Methods and systems for label-free imaging

Abstract

Provided herein is a microfluidic system that automates detection and response to transient organelle events. A chamber transports organelles in a fluid while an image capture device acquires time-sequenced images. A controller, executing machine-learning instructions, receives a first image sequence and determines no transient event is present, thereby leaving device operation unchanged. The controller subsequently receives a second image sequence and, based on differences between sequences, determines that a transient event has occurred. In response, the controller automatically triggers a solution exchange within the chamber.

Description

Atty Dkt.: 114198-0890METHODS AND SYSTEMS FOR LABEL-FREE IMAGINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of and priority to U.S. Provisional Application No. 63 / 734,023, filed December 13, 2024, which is incorporated herein by reference in its entirety.BACKGROUND

[0002] Microscopic imaging systems are widely used in biological research to study the structural and functional organization of cells. Advances in optical instrumentation and computational analysis have enabled high-resolution visualization of subcellular components. However, it remains challenging to acquire and process dynamic cellular information efficiently without perturbing the biological system under observation.SUMMARY OF THE DISCLOSURE

[0003] Organelle dynamics play a crucial role in cellular health and neurodegenerative diseases. However, current imaging technologies often lack the resolution and non-invasive capabilities needed to capture subtle alterations in living cells. Applicant presents a novel approach combining quantitative phase imaging (QPI) with deep learning (DL) to study organelle dynamics in primary skin fibroblasts from patients with Charcot-Marie-Tooth (CMT) disease and age-matched controls.

[0004] Applicant’s system integrates a high-performance QPI microscope with quasi - simultaneous fluorescence imaging, achieving ~200ms framerate videomicroscopy. This multimodal approach allows label-free imaging of subcellular structures via QPI, validated by targeted fluorescence imaging of specific organelles.

[0005] In one aspect, one can automate analysis of QPI data.

[0006] While this disclosure focuses on segmentation and tracking of mitochondria these methods and systems can be extended to other organelles to enable high-throughput, unbiased quantification of complex organelle behaviors from label-free imaging data.

[0007] Applying this technology to fibroblasts from CMT2A patients with mutations in the mitochondrial fusion protein MFN2 and age-matched controls, Applicant identified distinct patterns of mitochondrial dynamics. CMT2A patient cells exhibited increased mitochondrial fragmentation and clustering. Importantly, these phenotypes were detectable in QPI data and-1-4919-3750-8870.1Atty Dkt.: 114198-0890 confirmed by correlative fluorescence imaging, demonstrating the power of Applicant’s QPI approach for studying disease-related cellular changes without fluorescent labels.

[0008] This platform combines cutting-edge optical engineering and machine learning to achieve an unprecedented view of organelle dynamics in the context of disease. The ability to detect cellular changes without fluorescent labels opens new avenues for studying dysfunction in living patient-derived cells with minimal perturbation. Application of this technology will have broad impact across cell biology and neuroscience, enabling new discoveries about how organelle dysfunction contributes to pathological states. Moreover, Applicant’s system can serve as a foundation for future high-throughput screening assays to identify therapeutics that modulate organelle dynamics, potentially accelerating drug discovery efforts for CMT and other neurodegenerative disorders.

[0009] In one example, the disclosure relates to systems for detecting transient subcellular events and generating corresponding control actions in response to variations in organelle dynamics. Conventional microscopy systems can acquire image sequences at high spatial and temporal resolutions but often lack automation for dynamic event detection. As a result, transient organelle events such as mitochondrial fission, lysosomal fusion, or vesicle lysis can occur between set observation intervals, leading to missed or incomplete temporal records. Existing approaches that depend on frame review or thresholding of static image features can be inefficient and computer resource-intensive. Such limitations restrict quantitative, high- throughput analysis of transient events under varying physiological or disease conditions and hinder reproducibility across independent experimental sessions.

[0010] The techniques described herein can employ a combined microfluidic and deeplearning imaging architecture that couples a quantitative phase or fluorescence imaging instrument with an automated data-processing controller. In some implementations, the controller can receive consecutive image sequences of organelles within a microfluidic chamber and execute a machine-learning model trained to distinguish human age-related transient features from baseline dynamics. In response to detection of a transient event, the controller can trigger rapid solution exchange through a connected reservoir, inducing fixation, reagent exposure, or other predefined actions in real time. In some implementations, the same controller can continue monitoring post-trigger images to evaluate alterations in organelle-2-4919-3750-8870.1Atty Dkt.: 114198-0890 mobility, morphology, or interaction frequency, providing continuous feedback control during the imaging process.BRIEF DESCRIPTION OF THE FIGURES

[0011] FIGS. 1A - ID: Age-dependent changes to organelle dynamics in primary fibroblasts from humans of increasing age. (FIG. 1A) In human primary fibroblasts, mitochondrial network shapes were classified according to 3 main groups: fragmented, intermediate, or filamented. Applicant observed increased fragmentation in the older patient cells. (FIG. IB) Mitochondrial mobility in human primary fibroblasts was analyzed across different age groups using confocal timelapse microscopy in combination with particle image velocimetry (PIV). PIV was employed to quantify mitochondrial movement by tracking velocity fields generated by particle displacements over time, revealing an age-dependent decline in mitochondrial movement. (FIG. 1C) Applicant observed a correlated increase in superoxide activity in the same human primary fibroblasts used in (B), highlighting a correlation between impaired mitochondrial dynamics and health. (FIG. ID) Lysosomal mobility was assessed using confocal timelapse microscopy and multi-object tracking analysis, which demonstrated a comparable age-associated reduction in lysosomal mobility, suggesting a broader decline in organelle transport with aging. Individual dots represent single-cell biological replicates with each cell from a separate coverslip. N= 25 000 cells / coverslips / wells were used for the superoxide assay. *p<0.05, **p<0.001, ***p<0.0001. ns: not significant.

[0012] FIG. 2 : AD patient cells display age-dependent alterations to organelle morphology and inter-organelle contacts. A representative fibroblast derived from an 81 -year old AD patient labeled with mitotracker (red) and lysotracker (cyan) shows significant amounts of colocalization between lysosomes and mitochondria. Quantification of colocalization in fibroblasts of increasing age show corresponding increases in mitochondria-lysosome colocalization.

[0013] FIG. 3: Phototoxicity markedly impacts organelle mobility in unpredictable ways. Cells expressing a mitochondrial outer membrane targeted GFP (Fisl-EGFP) were imaged with simultaneous epifluorescence and quantitative phaseimaging (QPI). The non-transfected neighboring cells quickly displayed reduced organelle mobility and eventual cell death, while the cells expressing Fisl-GFP remained normal over the same time. In contrast, cells-3-4919-3750-8870.1Atty Dkt.: 114198-0890 expressing cytosolic GFP (EGFP) displayed reduced organelle mobility and death faster than neighboring non-transfected cells.

[0014] FIG. 4 : Overview of QPI-fluorescence imaging system. Applicant built a custom QPI+epifluorescence imaging system based on the Nikon Ti2e system (Japan) with a raised dual-wheel deck and the PhiOptics SLIM module, with optimized optics for high spatiotemporal resolution with minimal lag between QPI and fluorescence image acquisition to ensure sufficient overlay of QPI+fluo imaging of rapidly moving organelles.

[0015] FIG. 5 : Virtual staining of mitochondria from raw QPI input data. Live-cell QPI-fluo videomicroscopy data was used to test a deep learning-based model that predicts mitochondria signal from raw QPI data. Fluorescence imaging of mito-GFP was used to evaluate the results, which showed a Pearson’s of 0.93.

[0016] FIGS. 6A - 6B: Timelapses of quasi-simultaneous QPI+fluorescence imaging of multiple organelles. (FIG. 6A) U2OS cells expressing the mitochondrial marker COX8-GFP were imaged using the system described in Fig. 4. Yellow arrowheads highlight mitochondrial remodeling events observed in both QPI and fluorescence channels, highlighting Applicant’s ability to capture mitochondrial dynamics with sufficient spatiotemporal resolution to ensure proper alignment for downstream deep learning-based training tasks. (FIG. 6B) Rapidly moving lysosomal vesicles were imaged using lysotracker green dye. As shown in Figures 1- 2, alterations to organelle morphology and mobility such as these are directly correlated with aging.

[0017] FIG. 7: Quantitative Phase + fluorescence imaging of organelles. Squares depict different organelles: a. Mitochondria; b. Lipid droplets; c. Endoplasmic reticulum; d. Golgi; e. Vesicles (lysosomes, endosomes, peroxi somes). Lentiviral transduction with GFP-tagged organelle targeting sequences enables QPI+fluorescence videomicroscopy of organelles with high specificity.

[0018] FIG. 8: Deep learning-based in silica virtual staining of mitochondria from label- free QPI videomicroscopy data as validated by ground-truth fluorescence data. Applicant trained a deep convolutional neural network to predict the fluorescence signal from raw QPI videomicroscopy data acquired using thequasi-simultaneous QPI+fluorescence imaging system described in Aim 1. These results prove Applicant’s ability to track mitochondrial-4-4919-3750-8870.1Atty Dkt.: 114198-0890 dynamics (note the mitochondrial fusion event marked by white arrowheads) from QPI data alone, as the deep learning (DL)-based inference is nearly indistinguishable from the fluorescence signal, with a Pearson’s correlation coefficient of 0.93. Given these results were generated using minimal training data (10 timelapses), Applicant are optimistic Applicant is able to generate highly accurate models with the proposed volume of data collection in Aim 1.

[0019] FIG. 9: Deep learning-based imaging of organelle mobility defects. LEFT: Applicant’s model tracks instances across frames by extracting features with convolutional neural networks and learning associations which are used to generate trajectories for each instance. RIGHT: Workflow integrating label-free QPI imaging with deep learning-based tracking of organelle mobility. Applicant’s tracking algorithm surpasses the accuracy of commercial IMARIS software. Raw label-free QPI images are acquired, and either virtual organelles are extracted from the raw data or simultaneously imaged fluorescent organelles are directly tracked in healthy vs. patients with Charcot-Mari e-Tooth disease mutations that impact organelle mobility. Kymographs or organelle tracks are used to quantify mobility. These results are representative of hundreds of organelles (small circles) from cells imaged from n=4 biological replicates (large circles). p<0.05, two-way t-test with the mean value of n=4 biological replicates used for statistical calculations.

[0020] FIG. 10: Particle image velocimetry of organelles in label-free data. Using semiautomated selection of comparable regions of interests in different human fibroblasts, Applicant can quantify and compare and generate “fingerprints” of organelle dynamics between different cell lines. These “mobility heatmaps” are highly compressed representations of the content-rich datasets Applicant generated. In Aim 2, Applicant propose to use deep learning-based methods to classify videomicroscopy data with much higher accuracy and information.

[0021] FIG. 11: Mitochondrial fission events captured via simultaneous QPI + fluorescence microscopy. LEFT: Raw QPI images of a mitochondrion over the 2 second course of a fission event MIDDLE: Quasi-simultaneously captured fluorescence of COX8- EGFP during the same mitochondrial fision event shown on the left. RIGHT: The overlay of QPI+fluorescence imaging shows the excellent overlay capabilities of the quasi-simultaneous imaging system. Applicant annotated hundreds of timelapses to train a deep learning model to predict mitofission events.-5-4919-3750-8870.1Atty Dkt.: 114198-0890

[0022] FIGS. 12A - 12B: CMT patient fibroblasts display actin-dependent organelle immobilization and aberrant actin assembly. (FIG. 12A) Lysosomes from two CMT patient and healthy control fibroblasts were imaged for 5 min. Rainbow “dragontail” tracks show organelle tracks. The mean distance for each organelle was quantified. Error bars: 5.E.M, n2200 tracks / organelle for n=10 cells, n=3 biological replicates. Similar results were observed for endosomes & mitochondria (not shown) (FIG. 12B) INF2 CMT patient fibroblasts had significantly increased actin aggregates. These structures invariably contained organelles. n=100 cells, n=3 biological replicates per condition. ***"indicate pO.OOOl, Welch's t-test. Scale bars: lOum.

[0023] FIG. 13 illustrates a system for detecting age-related transient events in a fluid using label-free imaging, in accordance with some embodiments.

[0024] FIG. 14 illustrates a flowchart of a method for detecting age-related transient events in a fluid using label-free imaging, in accordance with some embodiments.

[0025] FIG. 15A illustrates a block diagram of a computing device, in accordance with some embodiments.

[0026] FIG. 15B illustrates a block diagram of a computing device, in accordance with some embodiments.DETAILED DESCRIPTION

[0027] Definitions

[0028] Unless defined otherwise, all technical and scientific terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Although any methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, the preferred methods, devices, and materials are now described. All technical and patent publications cited herein are incorporated herein by reference in their entirety. Nothing herein is to be construed as an admission that the invention is not entitled to antedate such disclosure by virtue of prior invention.

[0029] Throughout and within this application technical and patent literature are referenced by a citation. For certain of these references, the identifying citation is found at the end of this-6-4919-3750-8870.1Atty Dkt.: 114198-0890 application immediately preceding the claims. All publications are incorporated by reference into the present disclosure to more fully describe the state of the art to which this disclosure pertains.

