Biomedical Optics Express

High-resolution microscopy techniques are used across research and industry to analyse biological systems, from biomolecules to subcellular organelles, multicellular models and tissues. As multimodal imaging workflows and quantitative analysis of bioimaging data become increasingly widespread, there is a growing need for materials and methods to calibrate imaging systems and evaluate the fidelity of generated image data. Here, we present three-dimensional microscopy phantoms fabricated using two-photon photolithography from transparent resins that exhibit both broadband visible autofluorescence and Raman scattering across the fingerprint and C-H stretching regions. Suitable for analysis using optical profilometry, the phantoms were dimensionally calibrated with SI traceability using a metrological confocal microscope. Immersible in air and common aqueous imaging media, the phantoms are compatible with a wide variety of optical microscopy techniques, including one and two-photon excited fluorescence and coherent Raman scattering microscopy. We employed a forked wedge design to validate image deconvolution results and a stacked lattice phantom to recover image distortion matrices under realistic biological imaging conditions. We demonstrate the impact of correcting chromatic offsets and axial scaling errors for a representative application: analysis of a cell seeded scaffold using confocal laser scanning fluorescence microscopy. These phantoms provide a versatile platform for calibration, quality control and validation of multimodal imaging pipelines and improved quantitative optical microscopy.

Stimulated emission depletion (STED) microscopy is a super-resolution fluorescence imaging technique that achieves high spatial and temporal resolution by exploiting stimulated emission to induce fluorescence depletion (FD) and is expected to have substantial utility for imaging applications using fluorescent proteins. However, the compatibility of fluorescent proteins with STED microscopy systems has been understood primarily through empirical observations, and there is no established methodology for the rational selection of fluorescent proteins for STED microscopy. In this study, we systematically evaluated the compatibility of commonly used fluorescent proteins with STED microscopy systems by measuring FD properties using transient absorption spectroscopy and fluorescence dip spectroscopy, both of which are classified as two-color spectroscopy (TCS). Fluorescent proteins identified as compatible with the STED microscopy system based on the TCS measurements were employed for three-dimensional STED imaging of cellular samples expressing each protein. In all samples, three-dimensional spatial resolution was improved relative to confocal laser microscopy, with particularly marked improvements in z-axis resolution. These findings demonstrate that measurements of FD properties via TCS provide a robust approach for evaluating the compatibility of fluorescent proteins with the STED microscopy system and for selecting suitable fluorescent proteins for STED imaging.

The spatial resolution of optical imaging systems is fundamentally restricted by the diffraction limit. However, in widefield live-cell microscopy, the achievable resolution is further constrained by the specimen motion, which indicates the existence of a fundamental spatio-temporal resolution trade-off between signal accumulation during the full frame integration and the resulting motion blur. To improve the fidelity with which moving objects can be imaged, a quantitative understanding of this spatio-temporal trade-off is necessary. Here, we present a systematic analysis of motion-induced resolution dynamics measured with spectral signal-to-noise ratio (SSNR). We developed a simulation framework which models the image formation of objects undergoing arbitrary motion, to evaluate the degradation of the spatial resolution under translational and rotational dynamics. Our results demonstrate that for translating objects, the spatial resolution is anisotropically reduced as a function of the orientation of the object relative to the motion vector, leading to the spectral signal-to-noise ratio degrading by up to 50% and the resolution by up to 40% for a 90{degrees} change in the motion direction. Furthermore, we show that for rotational motion, conventional radially averaged metrics such as the Fourier Ring Correlation are not able to quantify the effects of angular blur. On the other hand, the SSNR is able to accurately quantify this degradation. These findings underscore the necessity of an object-oriented imaging approach, in which acquisition parameters such as exposure time are tuned to specific biological spatio-temporal characteristics to optimize the trade-off between motion blur and spatial fidelity.

