simultaneous label-free autofluorescence multi-harmonic (SLAM) microscopy, featuring four-channel multimodal imaging, inline monitoring, and pulse repetition-rate tuning.ApproachUsing large-core birefringent photonic crystal fiber spectral broadening prism compressor pre-chirping, this system allows users to independently adjust width, repetition rate, energy, which is useful optimizing imaging conditions towards no/minimal photodamage.ResultsIt demonstrates multichannel at one excitation per image pixel thus paves way improving speed by faster scanner with low photodamage. Moreover, grants flexibility autonomously fine-tune average power, free from interference, ensuring discovery optimal high SNR minimal phototoxicity across various applications.ConclusionsThe combination stable laser source, tunable ultrashort pulse, monitoring features, compact design makes robust, powerful, user-friendly platform.

Язык: Английский

---

Deep learning in mesoscale brain image analysis: A review

Computers in Biology and Medicine, Год журнала: 2023, Номер 167, С. 107617 - 107617

Опубликована: Окт. 27, 2023

Язык: Английский

---

Enhancing Total Optical Throughput of Microscopy with Deep Learning for Intravital Observation

Small Methods, Год журнала: 2023, Номер 7(9)

Опубликована: Май 15, 2023

Abstract The significance of performing large‐depth dynamic microscopic imaging in vivo for life science research cannot be overstated. However, the optical throughput microscope limits available information per unit time, i.e., it is difficult to obtain both high spatial and temporal resolution at once. Here, a method proposed construct kind intravital microscopy with throughput, by making near‐infrared‐II (NIR‐II, 900–1880 nm) wide‐field fluorescence learn from two‐photon based on scale‐recurrent network. Using this upgraded NIR‐II microscope, vessels opaque brain rodent are reconstructed three‐dimensionally. Five‐fold axial thirteen‐fold lateral improvements achieved without sacrificing light utilization. Also, tiny cerebral vessel dilatations early acute respiratory failure mice observed, an speed 30 fps.

Язык: Английский

---

PSSS-EEG: A Probabilistic-masking Self-Supervised Swin-transformer model for EEG-based drowsiness recognition

Pattern Recognition, Год журнала: 2024, Номер 158, С. 111005 - 111005

Опубликована: Сен. 10, 2024

Язык: Английский

---

Deep-prior ODEs augment fluorescence imaging with chemical sensors

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

Опубликована: Окт. 24, 2024

To study biological signalling, great effort goes into designing sensors whose fluorescence follows the concentration of chemical messengers as closely possible. However, binding kinetics are often overlooked when interpreting cell signals from resulting measurements. We propose a method to reconstruct spatiotemporal underlying in consideration process. Our fits data under constraint corresponding reactions and with help deep-neural-network prior. test it on several GCaMP calcium sensors. The recovered concentrations concur common temporal waveform regardless sensor kinetics, whereas assuming equilibrium introduces artifacts. also show that our can reveal distinct events distribution single neurons. work augments current highlights importance incorporating physical constraints computational imaging.

Язык: Английский

---

Video-rate 3D imaging of living cells using Fourier view-channel-depth light field microscopy

Communications Biology, Год журнала: 2023, Номер 6(1)

Опубликована: Дек. 12, 2023

Abstract Interrogation of subcellular biological dynamics occurring in a living cell often requires noninvasive imaging the fragile with high spatiotemporal resolution across all three dimensions. It thereby poses big challenges to modern fluorescence microscopy implementations because limited photon budget live-cell task makes achievable performance conventional approaches compromise between their spatial resolution, volumetric speed, and phototoxicity. Here, we incorporate two-stage view-channel-depth (VCD) deep-learning reconstruction strategy Fourier light-field microscope based on diffractive optical element realize fast 3D super-resolution reconstructions intracellular from single diffraction-limited 2D light-filed measurements. This VCD-enabled approach (F-VCD), achieves video-rate (50 volumes per second) at ~180 nm × 180 400 strong noise-resistant capability, which light field images signal-to-noise ratio (SNR) down -1.62 dB could be well reconstructed. With this approach, successfully demonstrate 4D organelle dynamics, e.g., mitochondria fission fusion, ~5000 times observation.

Язык: Английский

---

Spatial redundancy transformer for self-supervised fluorescence image denoising

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

Опубликована: Июнь 5, 2023

Abstract Fluorescence imaging with high signal-to-noise ratios has become the foundation of accurate visualization and analysis biological phenomena. However, inevitable photon shot noise poses a formidable challenge on sensitivity. In this paper, we provide spatial redundancy denoising transformer (SRDTrans) to remove from fluorescence images in self-supervised manner. First, sampling strategy based is proposed extract adjacent orthogonal training pairs, which eliminates dependence speed. Secondly, break performance bottleneck convolutional neural networks (CNNs), designed lightweight spatiotemporal architecture capture long-range dependencies high-resolution features at low computational cost. SRDTrans can overcome inherent spectral bias CNNs restore high-frequency information without producing over-smoothed structures distorted traces. Finally, demonstrate state-of-the-art single-molecule localization microscopy two-photon volumetric calcium imaging. does not contain any assumptions about process sample, thus be easily extended wide range modalities applications.

Язык: Английский

---