[0030] The practice of the present disclosure employs, unless otherwise indicated, conventional techniques of tissue culture, immunology, molecular biology, microbiology, cell biology and recombinant DNA, which are within the skill of the art. See, e.g., Sambrook and Russell eds. (2001) Molecular Cloning: A Laboratory Manual, 3rdedition; the series Ausubel et al. eds. (2007) Current Protocols in Molecular Biology; the series Methods in Enzymology (Academic Press, Inc., N.Y.); MacPherson et al. (1991) PCR 1 : A Practical Approach (IRL Press at Oxford University Press); MacPherson et al. (1995) PCR 2: A Practical Approach; Harlow and Lane eds. (1999) Antibodies, A Laboratory Manual; Freshney (2005) Culture of Animal Cells: A Manual of Basic Technique, 5thedition; Gait ed. (1984) Oligonucleotide Synthesis; U.S. Patent No. 4,683,195; Hames and Higgins eds. (1984) Nucleic Acid Hybridization; Anderson (1999) Nucleic Acid Hybridization; Hames and Higgins eds. (1984) Transcription and Translation; Immobilized Cells and Enzymes (IRL Press (1986)); Perbal (1984) A Practical Guide to Molecular Cloning; Miller and Calos eds. (1987) Gene Transfer Vectors for Mammalian Cells (Cold Spring Harbor Laboratory); Makrides ed. (2003) Gene Transfer and Expression in Mammalian Cells; Mayer and Walker eds. (1987) Immunochemical Methods in Cell and Molecular Biology (Academic Press, London); Herzenberg et al. eds (1996) Weir’s Handbook of Experimental Immunology; Manipulating the Mouse Embryo: A Laboratory Manual, 3rdedition (Cold Spring Harbor Laboratory Press (2002)); Sohail (ed.) (2004) Gene Silencing by RNA Interference: Technology and Application (CRC Press).

[0031] All numerical designations, e.g., pH, temperature, time, concentration, and molecular weight, including ranges, are approximations which are varied ( + ) or ( - ) by increments of 0.1 or 1.0, where appropriate. It is to be understood, although not always explicitly stated that all numerical designations are preceded by the term “about.” It also is to be understood, although not always explicitly stated, that the reagents described herein are merely exemplary and that equivalents of such are known in the art.-7-4919-3750-8870.1Atty Dkt.: 114198-0890

[0032] As used in the specification and claims, the singular form “a”, “an” and “the” include plural references unless the context clearly dictates otherwise. For example, the term “a cell” includes a plurality of cells, including mixtures thereof.

[0033] As used herein, the term “comprising” or “comprises” is intended to mean that the compositions and methods include the recited elements, but not excluding others. “Consisting essentially of’ when used to define compositions and methods, shall mean excluding other elements of any essential significance to the combination for the stated purpose. Thus, a composition consisting essentially of the elements as defined herein would not exclude trace contaminants from the isolation and purification method and pharmaceutically acceptable carriers, such as phosphate buffered saline, preservatives and the like. “Consisting of’ shall mean excluding more than trace elements of other ingredients and substantial method steps for administering the compositions of this invention or process steps to produce a composition or achieve an intended result. Embodiments defined by each of these transition terms are within the scope of this disclosure.

[0034] The term “isolated” as used herein with respect to nucleic acids, such as DNA or RNA, refers to molecules separated from other DNAs or RNAs, respectively that are present in the natural source of the macromolecule. The term “isolated nucleic acid” is meant to include nucleic acid fragments which are not naturally occurring as fragments and would not be found in the natural state. The term “isolated” is also used herein to refer to polypeptides, proteins and / or host cells that are isolated from other cellular proteins and is meant to encompass both purified and recombinant polypeptides. In other embodiments, the term “isolated” means separated from constituents, cellular and otherwise, in which the cell, tissue, polynucleotide, peptide, polypeptide, protein, antibody or fragment(s) thereof, which are normally associated in nature. For example, an isolated cell is a cell that is separated form tissue or cells of dissimilar phenotype or genotype. As is apparent to those of skill in the art, a non-naturally occurring polynucleotide, peptide, polypeptide, protein, antibody or fragment(s) thereof, does not require “isolation” to distinguish it from its naturally occurring counterpart.

[0035] “Homology” or “identity” or “similarity” refers to sequence similarity between two peptides or between two nucleic acid molecules. Homology can be determined by comparing a position in each sequence which may be aligned for purposes of comparison. When a position in the compared sequence is occupied by the same base or amino acid, then the molecules are -8-4919-3750-8870.1Atty Dkt.: 114198-0890 homologous at that position. A degree of homology between sequences is a function of the number of matching or homologous positions shared by the sequences. An “unrelated” or “non-homologous” sequence shares less than 40% identity, or alternatively less than 25% identity, with one of the sequences of the present disclosure.

[0036] A polynucleotide or polynucleotide region (or a polypeptide or polypeptide region) has a certain percentage (for example, 70%, 75%, 80%, 85%, 90%, 95%, 98% or 99%) of “sequence identity” to another sequence means that, when aligned, that percentage of bases (or amino acids) are the same in comparing the two sequences. This alignment and the percent homology or sequence identity can be determined using software programs known in the art, for example those described in Ausubel et al. eds. (2007) Current Protocols in Molecular Biology. Preferably, default parameters are used for alignment. One alignment program is BLAST, using default parameters. In particular, programs are BLASTN and BLASTP, using the following default parameters: Genetic code = standard; filter = none; strand = both; cutoff = 60; expect = 10; Matrix = BLOSUM62; Descriptions = 50 sequences; sort by = HIGH SCORE; Databases = non-redundant, GenBank + EMBL + DDBJ + PDB + GenBank CDS translations + SwissProtein + SPupdate + PIR. Details of these programs can be found at the following Internet address: ncbi.nlm.nih.gov / cgi-bin / BLAST.

[0037] An equivalent or biological equivalent nucleic acid, polynucleotide or oligonucleotide or peptide is one having at least 80 % sequence identity, or alternatively at least 85 % sequence identity, or alternatively at least 90 % sequence identity, or alternatively at least 92 % sequence identity, or alternatively at least 95 % sequence identity, or alternatively at least 97 % sequence identity, or alternatively at least 98 % sequence identity to the reference nucleic acid, polynucleotide, oligonucleotide or peptide.

[0038] “Detectable label”, “label”, “detectable marker” or “marker” are used interchangeably, including, but not limited to radioisotopes, fluorochromes, chemiluminescent compounds, dyes, and proteins, including enzymes. Detectable labels can also be attached to a polynucleotide, polypeptide, antibody or composition described herein.

[0039] Examples of suitable fluorescent labels include, but are not limited to, fluorescein, rhodamine, tetramethylrhodamine, eosin, erythrosin, coumarin, methyl-coumarins, pyrene, Malacite green, stilbene, Lucifer Yellow, Cascade Blue™, and Texas Red. Other suitable-9-4919-3750-8870.1Atty Dkt.: 114198-0890 optical dyes are described in the Haugland, Richard P. (1996) Handbook of Fluorescent Probes and Research Chemicals (6th ed.).

[0040] In some embodiments, the fluorescent label is functionalized to facilitate covalent attachment to a cellular component present in or on the surface of the cell or tissue such as a cell surface marker. Suitable functional groups, include, but are not limited to, isothiocyanate groups, amino groups, haloacetyl groups, maleimides, succinimidyl esters, and sulfonyl halides, all of which may be used to attach the fluorescent label to a second molecule. The choice of the functional group of the fluorescent label depends on the site of attachment to either a linker, the agent, the marker, or the second labeling agent.

[0041] Modes For Carrying Out The Disclosure

[0042] Aging is characterized by progressive cellular dysfunction, with alterations in organelle dynamics playing a crucial role in age-related disease. However, Applicant’s understanding of how organelles change with age is limited by current imaging technologies, which lack the resolution and non-invasive capabilities needed to capture subtle alterations in live cells. This technological gap hinders progress in aging research and the development of interventions to promote healthy aging. There is an urgent need for innovative imaging approaches that can reveal age-related changes in organelle behavior with unprecedented detail and minimal perturbation.

[0043] This disclosure aims to develop cutting-edge computational imaging tools to study age- related changes in organelle dynamics using primary human patient skin fibroblasts from donors across the lifespan. By combining label-free quantitative phase imaging (QPI) with advanced deep learning (DL), Applicant created a new paradigm for high-content, minimally invasive imaging of age-related cellular changes. Applicant’s long-term goal is to establish datasets and toolkits that empower researchers to uncover fundamental principles of organelle aging in diverse cellular contexts. Following the requirements for PAR-22-127, Applicant’s three aims are focused solely on technology development:Aim 1: Develop a high-performance multimodal QPI-fluorescence imaging system optimized for primary human fibroblasts.

[0044] Applicant developed a QPI microscope integrated with quasi-simultaneous fluorescence imaging to achieve:-10-4919-3750-8870.1Atty Dkt.: 114198-0890• Spatial resolution of <250 nm for QPI and <120 nm for fluorescence• Temporal resolution of >5 frames per second when combined• Field of view >130 x 130 pm to capture entire fibroblasts• Generate QPI-fluorescence datasets for multiple organelles including mitochondria, lysosomes, autophagosomes, endoplasmic reticulum, nuclear envelope, histones, and condensates.

[0045] Benchmark: Acquire correlative QPI-fluorescence datasets of key organelles in ^500 primary fibroblasts from donors aged 20-90 years, imaged over 24-hour periods with <5% photobleaching or phototoxicity.Aim 2: Create tools for automated analysis of age-related organelle changes in QPI data.

[0046] Applicant developed and validate DL models for:• Segmentation of individual organelles with90% accuracy compared to manual annotation• Tracking of organelle movements with 2s 85% accuracy over 1000-frame videos• Classification of organelle types and morphologies with 95% accuracy• Detection of age-related organelle features (e.g., enlarged lysosomes, fragmented mitochondria) with80% accuracy

[0047] Benchmark: Process QPI videos of >5000 fibroblasts across the age spectrum, automatically extracting quantitative data on organelle morphology, dynamics, and interactions. Generate deep learning-based in silico virtual staining of each organelle, enabling label-free tracking of each organelle. Identify at least 3 novel age-related organelle phenotypes that predict a patient's age, validated by holdout datasets.Aim 3: Engineer a "self-driving" microscope for adaptive imaging of age-related organelle events.

[0048] Applicant created a closed-loop system integrating real-time DL analysis with microfluidics to:-11-4919-3750-8870.1Atty Dkt.: 114198-0890• Detect specific age-related organelle events (e.g., mitochondrial dynamics) within 100 ms of occurrence• Trigger microfluidic solution exchange within 500 ms of event detection• Achieve 90% success rate in capturing targeted events

[0049] Benchmark: Automatically detect and chemically fix >500 fibroblasts within 1 second of identified age-related organelle events, enabling downstream correlative light and electron microscopy analysis. Compare event frequencies between young (20-40 years), middle-aged (41-60 years), and old (61+ years) donor cells.

[0050] This innovative approach combines expertise in optical engineering, machine learning, and aging biology to achieve an unprecedented view of age-related changes in organelle dynamics.

[0051] The resulting toolkit has a broad impact, enabling researchers across disciplines to make new discoveries about organelle function in aging and age-related diseases. Moreover, the intelligent microscopy platform serves as a foundation for high-throughput screening assays to identify interventions that modulate age-related organelle changes, accelerating the development of therapies to promote healthy aging.Beyond Mitochondria: The Multi-Organelle Impact of Aging

[0052] Aging is characterized by progressive cellular dysfunction, with alterations in organelle dynamics playing a crucial role in age-related decline. Most studies focus on mitochondria which are directly implicated in aging. Emerging evidence suggests inter-organelle contacts between mitochondria, the ER, and lysosomes mediates different subtypes of mitochondrial fission and mitophagy. However, Applicant’s understanding of how these and other organelles change with age is limited by current imaging technologies, which lack the resolution and non- invasive capabilities needed to capture subtle alterations in living cells over extended periods. This technological gap hinders progress in aging research and the development of interventions to promote healthy aging. There is an urgent need for innovative imaging approaches that reveal age-related changes in organelle behavior with unprecedented detail and minimal perturbation. Applicant’s data show that not only do mitochondrial dynamics and function decrease with age, but so does lysosome mobility (FIG. 1).-12-4919-3750-8870.1Atty Dkt.: 114198-0890

[0053] These results showing alterations to not just mitochondria, but also lysosome dynamics, highlight the need to investigate the dynamics of multiple organelles simultaneously to capture the full landscape of age-related changes. In further support of this claim, Applicant observed an age-dependent increase in association between mitochondria and lysosomes in Alzheimer’s disease patients (FIG. 2).

[0054] Detailed mechanistic investigations of organelle movements are hindered by challenges in observing and analyzing these dynamic processes. The highly dynamic nature of organelles within live cells necessitate both fast and precise monitoring and quantification strategies. While fluorescence microscopy offers exceptional resolution through fluorescent labeling, its inherent drawbacks, including perturbation of cellular processes and limitations in live-cell imaging, hinder comprehensive understanding. Moreover, the constrained number of observable channels restricts the breadth of investigation to usually just 2 or 3 organelles at a time, greatly reducing the throughput and context of each experiment. Further, labeling organelles with fluorescent proteins or dyes involves sample manipulations that are laborious, expensive, and often perturb cellular physiology in unpredictable ways. For example, Applicant discovered that organelle mobility decreases as an acute response to phototoxicity, introducing a potentially underappreciated confounding factor for all organelle studies to date. Just as importantly, Applicant discovered the phototoxic response is highly variable depending on the location of the GFP molecule; localizing GFP to the outer mitochondrial membrane paradoxically protected cells against phototoxicity, whereas cytosolic GFP sensitized cells to phototoxicity (FIG. 3).

[0055] To address these challenges, Applicant developed quantitative phase imaging (QPI) label-free imaging assays capable of visualizing organelles in their native state without the need for invasive techniques. QPI minimizes sample manipulation, enabling simultaneous observation of multiple biological structures with reduced phototoxicity and no photobleaching, rendering it ideal for continuous live-cell imaging. QPI enables the monitoring of all organelles within a cell simultaneously, yet they lack the specificity required to focus on individual organelles. To address this limitation, Applicant built a custom imaging system for quasi-simultaneous QPI+fluorescence microscopy (FIG. 4). Quasi-simultaneity is essential for capturing live moving organelles with near-perfect overlays of QPI and fluorescence images-13-4919-3750-8870.1Atty Dkt.: 114198-0890 that are suitable for deep learning models that virtually stain organelles in QPI datasets. Applicant’s data on mitochondria show feasibility of Applicant’s approach (FIG. 5).