Tissue phantoms that mimic microvasculature and perfusion are essential for modelling vascular function, guiding interventions, and calibrating imaging systems, which require faithful replication of vascular geometry and flow. Conventional fabrication strategies, including wire-based molding, lithographic micromachining, and additive manufacturing, offer useful capabilities but remain constrained by predefined designs, rectangular channel cross-sections, limited scalability, and high production costs. Reliance on predefined digital vascular models restricts design flexibility and limits the ability to capture the natural variability and complexity of real vascular systems. Here, we present a lithography-free, fractal-generating approach based on a modified Lifted Hele-Shaw Cell (LHSC) technique, in which vascular networks emerge spontaneously via interfacial fluid instabilities. Unlike pre-designed methods, these structures are governed by fluid properties and flow conditions, enabling adaptive, physiologically relevant geometries with smooth Gaussian cross-sections and natural diameter tapering. We demonstrate four phantom designs: a planar vascular tree, an anatomically guided cerebral network, a retinal vascular model, and a conformable curved substrate phantom. Validation using Laser Speckle Contrast Imaging confirms structural fidelity and physiologically relevant flow consistent with Murrays law. This platform uniquely integrates realistic vascular architecture with emergent, fractal driven formation, highlighting its potential as a reproducible and biologically relevant alternative to conventional vascular phantom fabrication. Furthermore, the availability of such realistic in vitro vascular models can reduce reliance on animal experiments and contribute towards more ethical and sustainable preclinical research.

Visual prostheses face a critical miniaturisation challenge: converting photoreceptor signals to biologically appropriate retinal ganglion cell (RGC) stimulation patterns within the spatial constraints of intraocular implants. Existing systems rely on external microcontrollers for signal processing, limiting scalability for high-density pixel arrays. This paper presents an integrated per-pixel circuit architecture that directly converts photocurrent into frequency-modulated current pulses that match RGC activation thresholds. The design targets are established through NEURON computational modelling of red-green colour-opponent midget RGCs, identifying stimulation thresholds of +0.1nA to +3.5nA for depolarisation and -0.1nA for repolarisation. The proposed circuit combines a transimpedance amplifier, a voltage-controlled oscillator with a Schmitt trigger, and a current-controlled output stage to generate biphasic pulses within these thresholds. A complementary output provides lateral inhibition, reducing crosstalk between adjacent RGC stimulation sites. Photoreceptor integration is achieved using P3HT:PCBM organic photodiodes for cone-associated RGCs and phototransistors for rod-associated RGCs, validated through OghmaNano finite element simulations. The photodiode circuit produces output frequencies of 2.5Hz (dark) to 600Hz (100 W/m2), matching reported RGC response ranges. This architecture eliminates external processing requirements, enabling scalable high-density retinal prostheses design.

For decades, the photon counting histogram (PCH) was used as the sole method to quantify fluorophore numbers in a diffraction-limited focal volume. This technique combines spatial excitation profiles, and the distribution of photon counts to register the photon emission statistics of individual fluorophores. However, this approach has not yet been transferred to widefield fluorescent imaging due to the lack of fast and single photon sensitive camera sensors which can capture the photon emission statistics of a single fluorophore. Here, we explore avenues towards quantitative analysis of the active fluorophore number by leveraging recent advancements in single photon avalanche diode (SPAD) array technology. Binary exposures of a SPAD array can be synchronized with picosecond laser pulses to measure the PCH in a widefield setting. Then, by modeling the statistical relationship between the active fluorophore number and the PCH in a region of interest following a laser pulse, we can perform Bayesian inference of this number. The model is demonstrated experimentally by counting quantum dots and various numbers of fluorescent dye molecules bound to DNA origamis. We find that this method has several important applications in widefield microscopy, including enhanced localization microscopy and constrained fitting of multiple unresolvable fluorescent emitters.

Manufacturer-defined signal-strength indices are frequently employed as quality benchmarks for automated optical coherence tomography analysis, yet their empirical relationship with deep learning segmentation accuracy remains unclear. Because these metrics were originally developed for conventional image-processing pipelines, their ability to predict modern model-based segmentation accuracy has not been empirically validated. To address this gap, we evaluated the Heidelberg Spectralis Q-score against U-Net segmentation performance across 5,047 B-scans from 103 eyes for three anatomical boundaries of the posterior segment of the eye: the Ellipsoid Zone (EZ), Bruch's Membrane (BM), and Choroid Outer Boundary (COB). Alongside standard boundary agreement metrics (MAE, MSE, Dice Similarity Coefficient), we adapted the Earth Mover's Distance (EMD) from optimal transport theory as a boundary evaluation metric. Unlike column-wise averages, EMD quantifies boundary agreement as a 2-D geometric displacement, directly measuring residual spatial displacement between the model segmented boundary and the ground-truth boundary. Our results demonstrate that the Q-score - originally designed to gate image-processing-based automated analysis - is a poor predictor of deep learning boundary segmentation accuracy, with explained variance (R2) failing to exceed 1.4% across all three boundaries. We further observed a monotonically increasing error hierarchy with anatomical depth (EZ < BM < COB), consistent across metrics, which is unexplained by the signal strength. At the COB, correlations were paradoxically positive, explained by a B-scan-level mediation chain: higher Q-scores correspond to greater choroidal thickness (r=0.113, {rho}=0.158), which in turn predicts higher COB segmentation error (r=0.165, {rho}=0.191) - a localization difficulty that global signal strength cannot capture. Collectively, these findings challenge the implicit assumption that signal-strength-based quality thresholds are a reliable proxy for deep learning model performance, and motivate a shift toward task-specific acquisition quality criteria calibrated to model performance rather than signal interpretability.

BackgroundPrevious machine learning models to intraoperatively predict the molecular status of gliomas using stimulated Raman histology (SRH), such as DeepGlioma, have achieved high performance (91.5% accuracy) on curated datasets. However, when used intraoperatively, DeepGlioma (162M parameters) runs slowly on current SRH hardware and underperforms due to its lack of an image rejection mechanism and its validation on curated images. Here, we introduce SRH-Informed Glioma classificatioN with Attention Learning (SIGNAL) (27M parameters), a lighter model with a built-in attention-based rejection mechanism that outperforms DeepGlioma on uncurated clinical datasets. MethodsSIGNAL was developed using 1.56 million SRH fields-of-view from 967 adult diffuse glioma patients collected between December 2017 and July 2025. We used 412 patients from NYU for training and internal validation and a multi-institutional, international cohort of 555 patients for testing. SIGNAL uses a ResNet50 backbone pretrained using a hierarchical contrastive loss function followed by a multi-head multi-layer perceptron (MLP). Using a patch-based attention threshold of 0.6, a final MLP was trained to predict glioma subtypes: glioblastoma, oligodendroglioma, or astrocytoma. ResultsSIGNAL outperformed DeepGlioma, achieving greater overall accuracy (90.10% vs. 72.59%) while running faster (16.0 vs. 6.7 patches/s). SIGNAL also outperformed DeepGlioma on all three molecular classification tasks, including IDH mutation (accuracy: 93.51% vs. 79.22%), 1p19q codeletion (93.51% vs. 88.31%), and ATRX loss (89.61% vs. 83.98%). SIGNALs attention mechanism had a strong positive linear correlation with mean patch cellularity (r=0.96, p<0.001) and a strong negative correlation with patch blood coverage (r=-0.99,p<0.001). Finally, subtype and molecular accuracy between tumor core and margin samples were equivalent despite significantly lower patch retention in tumor margins (44.5% vs 60.2%, p<0.0001). ConclusionSIGNAL is a lightweight model for intraoperative molecular classification of gliomas using SRH imaging. Its attention-based image quality filter allows for excellent performance, quick processing, and highly interpretable outputs critical for reliable use in intraoperative workflows. Brief 1-2 Sentence DescriptionWe present SIGNAL, a lightweight machine learning model for intraoperative molecular classification of diffuse gliomas using stimulated Raman histology, whose core innovation is a learned attention mechanism that filters diagnostically uninformative tissue, such as blood and acellular regions, before classification, enabling robust real-world generalizability. Validated on 555 patients across four international centers, SIGNAL outperforms the previous state-of-the-art model DeepGlioma on glioma subtype classification (90.10% vs. 72.59% accuracy) while running 2.4 times faster on intraoperative hardware.