[0056] Applicant’s technology development directly addresses these challenges and needs by creating a powerful new toolkit for studying age-related changes in organelle dynamics. By combining label-free quantitative phase imaging (QPI) with advanced deep learning (DL), Applicant enable researchers to:

[0057] 1. Visualize and quantify age-related changes in organelle morphology, dynamics, and interactions with high spatiotemporal resolution and minimal phototoxicity. This allows for long-term studies of how organelle behavior changes throughout the lifespan of cells, tissues, and organisms.

[0058] 2. Automatically detect and analyze subtle age-related phenotypes in organelle function that may be precursors to cellular dysfunction. The high-throughput nature of Applicant’s approach facilitates the identification of early biomarkers of cellular aging.

[0059] 3. Capture transient organelle events associated with aging processes, such as mitochondrial fragmentation or lysosomal fusion, with unprecedented temporal precision. This provides new insights into the mechanisms underlying age-related organelle dysfunction.

[0060] 4. Perform large-scale, unbiased analyses of organelle dynamics across different cell types, tissues, and ages, potentially uncovering novel patterns and relationships in cellular aging.

[0061] The significance of this technology development for aging research is multifaceted:

[0062] Mechanistic Insights: By enabling detailed, long-term observation of organelle dynamics in aging cells, Applicant’s toolkit helps elucidate the mechanisms underlying age- related cellular dysfunction. This could lead to the identification of new therapeutic targets for age-related diseases.

[0063] Biomarker Discovery: The ability to detect subtle changes in organelle behavior could reveal new biomarkers of cellular aging, potentially allowing for earlier diagnosis and intervention in age-related disorders.-14-4919-3750-8870.1Atty Dkt.: 114198-0890

[0064] Comparative Studies: Applicant’s high-throughput, label -free approach facilitates large-scale studies comparing organelle dynamics across different ages, cell types, and genetic backgrounds. This could uncover cell type-specific or tissue-specific aspects of aging.

[0065] Intervention Testing: The non-invasive nature of Applicant’s imaging technology makes it ideal for testing interventions aimed at modulating age-related changes in organelle function. This could accelerate the development of therapies to promote healthy aging.

[0066] Integration with Other Aging Research Tools: Applicant’s platform can be readily integrated with other cutting-edge tools in aging research, such as single-cell sequencing or proteomics, providing a multi-modal view of cellular aging processes.

[0067] By developing this innovative imaging and analysis toolkit, Applicant provides the aging research community with powerful new capabilities to study the fundamental cellular processes underlying aging and age-related diseases. The nature of Applicant’s tools ensures widespread adoption and continued development, maximizing the impact on the field. Ultimately, this technology has the potential to accelerate discoveries in aging biology and contribute to the development of interventions that promote healthier aging.

[0068] This disclosure introduces several highly innovative aspects that significantly advances the field of aging research, particularly in the study of age-related changes in organelle dynamics:

[0069] Label-free, high-resolution imaging of multiple organelles over extended periods: Applicant’s multimodal QPI-fluorescence system represents a significant leap forward in label- free imaging technology for aging research. By achieving spatial resolution of -120 nm for fluorescence and -250 nm for QPI, with temporal resolution of -5 frames per second, Applicant enables visualization of age-related changes in organelle dynamics at an unprecedented level of detail without the need for exogenous labels. This approach overcomes the limitations of current fluorescence-based methods, which are restricted by phototoxicity and the number of spectrally distinct fluorophores that can be used simultaneously. Applicant’s system allows for long-term imaging of multiple organelles in their native state throughout the aging process, providing a more comprehensive and physiologically relevant view of age- related cellular changes.-15-4919-3750-8870.1Atty Dkt.: 114198-0890

[0070] Deep learning-based analysis of age-related changes in QPI data: While QPI generates information-rich datasets, extracting meaningful biological information about aging processes from these images remains challenging. Applicant’s innovative application of deep learning to QPI data analysis transforms the utility of this imaging modality for aging research. By developing models for organelle segmentation, tracking, and classification with high accuracy (>90% for segmentation, >85% for tracking, and >95% for classification), Applicant enables automated, high-throughput analysis of age-related changes in organelle dynamics from label-free images. This approach not only increases the speed and scale of data analysis but also has the potential to reveal novel patterns and relationships in organelle aging not obvious to human observers.

[0071] Intelligent "self-driving" microscopy system for capturing age-related organelle events: The development of a closed-loop system that can detect specific age-related cellular events and trigger automated responses in real-time represents a paradigm shift in aging research microscopy. By integrating real-time deep learning analysis with microfluidics, Applicant’s system is capable of detecting events such as mitochondrial fragmentation or lysosomal fusion within 100 ms and triggering solution exchange within 1 second. This level of intelligent automation enables entirely new experimental paradigms in aging research, such as precise temporal capture of transient cellular events associated with the aging process for downstream analysis.

[0072] Correlative QPI-fluorescence imaging for model training and validation in aging studies: Applicant’s approach of using correlative QPI-fluorescence imaging to train and validate deep learning models for QPI data analysis is uniquely innovative in the context of aging research. This allow us to leverage the specificity of fluorescence labeling to inform the interpretation of label-free QPI data, enabling the extraction of age-related biological information from unlabeled living cells over extended periods. This is particularly crucial for aging studies, where long-term observation of cellular processes is essential.

[0073] Toolkit for broad accessibility in the aging research community: By developing Applicant’s imaging system and analysis tools as a toolkit, Applicant fosters innovation and collaboration across the aging research community. This accelerates the adoption and further development of Applicant’s technology, maximizing its impact on the field of aging biology.-16-4919-3750-8870.1Atty Dkt.: 114198-0890

[0074] Potential for high-throughput screening of age-related interventions: The combination of label-free imaging, automated analysis, and intelligent microscopy control opens up innovative possibilities for high-throughput screening assays in aging research. Applicant’s system could be used to rapidly assess the effects of large libraries of compounds on age-related changes in organelle dynamics, potentially accelerating the discovery of interventions that promote healthy aging.

[0075] In summary, Applicant’s proposed technology development represents a convergence of cutting-edge optical engineering, machine learning, and aging biology that creates an unprecedented view of age-related changes in organelle dynamics in living cells. By overcoming current limitations in resolution, specificity, and throughput, Applicant’s innovations enables new discoveries about fundamental cellular processes underlying aging and their dysregulation in age-related diseases. The resulting toolkit has a broad impact, empowering researchers across disciplines to make transformative discoveries in aging biology and biomedicine.Overview

[0076] This disclosure outlines the development of an innovative imaging and computational toolkit to study age-related changes in organelle dynamics, addressing a critical gap in aging research. By combining label-free quantitative phase imaging (QPI) with advanced deep learning (DL) and intelligent microscopy control, Applicant aims to empower researchers to visualize, quantify, and understand the intricate relationship between organelle behavior and the aging process. The proposed technology development adheres to the PAR-22-127 guidelines, focusing solely on creating novel tools with broad applicability in aging research. The working prototypes resulting from this project significantly advances the state-of-the-art in live-cell imaging and automated image analysis, facilitating discoveries in aging biology and age-related diseases.1. Develop a High-Performance Multimodal QPI-Fluorescence Imaging System Optimized for Primary Human Fibroblasts

[0077] Current imaging technologies lack the resolution, sensitivity, and non-invasive capabilities required to capture subtle age-related alterations in organelle dynamics within living cells with minimal perturbation. The multimodal QPI-fluorescence system Applicant-17-4919-3750-8870.1Atty Dkt.: 114198-0890 proposes represents a significant advancement, enabling high-resolution, label-free visualization of multiple organelles simultaneously, alongside targeted fluorescence imaging for validation and deep learning model training. This approach overcomes limitations of existing methods, providing a comprehensive, physiologically relevant view of age-related cellular changes with minimal perturbation.Data

[0078] Applicant’s team has extensive experience in developing and implementing advanced, customized QPI and fluorescence microscopy systems for live-cell imaging with minimal lag between QPI and fluorescence imaging on a frame-by-frame basis (FIG. 6). Applicant has successfully demonstrated the feasibility of combining QPI with fluorescence imaging to study dynamics of multiple organelles in multiple cell types. Applicant’s data show that QPI captures the subtle changes in organelle morphology and mobility associated with aging in primary human fibroblasts (FIG. 5).

[0079] Applicant has generated lentiviruses with GFP-tagged organelle markers for ER, Golgi, mitochondria, lysosomes, endosomes, autophagosomes, lipid droplets, nuclear envelope, and histones. Applicant images each of these organelles with quasi-simultaneous QPI+fluorescence microscopy in human primary fibroblasts to build a complete map of age-related changes to each subcellular compartment while at the same time building training data for deep learningbased label-free in silico virtual staining (FIG. 7).Experimental Approach

[0080] Applicant can develop a state-of-the-art multimodal QPI-fluorescence microscope optimized for imaging primary human fibroblasts, addressing the unique challenges posed by these cells. The system is engineered to achieve the following benchmarks:

[0081] 1.1 Spatial Resolution Optimization• QPI resolution: < 250 nm• Fluorescence resolution: < 120 nm

[0082] Methodology: Applicant employs a combination of optimized optical design and deconvolution algorithms to minimize aberrations and maximize resolution. The system is-18-4919-3750-8870.1Atty Dkt.: 114198-0890 rigorously validated using standardized test samples, including fluorescent beads and resolution test patterns, ensuring exceptional image quality across both modalities.

[0083] 1.2 Temporal Resolution Optimization• Target: > 5 frames per second for both QPI and fluorescence

[0084] Methodology: Applicant implemented high-speed scientific CMOS cameras, fastswitching transmitted and reflected LED light sources, and optimize data acquisition and processing pipelines to achieve the target frame rate. The system's performance is validated using high-speed test samples, such as lysosomes in living cells, ensuring the ability to capture rapid organelle dynamics in real time.

[0085] 1.3 Field of View Optimization• Target: > 100 x 100 pm to capture representative fields of view

[0086] Methodology: Applicant carefully balanced optical design considerations to maximize the field of view while maintaining high resolution across the entire imaging area. This enables simultaneous visualization of multiple subcellular compartments, increasing throughput and content depth in the study of subcellular changes during aging.

[0087] 1.4 Quasi-simultaneous QPI+Fluorescence Imaging of Key Subcellular Compartments / Organelles

[0088] Target: Image all key subcellular compartments implicated in aging with quasi- simultaneous QPI+fluorescence

[0089] Methodology: Applicant generates lentiviruses expressing GFP-tagged markers for each key subcellular compartment in the cell, including: mitochondria, lysosomes, endosomes, autophagosomes, endoplasmic reticulum, golgi apparatus, nuclear envelope, histones, actin and microtubule filaments, plasma membrane, endocytic pits, stress granules, and peroxisomes. This enables us to correlate label-free QPI data with each subcellular compartment, informing deep learning model training and validation efforts in Aim 2.

[0090] 1.5 System Integration and Validation-19-4919-3750-8870.1Atty Dkt.: 114198-0890

[0091] Benchmark: Acquire correlative QPI-fluorescence datasets of each subcellular compartment in > 500 primary fibroblasts from donors aged 20-90 years, imaged with sufficient spatiotemporal resolution to capture organelle remodeling and movement.

[0092] Methodology:• Cell Preparation: Primary human skin fibroblasts from donors across the age spectrum is cultured under standard conditions and stably transfected with fluorescent organelle markers.• Imaging Protocol: An automated imaging sequence is developed, alternating between QPI and fluorescence modes at high speed (> 5 fps) to capture both label-free and labeled organelle dynamics. Long-term time-lapse imaging (up to 24 hours) is performed to assess cell-cycle related changes.• Phototoxicity and Photobleaching Assessment: Cell morphology, organelle dynamics, and viability is monitored. Photobleaching is quantified using fluorescence intensity measurements.• Data Analysis: Custom software is developed to align and correlate QPI and fluorescence data, enabling quantitative comparison of organelle detection, tracking between modalities, and downstream training of deep learning-based networks for in silico virtual staining of organelles (Aim 2).

[0093] Conclusions

[0094] Applicant anticipates achieving the target specifications for spatial resolution, temporal resolution, field of view, and multi-organelle fluorescence imaging. The integrated system enables long-term, non-invasive visualization of multiple organelles in primary human fibroblasts, providing a rich dataset for studying age-related changes in organelle dynamics.

[0095] 2. Create Tools for Automated Analysis of Age-Related Organelle Changes in QPI Data

[0096] Rationale

[0097] Extracting meaningful biological insights from QPI data requires sophisticated image analysis tools. Applicant has developed and applied deep learning models for biomedical image-20-4919-3750-8870.1Atty Dkt.: 114198-0890 analysis. Applicant leverages the power of deep learning to develop software for automated in silico virtual staining, tracking, classification, and age-related feature detection for each subcellular compartment in QPI videomicroscopy of primary human fibroblasts. These tools enables high-throughput, unbiased quantification of organelle dynamics, facilitating the discovery of novel age-related phenotypes and biomarkers.

[0098] Data

[0099] Using mitochondria as a test case, Applicant successfully demonstrated the feasibility of using DL to virtually stain and track organelles in QPI data (FIG. 8).

[0100] Applicant also generated a workflow for automated tracking of multiple objects or organelles in videomicroscopy data using global tracking transformers. This approach greatly outperforms commercial software in the challenging task of properly assigning object tracks which is essential for quantifying differences in organelle mobility between healthy and diseased states, as illustrated in FIG. 9 showing results with CMT patient cells.

[0101] Results show promise in detecting age-related morphological and dynamic changes to subcellular compartments in fibroblasts. Using particle image velocimetry of raw QPI videomicroscopy data, Applicant generated “mobility heatmaps” reflecting differences in organelle mobility (FIG. 10)

[0102] Applicant developed and validated a suite of DL models tailored for analyzing QPI data from primary human fibroblasts, addressing the specific challenges associated with these cells and the aging process. The models is trained and evaluated using the correlative QPL fluorescence datasets acquired in 1 above, ensuring high accuracy and biological relevance.

[0103] 2.1 Organelle Virtual Staining

[0104] Target: > 90% accuracy compared to manual annotationMethodology:• Dataset Preparation: A large, diverse dataset of correlative QPI-fluorescence images is curated using the different labels described in 1, featuring fibroblasts from donors across the age spectrum. Organelles is segmented from the fluorescence images to create ground truth masks.-21-4919-3750-8870.1Atty Dkt.: 114198-0890• Model Development: State-of-the-art segmentation architectures, such as U-Net, DenseNets, Transformer architectures, or Mask R-CNN, is implemented and trained on the QPI data, using the fluorescence-based annotations as supervision.• Validation: Model performance can be evaluated on a held-out test set, comparing segmentation results to manual annotations using metrics like Intersection over Union (loU) and Fl score.

[0105] 2.2 Organelle Tracking• Target: > 85% accuracy over 1000-frame videos

[0106] Methodology:• Dataset Preparation: Long-term QPI videos (1000+ frames) capturing dynamic organelle movements in fibroblasts is acquired.• Model Development: Multiple object tracking algorithms, such as DeepSORT or FairMOT, is adapted and optimized for tracking organelles in QPI data, addressing challenges like occlusion and motion blur.• Validation: Tracking accuracy is evaluated using metrics like Multiple Object Tracking Accuracy (MOTA) and ID switches, comparing to manual tracking on a subset of frames.

[0107] 2.3 Organelle Classification• Target: > 95% accuracy

[0108] Methodology:• Dataset Preparation: The correlative QPI-fluorescence datasets is leveraged to create a comprehensive collection of organelle images with known classifications.• Model Development: Convolutional neural network (CNN) classifiers, such as ResNet or EfficientNet, is trained to distinguish between different organelle types based solely on QPI data.• Validation: Classification accuracy is assessed on a held-out test set, using metrics like precision, recall, and Fl score.

[0109] 2.4 Detection of Age-Related Organelle Features-22-4919-3750-8870.1Atty Dkt.: 114198-0890• Target: > 80% accuracy in detecting age-related organelle features (e.g., enlarged lysosomes, fragmented mitochondria)

[0110] Methodology:• Dataset Preparation: The extensive collection of QPI images from fibroblasts across the age spectrum is leveraged. Expert biologists annotate specific age-related organelle features in these images, providing ground truth for model training.• Model Development: Applicant explored a combination of traditional image processing techniques and deep learning approaches to identify and quantify age-related features in QPI data. This involved the use of convolutional neural networks (CNNs) for feature extraction and classification, or the development of custom algorithms tailored to specific organelle phenotypes.• Validation: Model performance is rigorously assessed by comparing automated feature detection to expert annotations, using metrics such as precision, recall, and Fl score. The models is further validated on independent datasets to ensure generalizability across different fibroblast populations and experimental conditions.

[0111] 2.5 Integrated Analysis Pipeline and Benchmark Dataset

[0112] Process QPI videos of > 5000 fibroblasts across the age spectrum, automatically extracting quantitative data on organelle morphology, dynamics, and interactions. Identify at least 3 novel age-related organelle phenotypes that predict a patient's age, validated by holdout datasets.

[0113] Methodology:• Pipeline Integration: A user-friendly software package can be developed, seamlessly integrating all the DL models into a cohesive analysis pipeline. The software is designed to be modular and adaptable, allowing researchers to customize the analysis workflow for their specific needs.• Large-Scale Dataset: The extensive QPI video dataset acquired in 1 serves as the benchmark for evaluating the pipeline's performance. Additional QPI videos from diverse fibroblast populations and experimental conditions may be incorporated to enhance the robustness and generalizability of the analysis tools.-23-4919-3750-8870.1Atty Dkt.: 114198-0890• Automated Analysis: The integrated pipeline is applied to the benchmark dataset, automatically extracting quantitative data on organelle morphology (e.g., size, shape, texture), dynamics (e.g., velocity, displacement, interactions), and age-related features.• Validation and Novel Phenotype Discovery: The accuracy and reliability of the automated QPI-based analysis is validated by comparing results to manual analysis of corresponding fluorescence data for a subset of cells. Advanced statistical and machine learning techniques is employed to identify novel age-related organelle phenotypes within the QPI data, and their predictive power for determining a patient's age is assessed using holdout datasets.

[0114] This disclosure has applications to empower researchers to efficiently and objectively quantify age-related changes in organelle dynamics, facilitating the discovery of novel biomarkers and therapeutic targets for age-related diseases.

[0115] 3: Engineer a "Self-Driving" Microscope for Adaptive Imaging of Age-Related Organelle Events

[0116] Capturing rare and transient age-related organelle events, such as mitochondrial fragmentation or lysosomal fusion, requires real-time detection and rapid response capabilities. Applicant developed an intelligent, closed-loop microscopy system that integrates real-time DL analysis with microfluidics to automatically detect and respond to specific organelle events, enabling precise temporal capture and downstream analysis.

[0117] Data

[0118] Applicant’s team has experience in developing custom microscopy systems and integrating them with microfluidic devices. In prior publications Applicant has successfully demonstrated the feasibility of using real-time image analysis to trigger microfluidic events in other biological contexts in service of live-to-fixed cell CLEM. Applicant’s data show that DL models can detect organelle events in QPI videos with high accuracy, providing a foundation for real-time analysis (FIG. 11).

[0119] A "self-driving" microscope that seamlessly integrates real-time DL analysis with microfluidic control, enabling automated detection and response to age-related organelle-24-4919-3750-8870.1Atty Dkt.: 114198-0890 events in primary human fibroblasts is engineered. The system is designed to achieve the following:

[0120] 3.1 Real-Time Event Detection• Target: Detect specific age-related organelle events (e.g., mitochondrial fragmentation, lysosomal fusion) within 100 ms of occurrence

[0121] Methodology:• Hardware Optimization: Applicant leveraged GPU acceleration and optimized Applicant’s image processing pipeline to minimize latency, ensuring rapid analysis of incoming QPI data.• Model Optimization: The event detection models developed in 2 is adapted for real-time inference, potentially employing techniques like model pruning or quantization to achieve the desired speed without sacrificing accuracy.• Validation: The system's real-time performance is rigorously evaluated using simulated data and pre-recorded videos, measuring the time between an event occurring and its detection.

[0122] 3.2 Rapid Microfluidic Actuation• Trigger microfluidic solution exchange within 500 ms of event detection.

[0123] Methodology:• Microfluidic Design: A custom microfluidic chip is designed and fabricated, optimized for compatibility with Applicant’s QPI system and featuring rapid solution exchange capabilities. The chip enables precise and controlled delivery of reagents or fixatives to the cells upon event detection.• Control System: A low-latency control system is developed to interface between the realtime analysis pipeline and the microfluidic device, ensuring rapid and reliable triggering of solution exchange upon event detection.• Validation: The speed and accuracy of microfluidic actuation is validated using fluorescent dyes, measuring the time between a software trigger and the completion of solution exchange.-25-4919-3750-8870.1Atty Dkt.: 114198-0890

[0124] 3.3 System Integration and Performance• Achieve > 90% success rate in capturing targeted events

[0125] Methodology:• Full System Integration: The real-time analysis pipeline, microfluidic control system, and QPI microscope is seamlessly integrated into a cohesive, automated platform.• Automated Experimental Protocol: Software is developed to manage the entire workflow, from continuous QPI monitoring to event detection, microfluidic actuation, and follow-up imaging.• Performance Evaluation: Rigorous experiments is conducted, targeting specific age- related organelle events. The success rate of event capture and subsequent microfluidic response is meticulously quantified, ensuring the system's reliability and effectiveness.

[0126] 3.4 Correlative Light and Electron Microscopy (CLEM) Workflow

[0127] Automatically detect and chemically fix > 500 fibroblasts within 1 second of identified age-related organelle events, enabling downstream correlative light and electron microscopy analysis. Compare event frequencies between young (20-40 years), middle-aged (41-60 years), and old (61+ years) donor cells.

[0128] Methodology:• Rapid Fixation Protocol: A chemical fixation compatible with the microfluidic system is optimized, ensuring rapid and effective preservation of cellular ultrastructure within 1 second of event detection.• CLEM Workflow: A streamlined protocol is established for retrieving fixed cells from the microfluidic chip and preparing them for CLEM analysis, enabling high-resolution visualization of organelle events in their native context.• Validation and Comparative Analysis: The ultrastructure of cells fixed immediately post-event is compared to control cells, evaluating the preservation of transient structures associated with age-related organelle events. Event frequencies is quantified and compared across different age groups (young, middle-aged, and old) to gain insights into the dynamics of organelle aging.-26-4919-3750-8870.1Atty Dkt.: 114198-0890

[0129] Applicant anticipates achieving high speed and accuracy in event detection, microfluidic actuation, and overall system performance. The intelligent microscope enables automated capture and preservation of rare and transient age-related organelle events, facilitating in-depth CLEM analysis and providing novel insights into the mechanisms of organelle aging.

[0130] Integration and Synergy

[0131] The three aims of this disclosure are highly interconnected and synergistic, forming a cohesive platform for studying age-related changes in organelle dynamics.• 1 provides the foundation by developing a high-performance imaging system capable of capturing both label-free and fluorescence data from multiple organelles in primary human fibroblasts.• 2 leverages the data from Aim 1 to train and validate deep learning models for automated analysis of QPI data, enabling high-throughput quantification of organelle dynamics and age-related features.• 3 integrates the DL models from Aim 2 with real-time image analysis and microfluidics to create an intelligent microscope capable of autonomously detecting and responding to specific organelle events, facilitating in-depth CLEM analysis.

[0132] This integrative approach ensures that each technological advance builds upon and enhances the others, creating a powerful and versatile toolkit for aging research.

[0133] Organelle Dysfunction at the Nexus of Neurodegeneration

[0134] The precise regulation of organelle activity, positioning, and turnover is paramount for cellular health and survival. Organelle dynamics, including fission / fusion and mobility, are orchestrated by the actin cytoskeleton and its regulatory proteins. In peripheral neurons, with their extensive projections demanding meticulously positioned organelles and extraordinarily long travel distances, the importance of organelle mobility and turnover is amplified1'4. Consequently, mutations that disrupt organelle mobility / turnover are implicated in a multitude of peripheral neurodegenerative diseases, including Charcot-Marie-Tooth (CMT) disease5'7, amyotrophic lateral sclerosis (ALS), and others. However, significant knowledge gaps remain to be filled to fully understand the fundamental mechanisms5, 6’8'10. For instance, conflicting-27-4919-3750-8870.1Atty Dkt.: 114198-0890 conclusions have been drawn regarding the impact of RAB7A mutations in CMT2B: 1) the functional consequences of these mutations on RAB7A activity are haploid insufficiency11vs gain of toxicity12'14. 2) The RAB7A mutations promote mitochondrial elongation15, 16vs fragmentation17, 18. By using innovative techniques and physiological models, our proposal is aimed at resolving these important mechanistic issues and informing future therapies.

[0135] CMT: A Focus on Axonal Degeneration

[0136] CMT, the most prevalent inherited neurological disorder, manifests as a progressive deterioration of peripheral nerves, leading to motor and sensory impairments6, 7’9. Despite the ubiquitous presence of CMT-implicated proteins throughout the body, the disease’s predilection for peripheral nerves with iong projections underscores the critical role of organelle mobility in its pathogenesis. Notably, genes like INF219’20and Rab7a ’13, 21, 22. 22, known to influence organelle fission / mobility / tumover, are mutated in CMT.

[0137] INF2 CMT patient fibroblasts display actin-dependent organelle immobilization and aberrant actin assembly. Live cell imaging of lysosomes in CMT patient and healthy control fibroblasts shows decreased lysosome mobility in CMT patients. This mobility defect is rescued by treatment with Latrunculin B, an actin depolymerizing drug. Additionally, INF2 CMT patient fibroblasts display a significant increase in actin aggregates compared to healthy controls. These actin aggregates often contain organelles, suggesting a link between aberrant actin assembly and organelle immobilization in INF2 CMT (FIG. 12).

[0138] Conclusion

[0139] This disclosure outlines the development of an innovative imaging and computational toolkit to study age-related changes in organelle dynamics, addressing a critical need in aging research. By combining label-free QPI, deep learning, and intelligent microscopy control, Applicant created a powerful and versatile platform for visualizing, quantifying, and understanding the complex relationship between organelle behavior and the aging process. The resulting toolkit empowered researchers to make transformative discoveries in aging biology and contribute to the development of interventions that promote healthier aging.-28-4919-3750-8870.1Atty Dkt.: 114198-0890

[0140] Example Embodiment

[0141] This disclosure relates to imaging and computational techniques for detecting transient biological events within live-cell environments. Cellular imaging is widely used to study the dynamic behavior of intracellular structures such as mitochondria, lysosomes, and other organelles. Various optical methods, including fluorescence imaging and quantitative phase imaging, can acquire sequential image data that represent changes in morphology, mobility, or interaction patterns of these subcellular structures. Computational pipelines can process imaging data to extract quantitative parameters of these biological processes. The ability to acquire high-speed image sequences and analyze them in real time can provide valuable insights into fundamental cell biology and disease mechanisms.