Light sheet fluorescence microscopy enables volumetric imaging with high imaging speed, optical sectioning capability, and reduced photobleaching and phototoxicity, and has become a workhorse in bioimaging. However, widely adopted Gaussian light sheets face an inherent trade-off between axial resolution and field-of-view due to diffraction. State-of-the-art nondiffracting light sheets--including Bessel beam, Airy beam, and lattice light sheet--alleviate this trade-off but suffer from optical aberrations that compromise performance with increasing imaging depth. While the integration of adaptive optics offers a promising solution, such integrated systems are typically complex, expensive, and slow due to the need for serial mapping and correction of spatially varying aberrations across the specimen. Here, we present polarization-engineered aberration-resilient light sheet (PEARLS), a new class of monochromatic nondiffracting light sheet with temporally invariant profile and robustness to optical aberrations. In comparison with existing light sheets, PEARLS showed significantly reduced photobleaching and enhanced aberration-resilience, permitting imaging of three-dimensional subcellular dynamics in optically complex environments. We applied PEARLS for noninvasive observations of biological dynamics in various living systems, revealing phenotypic diversity across spatial and temporal scales--from rapid membrane dynamics and organelle interactions in cultured cells to coordinated mitosis and cell migrations in developing embryos.

We find that multi-site temporal control of optogenetic photostimulation in peripheral nerves can enhance firing rates by overcoming the intrinsic limitation of opsin photophysics. The benefits of multi-site optogenetic stimulation were demonstrated with three approaches: (1) in silico modeling, (2) ex vivo in the sciatic nerve, and (3) in vivo in the vagus nerve. An in silico model of multi-site optogenetic stimulation was developed in two Hodgkin and Huxley type neuron models, that supported our hypothesis. The ex vivo sciatic nerve showed an increase in firing frequency that is physiologically relevant for functional control. The technique was then applied in vivo for optogenetic vagus nerve stimulation resulting in significant changes in heart rate compared with standard methods of single-site stimulation. Improving the control of optogenetically induced neural firing will have broad impacts for future developments in optical nerve interfaces and brain-machine interfaces.

Zebrafish (Danio rerio) are an important vertebrate model for vision and neuroscience research. In the larval stages, the aquatic species begins to elicit the optomotor response (OMR) to stabilize themselves in water -- a behaviour that may be exploited in the laboratory to measure visual acuity. However, up to now, the measurement of the OMR in juvenile and adult zebrafish has been limited due to their behavioural complexity. Here, we optimize a protocol to assay zebrafish aged between 4 and 9 weeks-post-fertilization, by displaying sinusoidal gratings parallel to the zebrafish eye to elicit a robust OMR. We assessed the visual spatial-frequency tuning function of an environmentally induced myopia model to confirm the sensitivity and robustness of the protocol. Additionally, we show the OMR is sensitive to the contrast and temporal resolution of the sinusoidal gratings. Furthermore, we found that the time between stimulus presentations impact the spatial-frequency tuning function likely as time is required for zebrafish to return to baseline swimming after eliciting the OMR. Finally, we found that the OMR after ten versus twenty seconds of stimulus onset appears comparable; indicating that robust OMR responses in zebrafish can be elicited through relatively short stimulus presentations. Through the experiments conducted, we present an optimized protocol specific to zebrafish. The protocol may be used to follow the progression or treatment efficacy of progressive neurological disorders including specific visual disorders and higher brain functions with visual endophenotypes. Ultimately, this protocol allows for high-throughput robust measures of visual and neural function in zebrafish.

We introduce a workflow to identify oligomeric structures that are recorded with single-molecule localization microscopy (SMLM) under cryogenic conditions. Typically, these oligomers are assumed to consist of protomers arranged as equilateral two-dimensional polygons and every protomer is labeled with a dye molecule for visualization. Unlike previous work, we consider scenarios in which the sample plane has an unknown orientation relative to the focal plane. Our contribution is a high-precision plane-fitting algorithm to determine the sample plane, combined with geometrical transformations and two circle-fitting algorithms to identify the oligomeric structures. Our simulations on synthetic data demonstrate that the proposed workflow achieves high accuracy in estimating both the unknown tilted plane and the oligomer size.