[0142] Existing imaging workflows can acquire large volumes of data but may require extensive processing to identify biologically relevant events. Many transient organelle phenomena occur on millisecond timescales and can be easily missed between acquisition intervals. Image segmentation methods based on static thresholds or predefined filters may lack adaptability to variations in illumination, focus, or morphology between samples. Traditional control software can operate imaging sequences in a predetermined fashion without considering the content of the captured images. As a result, experimental workflows may be unable to react to conditions of interest while they occur, limiting the ability to capture or preserve short-lived organelle events.

[0143] Although conventional microscopy instruments can record high-quality image sequences, the identification of transient biological phenomena within such data can remain technically challenging. Many events such as organelle fission, fusion, or vesicle rupture occur rapidly and intermittently, often within milliseconds. Processing large image sets can introduce delays, variability, and incomplete coverage of such transient events. Static threshold methods based on intensity change or displacement can fail under variations in focus, brightness, or intracellular crowding. As a result, conventional imaging pipelines may be unable to isolate specific temporal patterns or react in real time to the occurrence of critical subcellular events.

[0144] The techniques described herein can provide automated detection and response to transient organelle events through an integrated imaging and control system. The approach can combine real-time image acquisition with a learning-based analysis process that predicts the-29-4919-3750-8870.1Atty Dkt.: 114198-0890 occurrence of rapid biological phenomena based on temporal sequences of cellular images. The controller of the system can execute trained computer models that identify specific signatures of subcellular activity, such as organelle fusion or fragmentation, and can take immediate corrective or investigative actions upon detection. Such actions can include initiating a rapid solution exchange or fixation step within a connected fluidic environment to preserve the state of the observed organelles at the time of detection.

[0145] The implementation can provide a direct technical improvement over conventional passive imaging workflows. By coupling machine-learned event detection with low-latency microfluidic control, the techniques can capture transient subcellular phenomena that would otherwise remain unrecorded. The system can thereby reduce latency between detection and actuation. The integration of automated inference and environmental adjustment can generate datasets that represent biological processes more accurately in both spatial and temporal domains. These technical improvements can support quantitative mapping of organelle dynamics for a broad range of cellular imaging applications.

[0146] Referring now to FIG. 13, illustrated is a block diagram of an example microfluidic imaging system 1300 for automated detection of transient organelle events. The system 1300 can include a controller 1302, an image capture device 1304, a solution reservoir 1306, a microfluidic channel 1308, a chamber 1310, a cell 1312 located within the chamber 1310, and an organelle 1314 within the cell 1312. The system 1300 can include any number of components in any configuration.

[0147] The system 1300 can include a controller 1302. The controller 1302 can be or include one or more processors coupled to memory that execute machine-learning-based image analysis and real-time control of microfluidic operations. The controller 1302 can be the same as or may be implemented by the computing device 1500, shown and described with reference to FIGS. 15A and 15B. In some implementations, the controller 1302 can execute stored instruction sequences to process time-sequenced image data, generate feature representations for individual organelles, and evaluate temporal changes within those representations to distinguish transient organelle events from baseline dynamics. For example, the controller 1302 can execute a trained computational model that processes pixel-level intensity variations or phase-shift patterns in consecutive image frames to determine the presence or absence of a transient organelle event in a biological image sequence. In some implementations, the -30-4919-3750-8870.1Atty Dkt.: 114198-0890 controller 1302 can allocate computational resources among central processing units, memory subsystems, and peripheral interfaces of the computing device 1500 to maintain low-latency inference cycles and command issuance for subsequent actuation within the microfluidic system 1300.

[0148] The system 1300 can include an image capture device 1304. The image capture device 1304 can acquire time-sequenced image data representing biological material positioned in the chamber 1310 using optical, fluorescence, and / or quantitative phase imaging modalities, among others. In some implementations, the image capture device 1304 can include a highspeed camera coupled to a quantitative phase imaging microscope that facilitates detection of transient subcellular changes at submicron resolution. For example, the image capture device 1304 can incorporate an sCMOS or EMCCD detector integrated with a quasi-simultaneous QPI-fluorescence optical assembly that captures cell or organelle images at resolutions of 250 nanometers or less. In some implementations, the image capture device 1304 can generate synchronized frame sequences through a high-bandwidth interface such as Camera Link or 10 Gb Ethernet and transmit the corresponding data to the controller 1302 with latency below one frame interval. For example, the image capture device 1304 can output alternating phase and fluorescence images every 200 milliseconds during continuous observation of mitochondrial or lysosomal motion. The integration of the optical head, detector, and transmission interface in the image capture device 1304 can maintain temporal alignment between channels used for quantitative phase imaging and fluorescence validation throughout extended imaging sessions.

[0149] The image capture device 1304 can transmit individual frames or continuous sequences of frames to the controller 1302 for algorithmic evaluation. Each transmitted frame can represent an optical phase or fluorescence image depicting variations in refractive index or intensity within cellular structures over a defined time interval. In some implementations, the controller 1302 can temporally align successive frames and extract differential image features used to identify organelle activity changes. For example, the controller 1302 can detect pixellevel phase shifts corresponding to mitochondrial or lysosomal membrane deformation and quantify such displacements to assess transient organelle dynamics in the acquired dataset. In some implementations, the image capture device 1304 can further provide frame metadata specifying acquisition timestamps, illumination mode, or exposure conditions, enabling the-31-4919-3750-8870.1Atty Dkt.: 114198-0890 controller 1302 to correlate observed optical phase variations with recorded imaging parameters for accurate temporal association of detected events.

[0150] The image capture device 1304 can communicate with the controller 1302 through a high-speed data connection such as Peripheral Component Interconnect Express (PCIe) or Ethernet to maintain minimal frame delay between acquisition and analysis. In some implementations, the image capture device 1304 can stream each acquired frame directly into a controller-managed memory buffer that allows sequential inference processing within a single frame interval. For example, the controller 1302 can receive quantitative phase or fluorescence images through a PCIe interface operating at bandwidths exceeding 10 gigabits per second, thereby sustaining real-time transfer of full-resolution data for continuous analysis. In some implementations, the image capture device 1304 can include synchronization circuitry that aligns internal exposure clocks with those of the controller 1302 to achieve quasi-simultaneous acquisition of phase and fluorescence images within a temporal offset of a few milliseconds. For example, a trigger signal from the controller 1302 can initiate near-concurrent exposure cycles in separate sensor elements, enabling registration accuracy sufficient for overlaying quantitative phase and fluorescence channels during analysis of subcellular events.

[0151] The system 1300 can include a solution reservoir 1306. The solution reservoir 1306 can contain one or more liquids used to alter, fix, or chemically process biological materials observed within the chamber 1310. In some implementations, the solution reservoir 1306 can maintain separate compartments for buffer media, fixative solutions, or reagent formulations that are directed to the chamber 1308 through controlled micro-pumping elements. For example, one compartment of the solution reservoir 1306 can hold a phosphate-buffered saline used to stabilize cell morphology prior to reagent exchange, while another compartment can hold a paraformaldehyde-based fixative. Each compartment may connect to discrete delivery lines or valves that are actuated by the controller 1302 to achieve rapid and selective solution transfer to or from the chamber 1308.

[0152] In some implementations, the solution reservoir 1306 can operate under pressure or vacuum differentials applied by the controller 1302 to execute sub-second fluid exchanges. For example, the controller 1302 can generate a command pulse that activates a micro-pump downstream of the solution reservoir 1306, causing complete medium replacement in the -32-4919-3750-8870.1Atty Dkt.: 114198-0890 chamber 1308 within about one second after an event detection. The solution reservoir 1306 may include integrated pressure sensors or flow restrictors to maintain steady inflow rates compatible with high-magnification imaging. The physical structure of the solution reservoir 1306 may include chemically inert materials such as borosilicate glass, fluoropolymer tubing, or poly dimethyl siloxane (PDMS), maintaining compatibility with fixative agents and sustaining repeated fluid exchanges during extended imaging sessions.

[0153] The solution reservoir 1306 can execute rapid solution exchange operations within the microfluidic system 1300 under command of the controller 1302. In some implementations, the controller 1302 can transmit a digital control signal that actuates one or more microvalves or micropumps coupled to the fluidic line extending between the solution reservoir 1306 and the chamber 1308. For example, the controller 1302 can generate a pulse sequence that activates a pressure-driven actuator in the solution reservoir 1306, causing immediate displacement of fluid through the microfluidic channel 1308 into the chamber 1308. In some implementations, the controller 1302 can regulate the amplitude and duration of the control signal to achieve a specific exchange rate or fluid turnover ratio suitable for fixation or reagent delivery. For example, the control signal may drive a microperistaltic pump to deliver a reagent bolus that replaces the entire chamber volume within approximately one second of detecting a transient event. The coordination between the solution reservoir 1306 and the controller 1302 can maintain a continuous flow path during exchange while reducing turbulence in the chamber 1308 to preserve the imaging field and structural integrity of the observed organelles.

[0154] The solution reservoir 1306 can operate according to a combination of micro-pumps and electronically controlled flow valves that are connected to the microfluidic channel 1310 to deliver fluidic commands from the controller 1302. In some implementations, the solution reservoir 1306 can activate one or more piezoelectric or pneumatic pumps to generate a displacement pulse in the connected conduit and initiate high-speed fluid transfer toward the chamber 1308. For example, upon receiving a trigger signal from the controller 1302, the solution reservoir 1306 can emit a regulator-controlled pressure burst that displaces reagent fluid through the microfluidic channel 1310 until the original chamber fluid volume is fully exchanged within approximately 500 milliseconds after the command initiation. In some implementations, the flow valves coupled to the solution reservoir 1306 can operate in coordinated timing sequences that maintain a continuous flow path while minimizing fluidic-33-4919-3750-8870.1Atty Dkt.: 114198-0890 backpressure and mechanical vibration in the imaging region. For example, one valve may open fractionally before a corresponding valve closes to sustain laminar flow during rapid replacement cycles and prevent distortion in the optical field during high-magnification imaging.

[0155] The microfluidic channel 1310 can function as a fluid transport conduit that hydraulically couples the solution reservoir 1306 to the chamber 1308. In some implementations, the microfluidic channel 1310 can be fabricated as a micro-scale conduit with cross-sectional dimensions that maintain laminar flow while permitting sub-millisecond reagent or fixative delivery latency. For example, the microfluidic channel 1310 can include a photolithographically patterned flow path with a rectangular or serpentine geometry etched into a glass or polydimethylsiloxane substrate to minimize internal turbulence during liquid displacement. In some implementations, the microfluidic channel 1310 can include microscale inlet and outlet junctions that align precisely with the chamber 1308 to maintain optical stability during high-resolution imaging. The microfluidic channel 1310 can maintain continuous hydraulic connectivity with the solution reservoir 1306 to facilitate real-time exchange operations commanded by the controller 1302.

[0156] In some implementations, the microfluidic channel 1310 can include integrated elements that modulate fluid velocity, pressure, or direction during solution exchange. For example, the microfluidic channel 1310 can incorporate lithographically embedded microvalves or resistive flow segments formed from elastomeric layers that deform under applied pneumatic or piezoelectric actuation. The microfluidic channel 1310 can use these elements to achieve precise volumetric control over inflow and outflow, thereby maintaining stable flow rates during dynamic switching between media and reagents. In some implementations, the microfluidic channel 1310 can include backpressure regulators or bypass conduits that limit transient pressure fluctuations within the optical field of the chamber 1308. The geometry, wall thickness, and surface coating of the microfluidic channel 1310 can be selected to minimize adsorption of biomolecules and maintain chemical compatibility across repeated fluidic cycles.

[0157] The chamber 1308 can be a volumetric enclosure that provides a controlled environment for positioning biological samples subject to optical interrogation and fluid exchange. In some implementations, the chamber 1308 can maintain an optically transparent -34-4919-3750-8870.1Atty Dkt.: 114198-0890 structure formed from a glass substrate and a complementary sealing layer that collectively support high-resolution imaging modalities such as quantitative phase or fluorescence microscopy. For example, the chamber 1308 can define a microfluidic well fabricated on the substrate with a flat optical interface providing consistent refractive index characteristics across the imaging area. In some implementations, the chamber 1308 can include inlet and outlet junctions that align with the microfluidic channel 1310 to permit real-time introduction or removal of reagents without altering focal stability. For example, the chamber 1308 can incorporate micron-scale flow paths coupled to pneumatic control ports that generate laminar flow across cultured fibroblasts adhered to the chamber base. In some implementations, the chamber 1308 can further incorporate mechanical features such as patterned ridges or bonding regions that restrict sample drift during continuous imaging cycles, thereby maintaining positional stability necessary for sequential image acquisition and automated analysis. For example, the chamber 1308 can anchor one or more cells using surface coatings compatible with long-term live-cell imaging and repeated solution exchange events.

[0158] The chamber 1308 can receive solution inflow and outflow through the microfluidic channel 1310 during automated exchange operations. In some implementations, the chamber 1308 can alternate between different fluid compositions under programmatic commands generated by the controller 1302. For example, the chamber 1308 can sequentially circulate growth media, reagent formulations, or chemical fixatives through the microfluidic channel 1310 according to predefined trigger conditions identified by the controller 1302. In some implementations, each solution cycle can maintain controlled flow rates and temperature conditions to preserve imaging stability while permitting reproducible exchange sequences across consecutive experimental runs. For example, the controller 1302 can regulate valve openings or micro-pump pulse durations to establish laminar fluid motion within the chamber 1308, thereby maintaining a uniform reagent gradient during event-triggered switching between media, reagents, and / or fixatives, among others.

[0159] The chamber 1308 can provide an optically accessible environment that facilitates high-resolution imaging by transmitting light with minimal scattering and refraction. The optical interface of the chamber 1308 can be constructed from transparent substrates that maintain uniform refractive index properties across the imaging area to preserve phase stability during quantitative phase or fluorescence acquisition. In some implementations, the chamber-35-4919-3750-8870.1Atty Dkt.: 114198-08901308 can include an optically polished coverslip base joined to a rigid ceiling layer to reduce thermal and mechanical drift during extended imaging sessions. For example, the chamber 1308 can incorporate borosilicate glass or fused silica elements bonded with optically clear adhesives that maintain index matching across wavelengths used for transmitted or reflected illumination. The selection of materials and bond geometries can maintain high numerical aperture objectives in close proximity to the sample without inducing spherical aberration or focal distortion during repeated solution exchanges.

[0160] In some implementations, the chamber 1308 can include a high-index coverslip base paired with an optically neutral ceiling layer that equalizes refraction through the imaging path. For example, the ceiling layer can be fabricated from a polydimethylsiloxane (PDMS) membrane with an optical index selected to match the base substrate, thereby minimizing phase delay gradients during live-cell quantitative phase imaging. Additional mechanical stabilization can be achieved through reinforcement frames or peripheral clamps that maintain planar alignment between the base and ceiling layers while accommodating microfluidic ports coupled to the channel 1310. Such structural symmetry can retain consistent light path length and lateral resolution across the entire imaging field, enabling repeatable capture of organelle dynamics under varying microfluidic flow conditions.

[0161] The chamber 1308 can maintain or store fluid 1311 that circulates through the microfluidic channel 1310 during imaging or solution exchange operations. In some implementations, the fluid 1311 can be a biologically compatible medium formulated to sustain live cells such as fibroblasts, neurons, or epithelial cells under controlled culture conditions. For example, the fluid 1311 can include buffers, nutrients, and gas equilibrating agents that maintain physiological osmolarity and pH values suitable for quantitative phase or fluorescence imaging. In some implementations, the fluid 1311 can be replaced intermittently or continuously through programmable flow pulses initiated by the controller 1302 to introduce reagents or fixatives from the solution reservoir 1306. For example, the controller 1302 can initiate a synchronized inflow through the microfluidic channel 1310 to refresh the chamber 1308 volume at defined time intervals without disturbing the imaging field. The physical properties of the fluid 1311 such as viscosity, refractive index, and refractive homogeneity can be selected to minimize optical aberration and shear stress on the cell 1312 during extended microscopic observation.-36-4919-3750-8870.1Atty Dkt.: 114198-0890

[0162] The chamber 1308 can contain one or more cells 1312 suspended within the fluid 1311. The fluid 1311 can include single cells, multicellular clusters, or other particulate matter relevant to microscopic analysis. In some implementations, the fluid 1311 can include biological particulates such as vesicles, exosomes, or lipid droplets, and / or non-biological tracer particles used to calibrate optical measurements. For example, the fluid 1311 can transport patient-derived fibroblasts, neuronal spheroids, or synthetic calibration beads through the imaging field for comparative observation. The inclusion of such diverse cellular and particulate elements in the fluid 1311 can enable real-time optical detection of morphology, refractive index variation, and movement behavior across multiple sample types during continuous imaging operations.

[0163] The cell 1312 can exhibit transient organelle events that are monitored through the image capture device 1304 and evaluated by the controller 1302. In some implementations, the image capture device 1304 can capture consecutive image frames of cellular processes that include alterations to subcellular structures, and the controller 1302 can process the frames to detect signatures of dynamic events. For example, the controller 1302 can determine temporal displacements or frequency-domain variations in pixel intensities corresponding to morphological changes associated with mitochondrial fission, lysosomal fusion, or other transient transitions of intracellular mobility. In some implementations, the controller 1302 can apply learned temporal patterns within a trained deep learning network to classify each detected morphological shift as a probable fission, fusion, or repositioning event. For example, the controller 1302 can evaluate sequences of phase contrast or fluorescence data that reveal fragmentation of mitochondrial networks or directional displacement of lysosomal vesicles along cytoskeletal tracks, confirming that the event corresponds to organelle mobility transitions rather than steady-state drift.

[0164] The cell 1312 can adhere to the interior surface of the chamber 1308 to maintain positional stability during long-term live-cell imaging operations controlled by the system 1300. In some implementations, the controller 1302 can regulate optical intensity and illumination timing to maintain low-phototoxic conditions suitable for continuous quantitative phase imaging. For example, the image capture device 1304 can operate at frame rates between one and five frames per second using low-power light-emitting sources that limit energy exposure to less than five percent of the photobleaching threshold for fluorescently labeled-37-4919-3750-8870.1Atty Dkt.: 114198-0890 proteins. In some implementations, the chamber 1308 can include a temperature-controlled substrate or gas-permeable layer that facilitates exchange of oxygen and carbon dioxide within the fluid 1311 to sustain fibroblast viability for up to twenty-four hours of uninterrupted imaging. For example, the controller 1302 can maintain the thermal environment of the chamber 1308 at approximately 37 degrees Celsius and adjust fluidic exchange intervals to preserve metabolic activity and minimize stress-induced morphological changes during photothermal exposure.

[0165] The cell 1312 can include an organelle 1314 that performs subcellular functions detectable through optical imaging. The organelle 1314 can be a mitochondrion, lysosome, endosome, or other intracellular compartment whose morphology and displacement within the cytoplasm can be monitored for temporal changes. In some implementations, the organelle 1314 can undergo structural transformations such as elongation, constriction, or vesicular fusion that are observable as transient contrast variations within quantitative phase or fluorescence image sequences. For example, the organelle 1314 can represent one of several mitochondria within primary fibroblast samples that exhibit measurable fragmentation events under Charcot-Marie-Tooth disease conditions, where morphological discontinuities along mitochondrial filaments correspond to fission signatures detected by the imaging system. In some implementations, the organelle 1314 can maintain fluorescence markers or intrinsic refractive index differences that enable the controller 1302 to derive shape descriptors and motion vectors from the acquired frames for downstream analysis of organelle dynamics and transient event classification.

[0166] The organelle 1314 can be the primary feature evaluated by the controller 1302 during automated transient event detection using image data received from the image capture device 1304. In some implementations, the controller 1302 can execute a trained inference model to evaluate pixel-level variations associated with the organelle 1314 and identify morphological transitions indicative of transient fusion or fission activity. For example, the controller 1302 can monitor frame-to-frame intensity and contour fluctuations corresponding to a lysosomal fusion event and generate a control signal to actuate the solution reservoir 1306. In response, the solution reservoir 1306 can initiate an immediate fluid exchange to introduce a fixative or reagent into the imaging chamber within a predefined latency interval to preserve the morphological state of the organelle 1314 during capture.-38-4919-3750-8870.1Atty Dkt.: 114198-0890

[0167] The organelle 1314 can exhibit measurable optical phase or fluorescence variations that permit the controller 1302 to computationally extract morphological and motion descriptors from acquired images. In some implementations, the controller 1302 can generate spatial intensity maps or phase shift matrices corresponding to the organelle 1314 across sequential frames to quantify shape changes or refractive index variations within each imaging interval. For example, the controller 1302 can determine temporal displacement vectors for each organelle 1314 across an image sequence and construct a trajectory map that represents velocity, directionality, and positional continuity of motion for subsequent classification of transient events. In some implementations, the trajectory map can include frame-indexed coordinate pairs that define motion pathways, and the controller 1302 can correlate these motion descriptors with known morphological signatures of fission, fusion, or translocation detected in prior training datasets.

[0168] The controller 1302 can determine event occurrences based on measurable differences between consecutive image sequences and can generate a control signal that initiates immediate solution exchange within the microfluidic system 1300. The controller 1302 can detect frame-to-frame variations in intensity distributions or phase-shift maps corresponding to cellular morphological changes and can classify those variations using a trained machine-learning model. In some implementations, the controller 1302 can identify sequential deviations that exceed a probabilistic threshold associated with transient organelle events such as mitochondrial fission, lysosomal fusion, or vesicular rupture, among others. For example, the controller 1302 can evaluate normalized cross-correlation coefficients derived from consecutive image matrices, determine that the computed deviation surpasses a statistically defined transient event threshold, and issue a digital trigger command to an actuator coupled to the solution reservoir 1306.

[0169] In some implementations, the controller 1302 can interpret rapid morphological transitions of mitochondria or lysosomes as indicators of transient events and can transmit a trigger signal through an electrical or optical interface connected to the solution reservoir 1306. For example, the controller 1302 can detect a fragmentation pattern in a mitochondrial network by measuring localized curvature changes and perimeter discontinuities within successive images, classify the event as mitochondrial fission, and command the solution reservoir 1306 to initiate introduction of a fixative into the chamber 1308 within one second. In another-39-4919-3750-8870.1Atty Dkt.: 114198-0890 example, the controller 1302 can register a transient fusion of lysosomes by identifying an abrupt increase in overlapped fluorescence regions and can actuate microvalves to deliver a reagent pulse that stabilizes the detected region for subsequent imaging cycles. The automated correlation between event classification by the controller 1302 and execution of microfluidic exchange operations can maintain temporal precision required for live-cell preservation and downstream structural analysis.

[0170] The controller 1302 can transmit command signals through an electrical interface to valves and pumps of the microfluidic system 1300 to implement low-latency control responses. In some implementations, the controller 1302 can operate one or more digital-to-analog converters or transistor-driven switches that directly energize microvalve actuators in response to event-detection commands. For example, the controller 1302 can output a pulse-width- modulated signal to a solenoid valve that displaces reagent fluid through the channel 1308. The electrical interface can include conductors and connectors suitable for high-frequency signal transfer, such as shielded cable assemblies or printed-circuit traces linking the controller 1302 to embedded driver circuits. In some implementations, the electrical interface can maintain a synchronous timing bus that correlates image acquisition timestamps with actuation events to preserve sub-second alignment between analysis output and fluidic adjustments within the microfluidic system 1300.

[0171] In some implementations, the controller 1302 can generate digital trigger signals within about 500 milliseconds of identifying a transient event during real-time inference. For example, upon detecting a mitochondrial fission or lysosomal fusion event, the controller 1302 can assert a control line high for a defined pulse duration that actuates a micro-peristaltic pump to initiate rapid solution inflow through the channel 1308. The electrical signal can propagate through a low-impedance path that minimizes propagation delay and enables nearly instantaneous fluid displacement in the chamber. In some implementations, the control parameters of pulse width, amplitude, and repetition rate can be selected to displace a specific fluid volume or maintain laminar flow during reagent delivery. The combined operation of the controller 1302 and fluidic actuation hardware can maintain biochemical stability of the sample while enabling reproducible fluid-exchange reactions for subsequent imaging and event validation.-40-4919-3750-8870.1Atty Dkt.: 114198-0890

[0172] Referring now to FIG. 14, illustrated is a flowchart of an example method 1400 for detecting organelle transient events and conditionally triggering solution exchange in a microfluidic system. The method 1400 can be executed, performed, or otherwise carried out by a data processing system such as any of the computing systems or devices described herein (e.g., the controller 1302, shown and described with reference to FIG. 13). In brief overview of the method 1400, the method 1400 can include receiving a sequence of images of the one or more organelles in a chamber of a microfluidic device (1402), determining whether a transient organelle event is detected (1404), determining not to adjust operation of the microfluidic device if no transient event is detected (1406), and automatically triggering a solution exchange within the chamber if the transient event is detected (1408). The method 1400 can include any number of operations, and the operations can be performed in any order. The method 1400 can provide real-time, automated detection and response to transient organelle events, thereby minimizing latency between event identification and solution exchange to preserve dynamic cellular states with sub-second temporal precision.

[0173] At operation 1402, the data processing system can receive a sequence of images (e.g., a first sequence of images) of one or more organelles located in a chamber of a microfluidic device. The data processing system can receive the image sequence directly from an image capture device that is oriented toward a face (e.g., an optically transparent face) of the chamber. The image capture device can acquire the frames through a quantitative phase imaging or fluorescence imaging process. In some implementations, the data processing system can receive the image sequence during continuous observation of living fibroblasts or other cell types maintained within a chamber. For example, the data processing system can receive consecutive frames transmitted through a high-speed interface such as Peripheral Component Interconnect Express (PCIe) or 10 gigabit Ethernet immediately after each frame capture by the image capture device. In some implementations, each received sequence can include multiple consecutive frames representing organelle motion, fragmentation, fusion, or other rapid dynamic transitions. For example, the data processing system can receive image sequences acquired at intervals that permit near real-time inference cycles while maintaining uninterrupted imaging throughput and temporal alignment across sequential acquisitions.

[0174] At operation 1404, the data processing system can determine whether a transient organelle event occurs in the newly received sequence of images. The data processing system-41-4919-3750-8870.1Atty Dkt.: 114198-0890 can retrieve each frame immediately after acquisition from the image capture device oriented toward a chamber of a microfluidic device. The data processing system can perform the analysis using sequential evaluations of quantitative phase or fluorescence frames representing organelle morphology and motion patterns over time. The data processing system can organize each image set into a temporal sequence and define computational boundaries around every observable organelle before performing transient event inference. The operation can occur continuously to maintain real-time responsiveness during monitoring of live cells under dynamic flow conditions.

[0175] In some implementations, the data processing system can execute a trained machine learning model that receives a plurality of consecutive images and generates an event determination score. The model can include a convolutional backbone for spatial feature extraction and a temporal attention layer for identifying frame-to-frame changes. For example, the data processing system can apply convolutional operations to evaluate local membrane curvature, aspect ratio variation, or texture contrast that correlates with mitochondrial fragmentation or lysosomal fusion. The model output can produce a numerical confidence value that quantifies whether the observed temporal variation corresponds to a transient organelle event or baseline fluctuation. The data processing system can treat the inference score exceeding a defined threshold as an indication of event occurrence.

[0176] The data processing system can evaluate each received image sequence in real time, immediately after the final frame of that sequence is transmitted from the image capture device. In some implementations, parallel data buffers can allow the data processing system to infer one sequence while acquiring the next, eliminating idle intervals between detection cycles. For example, the data processing system can allocate one processing thread to perform feature extraction while another thread receives image data through a high-bandwidth network interface. This configuration can maintain sub-second latency between image acquisition and event determination. Continuous inference can preserve sensitivity to rapid morphological transitions such as fission or fusion events that last only a few frames.