Three-dimensional photoacoustic tomography (3D-PAT) enables noninvasive structural and functional imaging with optical absorption contrast and ultrasonic detection depth. However, its spatial resolution is limited by acoustic diffraction, and incomplete detection geometry can substantially degrade image fidelity and quantitative accuracy. Here, we present a ULM-guided model-based reconstruction framework, termed 3D-PAULMprior that incorporates sub-diffraction vascular priors from concurrent ultrasound localization microscopy (ULM) into 3D photoacoustic reconstruction. The method uses weighted regional Laplacian regularization to integrate high-resolution vascular information into the inverse problem, thereby enhancing vascular sharpness, suppressing limited-view artifacts, and improving blood oxygen saturation estimation. We validated 3D-PAULMprior using numerical simulations, tissue-mimicking phantoms, and in vivo mouse brain imaging. Compared with conventional reconstruction, 3D- PAULMprior improved spatial resolution by over 50%, increased contrast-to-noise ratio by 261.2%, and enhanced structural similarity index by 24.6%. In vivo, 3D-PAULMprior recovered vascular structures that were poorly resolved or missing in conventional reconstructions and produced more spatially confined sO2 maps. These results establish 3D-PAULMprior as a robust multimodal reconstruction strategy for high-resolution structural and functional photoacoustic imaging.

Species-level Candida identification can inform antifungal management, but reliable identification platforms remain inaccessible in many clinical microbiology laboratories, whereas phase contrast microscopy -- a common feature of routine laboratory microscopes -- is widely available. We asked whether this ubiquitous optical tool, combined with a consumer smartphone and deep transfer learning, could provide a feasible low-cost approach for preliminary Candida species discrimination. Fifteen clinical isolates of four species (C. albicans, C. glabrata, C. tropicalis, C. krusei) were collected from a single clinical microbiology laboratory and imaged using a consumer-grade smartphone coupled to a standard phase contrast microscope. Suspensions in human serum were imaged immediately after preparation (T0) and after 2-hour incubation at 37{degrees}C (T2). Pretrained vision backbone architectures were evaluated as fixed feature extractors under strict Leave-One-Strain-Out cross-validation. The best-performing model -- EfficientNet-B0 embeddings with a Linear Support Vector Machine applied to T2 images -- achieved an apparent internally cross-validated strain-level balanced accuracy of 0.833 and an overall strain accuracy of 86.7% (13/15 strains correctly classified). C. albicans, C. glabrata, and C. tropicalis were each identified with 100% recall. Both misclassified strains belonged to C. krusei -- the species with the smallest panel representation (n=3 strains) -- with misclassification attributable to limited strain diversity and suboptimal image quality. These findings demonstrate promising feasibility for preliminary image-based Candida species discrimination from smartphone-acquired phase contrast microscopy images, and support further evaluation in larger, externally validated strain collections.

Wide-field imaging (WFI) is a mesoscopic approach for monitoring cortex-wide activity with high temporal resolution and a broad field of view. Owing to its simple optical configuration and compatibility with chronic preparations, WFI has become an important tool in systems neuroscience and disease-model research. In this chapter, we describe practical protocols for chronic transcranial WFI in mice using two complementary optical signals: genetically encoded calcium indicators (GCaMP) and endogenous flavoprotein autofluorescence. Calcium imaging provides a robust readout of neuronal population activity, whereas flavoprotein imaging reflects mitochondrial redox dynamics and cellular metabolic demand. We detail procedures for animal preparation, skull clearing, headplate implantation, macroscope assembly, synchronized sensory stimulation, triggered image acquisition, and MATLAB-based data analysis. The analysis workflow includes {Delta}F/F normalization, reference-based signal correction, and artifact reduction, followed by trial averaging, atlas registration, and region-of-interest analysis. Because imaging is performed through the intact skull, the protocol enables repeated longitudinal measurements in the same animal over extended periods. This approach is reproducible, cost-effective, and adaptable to studies of cortical physiology and neurological disorders.