[0177] In some implementations, the data processing system can perform temporal differencing by computing the cumulative variation in pixel-level or phase values across consecutive frames. The data processing system can transform each temporal difference map into a numerical vector representing displacement magnitude, contour curvature, or optical-42-4919-3750-8870.1Atty Dkt.: 114198-0890 density change. For example, the data processing system can compare the vector representations of successive frames to identify abrupt shape bifurcations consistent with mitochondrial fission. The resulting temporal descriptor can serve as an input to the trained network for classification of the event type. The same process can apply to other organelles that exhibit dynamic morphological transformations within the imaging field.

[0178] In one example, the data processing system can execute the machine learning model to determine that the sequence of images does not indicate any transient events in organelles by quantifying the mean distance traveled by each organelle in the chamber using image analysis. The data processing system can apply frame-to-frame optical flow measurements to compute displacement vectors representing organelle movement across consecutive images. In some implementations, the data processing system can calculate trajectory paths for each organelle using centroid detection and spatial correlation algorithms applied to sequential frames. For example, the data processing system can measure the displacement of mitochondria over a two- second interval and compute the mean distance traveled based on pixel-to-micrometer calibration values established during system alignment. The calculated mean distances can be stored in a temporary data array for immediate comparison against reference statistical values corresponding to known patterns of stable or dynamic organelle behavior. In some implementations, the data processing system can exclude non-organellar particulates, such as imaging artifacts or background fluctuations, by applying threshold segmentation and shape filters prior to quantification.

[0179] The data processing system can compare the quantified distances with a reference dataset to identify abnormal organelle mobility. The reference dataset can include historical measurements of organelle displacement ranges obtained from previously imaged cells under normal physiological conditions. In some implementations, the data processing system can determine that the current sequence indicates absence of transient events when the computed distance distribution for all detected organelles remains within standard deviation limits defined by the reference dataset. For example, the data processing system can evaluate mitochondrial motility data gathered from healthy fibroblasts, compare the measured mean distances of test samples, and determine that no transient event has occurred when the observed mean distance variation falls below a defined movement threshold, such as five percent deviation from baseline. The comparison can be executed using a kernel-based similarity-43-4919-3750-8870.1Atty Dkt.: 114198-0890 function or z-score normalization to verify statistical equivalence between measured and reference distributions. The results can be used immediately by the machine learning model to confirm non-occurrence of events before acquiring the next image sequence.

[0180] In an example, the data processing system can execute a trained classifier that receives time-sequenced image frames depicting actin structures and organelles within the chamber. The classifier can apply feature extraction operations to identify regions of elevated filament density or fluorescence intensity indicative of actin aggregates. In some implementations, the classifier can incorporate convolutional filters that measure curvature, cross-linking, or filament thickness across spatial coordinates. For example, the data processing system can apply layer outputs from the classifier to generate probability maps that delineate filament- dense regions overlapping with organelle boundaries, thereby establishing co-localization between actin assemblies and organelles suspended within the imaging field. The classifier output can include a numerical confidence value that quantifies the likelihood of immobilization, which the data processing system can compare to a detection threshold to determine whether the observed patterns represent abnormal aggregate formation. The resulting classification can be used as a trigger condition for subsequent fluidic actuation operations within the microfluidic environment.

[0181] In some cases, the data processing system can segment the organelles for classification. For example, the data processing system can generate per-instance masks for individual organelles by executing a deep learning model trained to segment and isolate subcellular structures in time-sequenced images. The data processing system can apply convolutional feature extraction layers to detect spatial boundaries corresponding to each distinct organelle and output pixel-level segmentation masks that delineate morphological contours in each frame. In some implementations, the data processing system can refine these boundaries through iterative filtering to remove artifacts and merge adjacent pixels associated with the same subcellular compartment. For example, the data processing system can compare edge continuity and texture homogeneity within adjacent regions to determine whether fragmented regions correspond to a single organelle. After generating the segmentation masks, the data processing system can classify each identified instance as mitochondria, lysosome, or endosome using a trained classification network that evaluates normalized intensity, shape descriptors, and motion vectors derived from the segmented regions. In some implementations,-44-4919-3750-8870.1Atty Dkt.: 114198-0890 the data processing system can assign a probabilistic label to each instance based on model output confidence values and verify that no class-specific transient event signatures are present by comparing inferred morphological patterns against baseline representations stored in reference datasets. For example, the data processing system can confirm absence of classspecific signatures when measured contour deformation, signal overlap, and aspect ratio variation remain within predefined tolerance intervals for stable organelle states.

[0182] In some cases, the data processing system can track organelle movement for detecting transient events. For example, the data processing system can track organelle movement by generating time-linked trajectories for each segmented organelle across the first sequence of images to characterize positional changes over consecutive frames. The data processing system can compute coordinate vectors that represent the centroid location of each organelle at each frame, and can aggregate the coordinate vectors into continuous trajectories corresponding to motion paths over time. In some implementations, the data processing system can refine the trajectories by applying motion-smoothing filters or interpolation between frames to correct for subpixel displacement errors in high-speed image sequences. For example, the data processing system can apply optical flow estimation or nearest-neighbor frame correspondence to maintain continuity in trajectory mapping even when transient occlusions occur within the imaging field. The data processing system can align the time-linked trajectories using acquisition timestamps to ensure that the motion profiles are temporally synchronized for downstream analysis. In some implementations, the data processing system can apply temporal pattern analysis to the generated trajectories to detect characteristic fission, fusion, or lysis signatures associated with dynamic morphological transitions. The data processing system can compute temporal derivatives of shape descriptors such as curvature and aspect ratio, and can compare the resulting temporal features against reference event templates generated during model training. For example, the data processing system can identify a fragmentation event when a single trajectory bifurcates into two distinct motion paths with maintained intensity continuity, or a fusion event when two previously separate trajectories converge into one. The data processing system can conclude non-occurrence of a transient event when no temporal pattern exceeds a detection threshold established for event probability, and can record the analyzed trajectory data for subsequent evaluation during continuous monitoring cycles.-45-4919-3750-8870.1Atty Dkt.: 114198-0890

[0183] In some implementations, the data processing system can distinguish between transient events in different organelle types. The data processing system can do so, for example, by quantitatively differentiating event likelihoods across multiple organelle classes including mitochondria, lysosomes, and endoplasmic reticulum. The data processing system can execute a trained inference model that generates probability distributions for each class based on temporal and morphological descriptors extracted from sequential image frames. In some implementations, the data processing system can compute likelihood values for class-specific event types by comparing temporal intensity gradients, contour variations, or motion trajectories observed in the acquired images. For example, the data processing system can identify a higher transient event probability for a mitochondrion relative to a lysosome when the measured curvature deviation and fragmentation index exceed predefined statistical thresholds associated with mitochondrial fission. In some implementations, the data processing system can assign a confidence score to each class determination by normalizing the aggregated model output probabilities over all evaluated organelles, thereby quantifying the relative certainty of event classification. For example, the data processing system can compute a normalized confidence score expressed as a probability ratio between the top two predicted organelle classes to weight the final detection outcome. These operations improve the monitoring process of individuals by providing statistically validated, organelle-specific assessments that enable more precise interpretation of cellular dynamics across age and disease conditions.

[0184] In some implementation, the data processing system can reduce the impact of phototoxic effects and / or photobleaching indicators that may impact the quality of images that are input into the machine learning model. For example, the data processing system can monitor imaging conditions for phototoxicity or photobleaching during image acquisition by continuously evaluating frame intensity stability, spectral emission decay, and background signal variation across sequential image sets. In some implementations, the data processing system can compare each newly acquired frame against predefined baseline illumination profiles to identify deviations indicative of excessive excitation energy or cumulative fluorescence loss. For example, the data processing system can measure the rate of fluorescence intensity reduction over time, determine that the slope of the reduction exceeds a pre-calibrated photobleaching threshold, and interpret the change as an indication of phototoxic-46-4919-3750-8870.1Atty Dkt.: 114198-0890 stress. In some implementations, the data processing system can adapt imaging parameters in real time by modifying excitation power, illumination duration, or frame acquisition intervals to maintain signal quality and optical phase accuracy without exceeding safe exposure levels. For example, the data processing system can automatically reduce light-emitting source intensity when a cumulative exposure index surpasses the target limit, or adjust the duty cycle of the illumination source to reduce radiation dose while preserving quantitative phase measurement precision. These operations improve the monitoring process of individuals by maintaining live-cell viability, minimizing optical damage, and ensuring consistent analytical fidelity across extended imaging sessions.

[0185] Responsive to determining the sequence of images did not depict a transient event, at operation 1406, the data processing system can determine not to adjust operation of the microfluidic device. For example, the data processing system can execute an evaluation cycle to determine whether adjustment of the microfluidic device is required after analyzing the image sequence. The data processing system can determine that no modification of the device operation is necessary when organelle dynamics remain within nominal mobility ranges or when no morphological indicator of fission, fusion, or lysis exceeds a defined detection threshold. In some implementations, the data processing system can compare detected displacement vectors and contour variations of each organelle to baseline statistical models derived from reference datasets. For example, the data processing system can determine that an aged fibroblast sample exhibiting reduced mitochondrial displacement represents stable dynamics and therefore maintain constant pressure and flow conditions in the microfluidic channel.

[0186] The data processing system can perform this determination immediately upon completion of each inference cycle before processing a subsequent sequence of images. In some implementations, the data processing system can reset transient variables that store intermediate confidence scores and initialize a new inference cycle for the following image sequence. For example, the data processing system can discard non-significant temporal variation values from the previous analysis and allocate processing resources to receive the next image set without altering active pump signals or valve states. This continuous evaluation procedure can maintain uninterrupted flow stability within the chamber while providing sequential image data to detect future transient organelle events.-47-4919-3750-8870.1Atty Dkt.: 114198-0890

[0187] The data processing system can repeat operations 1402 through 1406 in a continuous monitoring loop until determining that a sequence of images, such as a second sequence of images, indicates a transient event in the one or more organelles. In some implementations, the data processing system can initialize a recursive instruction cycle that sequentially acquires, evaluates, and classifies each consecutive image sequence until an event threshold is satisfied. For example, the data processing system can repeatedly receive image sequences from the image capture device, execute the machine learning model to compute event likelihood scores for each sequence, and reinitiate image acquisition for subsequent sequences when all evaluated scores fall below the predefined transient-event threshold. In some implementations, the data processing system can maintain queued image frames in volatile memory and discard previously analyzed frames at the start of each monitoring cycle to sustain real-time inference throughput. For example, the data processing system can allocate alternating memory buffers for active and standby image sequences, allowing uninterrupted transition between successive rounds of operations 1402 through 1406 until detection of a sequence representing a bona fide transient organelle event.

[0188] The data processing system can repeat the event-detection process for every continuous image sequence before beginning the subsequent capture cycle. Each inference cycle can initialize automatically after transfer of a new sequence of frames, guaranteeing uninterrupted analysis during experimental observation. In some implementations, the data processing system can archive inference results in transient memory for synchronization with the timing of microfluidic operations. For example, following detection of a positive event, the data processing system can immediately generate a control command for solution exchange. The iterative structure of successive inference cycles can maintain continuous feedback between image acquisition, analysis, and microfluidic control.

[0189] Responsive to determining a sequence of images indicate a transient event in the one or more organelles, at operation 1408, the data processing system can trigger (e.g., automatically trigger) a solution exchange within the chamber. To do so, the data processing system can transmit an electronic command through a control interface to microfluidic valve and pump systems coupled to the chamber. The control interface can include a signal driver circuit that converts digital command pulses into actuation voltages suitable for electropneumatic or piezoelectric microfluidic actuators. In some implementations, the data-48-4919-3750-8870.1Atty Dkt.: 114198-0890 processing system can schedule command issuance within a time interval measured in milliseconds after event recognition. For example, the data processing system can generate a control pulse propagating through a low-latency communication bus to trigger immediate displacement of fluid from the solution reservoir to the chamber.

[0190] In some implementations, the data processing system can generate a series of synchronized control signals that maintain consistent fluid flow and optical stability during exchange operations. The data processing system can regulate pulse width, amplitude, and duty cycle of the output signal according to pre-calibrated flow parameters. For example, the data processing system can select a control sequence that induces a pump stroke volume matching the chamber capacity while maintaining laminar flow conditions to prevent sample displacement. The generated control parameters can correspond to the physical characteristics of the microfluidic actuators, such as diaphragm stiffness or channel hydraulic resistance, enabling consistent reagent delivery or fixation within the defined sub-second target latency.

[0191] In some implementations, the data processing system can initiate a correlative light and electronic microscopy workflow responsive to detecting a transient event. For example, the data processing system can trigger chemical fixation of the organelles within one second of detecting a transient event to preserve subcellular structures for subsequent imaging. The data processing system can generate a control signal that activates microvalves and pump elements within the microfluidic network to deliver a fixative reagent at a pre-calibrated flow rate sufficient to achieve complete chamber exchange in less than one second. In some implementations, the fixative delivery process can maintain laminar flow conditions to avoid displacing or deforming cells during the initial preservation stage. The data processing system can further initiate a correlative light-and-electron microscopy workflow for the fixed sample by coordinating downstream imaging preparation steps, including the controlled retrieval of the fixed cells, resin infiltration, and positional mapping of optical regions of interest for electron microscopy alignment. In some implementations, the data processing system can synchronize image acquisition parameters between light and electron modalities, enabling reconstruction of subcellular ultrastructure corresponding to regions previously identified as transient event sites. These operations improve the monitoring process of individuals by enabling high-precision correlation between real-time dynamic events and ultrastructural-49-4919-3750-8870.1Atty Dkt.: 114198-0890 outcomes across cellular populations observed under different physiological or disease conditions.