Background: Quantitative lipid assessment is central to identifying rupture-prone coronary plaques and represents a therapeutic target for lipid-lowering therapy. Near-infrared spectroscopy (NIRS)-derived lipid core burden index (LCBI) is well validated and widely used for detecting lipid-rich lesions. Optical frequency domain imaging (OFDI) is increasingly adopted for guiding percutaneous coronary intervention (PCI) due to its high-resolution structural imaging capabilities. Depolarization-sensitive OFDI (depOFDI) provides intrinsic lipid contrast and may enable combined structural and compositional plaque characterization within a single OFDI-based platform. Objective: To define an OFDI-derived lipid metric and evaluate its agreement with NIRS-derived LCBI. Methods: Thirty-three patients underwent both polarization-sensitive OFDI and NIRS-intravascular ultrasound imaging during PCI. After exclusion of 4 datasets, 29 co-registered pullbacks were analyzed. A signal-to-noise-corrected depolarization metric was used to identify lipid-rich regions and generate depOFDI chemograms. maxLCBI4mm value and location, as well as total LCBI, were computed and compared with NIRS. Results: depOFDI demonstrated strong agreement with NIRS, showing high correlation for maxLCBI4mm (r^2 = 0.862) and total LCBI (r^2 = 0.867), along with strong spatial concordance for the location of the maxLCBI4mm (r^2 = 0.900). Bland-Altman analysis of LCBI4mm showed minimal bias (10.7) with 95% limits of agreement of [81.4 to 102.8]. Conclusions: depOFDI enables accurate quantification of lipid burden alongside the high-resolution structural information inherently provided by OFDI. Because depolarization metrics can be derived from polarization-diverse detection available in many commercial OFDI systems, this approach provides a practical pathway toward comprehensive plaque characterization within existing PCI workflows, without the need for additional imaging modalities.

To properly respond to their environment, cells adjust the activity of key regulatory proteins and rates of gene expression. Methods to detect and quantify these forms of regulatory dynamics in living cells are of central importance for understanding cellular signaling events in both physiological and pathological conditions. Current technologies in this field make use of fluorescent probes to track cell signaling dynamics. Although these technologies have been used for decades, challenges remain. In particular, the segmentation, tracking, and interpretation of single cell dynamic data are time-consuming, prone to subjective errors, and often lacking in standardization across experiments. Here, we present SPIFEE, a data pipeline that uses experiment-dependent parameters to smooth noise and quantify key features of fluorescence data from time-lapse imaging studies. Processing data in this manner enhances and accelerates quantification of live-cell gene and protein expression, simplifies data analysis, and facilitates hypothesis generation. Author SummaryCells adjust protein activity and gene expression levels over time to respond to changes in their environment, a process referred to as cell signaling dynamics. Quantifying cell signaling dynamics in living cells often uses fluorescent probes, such as green fluorescent protein (GFP) and its spectral variants, to track changes in gene expression or protein activity over time. Challenges inherent in analyzing fluorescence data from single cells stem from biological and experimental noise, time-consuming quantification, and subjective errors. To address these challenges, we developed a computational tool called Signal Processing and Integrated Feature Extraction (SPIFEE). The pipeline improves the quality of fluorescence data analysis by reducing noise and extracting signal features in a way that is both intuitive and objective. The pipeline provides more accurate, rapid, and unbiased quantification of time-lapse microscopy data.

Contactless assessment of cardiopulmonary function remains an unmet need, with current approaches relying either on subjective clinical examination or on resource-intensive imaging. We evaluated a novel multipoint airborne ultrasound surface motion camera (SMC) designed to map thoracic vibration patterns without contact and to extract clinically relevant information through data-driven analysis. In a prospective observational study, clinically characterised participants underwent short-duration acquisitions during natural breathing and externally induced oscillations. The resulting signals were transformed into spatially and frequency-resolved maps and analysed using machine learning models to discriminate healthy individuals from patients with respiratory or cardiac disease. The approach proved feasible in a clinical setting and achieved excellent discrimination between healthy individuals and respiratory patients (area under the receiver operating characteristic curve (AUC) 0.90 {+/-} 0.07), including in patients with subtle abnormalities not detected by pulmonary function testing. Discrimination between healthy individuals and cardiac patients ranged from acceptable to excellent (AUC 0.76-0.90 depending on subgroup), with the highest performance observed in aortic stenosis. Model interpretability analyses revealed spatial and spectral patterns consistent with the known physiological organisation of lung mechanics and cardiac auscultation areas, supporting a structure-function relationship between recorded signals and underlying processes. These findings indicate that thoracic vibration transmission encodes spatially and spectrally organised information that can be captured without contact and exploited through explainable data-driven modelling. While the results require confirmation in larger populations, this approach may represent an operator-independent, low-burden extension of bedside assessment, with potential applications in early detection, triage, and monitoring of cardiopulmonary disease.

Pressure injuries represent a significant healthcare challenge requiring early detection to prevent severe complications. While thermal imaging shows promise for detecting early pressure-related temperature changes, its robustness across varying imaging conditions and diverse patient populations remains unclear. This study systematically evaluated how imaging protocol variations (lighting, distance, positioning, camera type) and participant skin tone influence classification model performance for thermal cooling detection. Using a controlled cooling protocol to simulate early pressure injury temperature changes, we collected 1,680 images from 35 diverse participants across 12 imaging protocol variations. We compared two approaches: three deep learning models (MobileNetV2, InceptionNetV3, ResNet50) and a threshold-based approach using an optimal fixed threshold temperature differential. Deep learning models outperformed the threshold-based approach, achieving 98.6-99.6% accuracy compared to 95.6%, with superior performance across all imaging protocols and skin tone groups. Threshold-based approach showed camera-dependent misclassification patterns across skin tones. On the high-resolution FLIR E8XT, the MST 7-10 group had 8 of 11 misclassifications. This pattern shifted on the low-resolution FLIR ONE Pro, where the intermediate skin tone group (MST 6) had 22 of 44 total misclassifications.In contrast, deep learning models maintained consistent performance across all skin tone groups and imaging protocols. Visualization analysis of the deep learning models suggested that these models focused on thermal gradients at cooling region boundaries, suggesting that spatial temperature gradients, not single-value thresholds, are critical for accurate detection. These findings suggest the potential of deep learning-based approaches to maintain robust, equitable performance across diverse skin tones and imaging conditions.

Microbubble contrast-enhanced ultrasound (CEUS) relies on discriminating nonlinear bubble signals from linear tissue backscattering. While Singular Value Decomposition (SVD) filtering improves this discrimination, existing techniques often fail to retain the slowly-moving microbubble signals from static clutter. Here, we present a novel multi-stage singular value decomposition (MS-SVD) framework for ultrafast CEUS imaging. Our method employs plane-wave transmissions at multiple angles and acoustic pressure levels (implemented via duty-cycle modulation) and alternating transmit polarity. The beamformed data are then processed by three sequential SVD filters: (1) spatial-angular SVD to extract coherent signals across all transmit angles, (2) spatial-pressure SVD to separate linear fundamental and nonlinear harmonic components, and (3) spatiotemporal SVD to isolate moving microbubble echoes from tissue clutter. In in vitro flow phantoms and in vivo rat brain through a cranial window, MS-SVD dramatically improves microbubble detection compared to conventional SVD filtering, MS-SVD yields much stronger vascular contrast and suppresses tissue clutter to a greater extent. The resulting power-Doppler and super-resolution maps are notably cleaner and more complete: MS-SVD detects substantially more microbubble events in ULM, revealing finer vessel details and more accurate flow speeds. By capturing the full acoustic signature of microbubbles (both fundamental and harmonic), MS-SVD achieves higher contrast-to-noise and sensitivity in CEUS. These gains make it a powerful front-end for super-resolution ultrasound localization microscopy and other high-sensitivity microvascular imaging applications.