[0192] In some implementations, the data processing system can coordinate actuation timing with the completion of image acquisition cycles generated by the image capture device. The data processing system can transmit a trigger acknowledgment back to the image capture device to pause illumination while the fluidic medium is replaced within the chamber. For example, the data processing system can initiate a one-second fixation pulse that introduces a defined chemical reagent into the chamber to preserve mitochondrial or lysosomal morphology without disrupting cell position. Upon completion of the exchange, the data processing system can resume image capture in the subsequent frame interval to record post-trigger cellular responses and correlate them with the detected transient event.

[0193] In an example, the data processing system can execute a trained classifier to identify actin aggregates containing organelles in image sequences acquired from the chamber. The classifier can receive pixel matrices representing cytoskeletal regions and generate a probability map that distinguishes aggregated actin filaments from diffuse actin networks. In some implementations, the classifier can include convolutional layers that extract morphological features such as filament curvature, thickness, and intensity gradients, generating a classification mask that delineates areas of dense actin assembly. For example, the data processing system can analyze fluorescence images of organelles labeled with a membrane marker and actin labeled with a cytoskeletal probe, determining co-localization zones indicative of organelle entrapment within actin aggregates. The classification process can identify immobilized organelles based on spatial overlap between the actin and organelle masks, assigning an immobilization confidence score to quantify the likelihood of restricted motion due to abnormal filament assembly.

[0194] Responsive to a determination of immobilization, the data processing system can automatically trigger the introduction of an actin depolymerizing agent into the chamber. The system can generate a control signal that activates a microvalve or pump to deliver a measured volume of depolymerizing reagent, such as latrunculin B or cytochalasin D, within a defined time window following detection. In some implementations, the control parameters can specify injection rate, pressure, and reagent concentration to facilitate rapid dilution without disturbing optical focus or sample stability. For example, the data processing system can inject a reagent-50-4919-3750-8870.1Atty Dkt.: 114198-0890 pulse that replaces the chamber medium with a latrunculin B-containing solution capable of disrupting filamentous actin, followed by continued imaging to verify restoration of organelle mobility. Real-time feedback from subsequent image sequences can confirm a reduction in actin aggregate density and increased displacement of previously immobilized organelles, demonstrating a successful solution exchange initiated by automated event detection.

[0195] In some cases, the data processing system can monitor organelles in the chamber over time. For example, the data processing system can detect age-related organelle morphological features, including fragmentation and enlargement, by executing a trained deep learning model that classifies structural changes in sequential image data. In some implementations, the data processing system can generate pixel-level segmentation masks that isolate organelle boundaries and compute shape descriptors corresponding to curvature, aspect ratio, and surface area to quantify morphological variations. For example, the data processing system can compare the derived descriptors with reference templates associated with known agedependent morphologies to identify organelles exhibiting fragmentation or enlargement within each imaging sequence. The data processing system can record the count and spatial frequency of detected morphological features across multiple sequences of images of the chamber and store statistical aggregates that represent the distribution of feature occurrence over repeated acquisitions.

[0196] In some implementations, the data processing system can automatically quantify interorganelle interactions by evaluating spatial and temporal correlations among segmented organelle representations extracted from sequential image frames. The data processing system can generate pixel-level spatial overlap matrices that identify instances of colocalization between organelles within each frame and can aggregate frame-by-frame results to form a temporal association map representing the persistence and frequency of direct organelle contact. In some implementations, the data processing system can apply threshold-based filtering to exclude incidental proximity measurements and retain only statistically significant overlaps that indicate sustained interactions. For example, the data processing system can compare centroid distances and shared area ratios among organelle masks to classify an event as colocalized when overlap exceeds a normalized fraction of the smaller organelle’s area for a predefined number of consecutive frames. In some implementations, the data processing system can further compute interaction duration, recurrence intervals, and co-movement-51-4919-3750-8870.1Atty Dkt.: 114198-0890 trajectories to characterize the dynamics of association between organelle types such as mitochondria, lysosomes, and endosomes, among others. The data processing system can then generate a session report that summarizes interaction rates, distributions of spatial overlap, and quantitative parameters describing each class of inter-organelle contact over the total observation period. The automated quantification and reporting provide technical advantages by enabling reproducible, high-throughput assessment of organelle relationships across time without manual frame inspection, thereby increasing temporal accuracy and consistency in evaluating subcellular dynamics.

[0197] In some implementations, the data processing system can use the systems and methods described herein to monitor individuals by specific characteristics. For example, the data processing system can compare event frequencies and mobility profiles identified in multiple datasets obtained from separate experimental groups categorized by donor age or disease state. In some implementations, the data processing system can align image-derived quantitative parameters such as mean organelle displacement, fission-to-fusion ratio, or colocalization index across selected sample cohorts to establish standardized comparative features. For example, the data processing system can apply statistical normalization to event count distributions from young, middle-aged, and aged donor fibroblast populations or from patient- derived and control cell samples, generating pairwise or matrix-form comparisons that quantify relative differences in organelle dynamics. In some implementations, the data processing system can store the resulting comparative analytics and corresponding cohort labels within a database to retain longitudinal evaluation records across multiple imaging sessions. For example, each entry can associate computed mobility metrics and event frequency values with metadata for donor classification, sample identification, and experimental conditions, allowing future retrieval and cross-temporal analysis. These operations improve monitoring of individuals with specific characteristics by enabling systematic detection of progressive changes in organelle behavior across distinct age groups or disease-associated sample cohorts.

[0198] The automatic solution exchange provides significant technical advantages by enabling real-time, closed-loop intervention in response to detected organelle events. This automation ensures that chemical or environmental change, such as the introduction of reagents, fixatives, or therapeutic agents, are precisely timed to cellular events, minimizing latency. As a result, the system can capture transient or rare biological phenomena with high fidelity, improve-52-4919-3750-8870.1Atty Dkt.: 114198-0890 experimental reproducibility, and reduce phototoxicity or sample perturbation by limiting unnecessary exposure. Additionally, the integration of solution exchange with machine learning-driven event detection allows for adaptive experimental protocols, supporting high- throughput screening and advanced studies of dynamic cellular processes.

[0199] The data processing system can continue operation after a solution exchange by receiving a third sequence of images representing the same organelles previously analyzed in the first and second sequences. The data processing system can align this third sequence in time and spatial orientation with pre-exchange image data stored in a transient buffer to permit direct comparison of morphological and kinetic descriptors. In some implementations, the data processing system can use the same acquisition parameters applied during the second sequence, such as illumination intensity, focal depth, and acquisition frame rate, to generate a dataset that accurately reflects variations in organelle structure or position caused by fluid replacement. For example, the image capture device can capture quantitative phase or fluorescence frames at identical exposure durations to those used before solution exchange, enabling the data processing system to minimize photometric bias while assessing post-actuation morphological changes. The data processing system can organize the third sequence into frame sets corresponding to defined temporal intervals within the post-exchange period and can extract numerical descriptors such as area, aspect ratio, centroid displacement, and refractive index deviation for each individual organelle identified in prior sequences.

[0200] The data processing system can execute the machine learning model using the third sequence of images to determine whether the induced solution exchange altered organelle dynamics within the chamber. In some implementations, the machine learning model can compare vector fields derived from organelle trajectories before and after the exchange to compute differential displacement magnitudes indicative of motility changes. For example, when a fixative reagent introduced through the solution reservoir decreases mitochondrial movement within a cell, the data processing system can calculate a reduction in mean track velocity between the second and third sequences and classify the behavior as a successful immobilization response. In other implementations, the data processing system can apply convolutional filters to successive post-exchange frames to evaluate whether organelle boundaries remain stable, elongate, or fragment relative to pre-exchange states. The data processing system can output a determination indicating altered or unaltered dynamic behavior-53-4919-3750-8870.1Atty Dkt.: 114198-0890 for each organelle, which provides quantitative confirmation that the solution exchange event initiated by earlier inference cycles produced measurable biological or optical effects in the microfluidic system.

[0201] An example implementation of the microfluidic system can include a chamber in fluid communication with a solution reservoir through a microfluidic channel. The chamber can hold a suspension of fibroblast-derived cells containing organelles such as mitochondria, lysosomes, or endosomes, among others. The image capture device can be positioned opposite a transparent face of the chamber to acquire time-sequenced quantitative phase or fluorescence images through an objective lens coupled to a high-speed sensor. The data processing system, acting as the controller, can communicate with the image capture device through a low-latency data interface to receive continuous frame sequences. In some implementations, the data processing system can generate a first image sequence of images of the organelles in the fluid, segment subcellular features within those images, and temporally align the frames to produce a normalized dataset for inference analysis. For example, the data processing system can compute displacement or curvature descriptors for each segmented organelle, evaluate those descriptors using an embedded neural network, and determine that none of the detected temporal patterns exceed an internal threshold indicating a transient event.

[0202] Upon determining that the first sequence of images does not indicate any transient organelle events, the data processing system can maintain the current flow and temperature conditions of the microfluidic channel, while continuing the imaging process. The data processing system can subsequently receive a second image sequence representing an updated state of the same organelles and can execute the same trained machine learning model to quantify inter-frame differences. In some implementations, the model can process spatial derivatives of phase intensity or fluorescence overlap to detect a transient structural modification such as a mitochondrial fission or lysosomal fusion event. For example, when the normalized event probability generated by the model exceeds a predefined threshold value, the data processing system can classify the sequence as containing a transient event and can record the associated timestamp for synchronization with the imaging and fluidic hardware. The second sequence can thus serve as the detection trigger for subsequent control actions carried out by the system components.-54-4919-3750-8870.1Atty Dkt.: 114198-0890

[0203] Responsive to identifying a transient event, the data processing system can transmit one or more electrical signals to an actuator of the solution reservoir to initiate a rapid replacement of the fluid inside the chamber. The control command may activate microvalves or micropumps to displace a defined solution volume along the microfluidic channel within a sub-second interval. In some implementations, the new solution can carry a chemical fixative or reagent such as paraformaldehyde or an actin-depolymerizing compound suitable for preserving or modifying the detected state of the organelles. For example, the data processing system can drive the actuator sequence to complete the fluid exchange within approximately one second of event detection, thereby preserving the morphological configuration of the organelles as recorded in the triggering image sequence. This automated actuation provides a closed-loop control between image-based event detection and microfluidic fluid exchange within the system. The integration of automated actuation and closed-loop control enables precisely timed interventions in response to detected cellular events, minimizing latency and allowing the system to capture transient or rare biological phenomena with high fidelity while reducing sample perturbation and phototoxicity.

[0204] At least one aspect relates to a microfluidic system for automated organelle transient event detection. The system can include a chamber configured to transport one or more organelles in a fluid. The system can include an image capture device oriented toward a face of the chamber, where the image capture device can generate images of the one or more organelles in the chamber over time. The system can include a controller that can receive, from the image capture device, a first sequence of images of the one or more organelles in the chamber. The system can execute a machine learning model using the first sequence of images to determine that the first sequence of images does not indicate any transient events in the one or more organelles. The machine learning model can be trained to detect human age-based transient events in organelles. The system can determine not to adjust operation of the microfluidic device based on the determination that the first sequence of images does not indicate any transient events in the one or more organelles. The system can receive a second sequence of images of the one or more organelles in the chamber after the first sequence. The system can execute the machine learning model using the second sequence of images to determine, based at least on differences between the first and second sequences, that the second sequence indicates a transient event in the one or more organelles. The system can-55-4919-3750-8870.1Atty Dkt.: 114198-0890 automatically trigger a solution exchange within the chamber responsive to determining that the second sequence indicates the transient event in the one or more organelles.

[0205] In some implementations, the system can receive a third sequence of images of the organelles in the chamber after the solution exchange and can execute the machine learning model using the third sequence of images to determine whether the solution exchange altered organelle dynamics. In some implementations, the system can quantify the mean distance traveled by each organelle in the chamber using image analysis and can compare the quantified distances to a reference dataset to identify abnormal organelle mobility. In some implementations, the system can identify actin aggregates containing organelles in the images using a trained classifier and can trigger the introduction of an actin depolymerizing agent into the chamber if immobilization is detected. In some implementations, the system can generate per-instance masks for individual organelles using a deep learning model and can classify each instance as mitochondria, lysosome, or endosome, verifying that class-specific event signatures are absent. In some implementations, the system can generate time-linked trajectories for each segmented organelle across the first sequence and can apply temporal pattern analysis to detect fission, fusion, or lysis signatures and conclude non-occurrence when no event pattern exceeds a detection threshold. In some implementations, the system can detect age-related organelle morphological features including fragmentation and enlargement and can record the frequency of detected features for statistical aggregation across multiple sequences. In some implementations, the system can automatically quantify inter-organelle interactions including colocalization events over time and can generate a session report summarizing interaction rates and spatial overlap distributions. In some implementations, the system can compare event frequencies and mobility profiles between samples grouped by donor age or disease state and can store comparative analytics and cohort labels in a database for longitudinal evaluation. In some implementations, the system can trigger chemical fixation of the organelles within one second of detecting a transient event and can initiate a correlative light-and-electron microscopy workflow for the fixed sample. In some implementations, the system can differentiate event likelihoods across organelle classes including mitochondria, lysosomes, and endoplasmic reticulum and can assign a confidence score to the determination for each class based on model output probabilities. In some implementations, the system can monitor imaging conditions for phototoxicity or photobleaching indicators during acquisition and can adapt-56-4919-3750-8870.1Atty Dkt.: 114198-0890 imaging parameters in real time to minimize phototoxic effects while maintaining analysis fidelity.

[0206] At least one other aspect relates to a method. The method can be performed, for example, by one or more processors coupled to non-transitory memory. The method can include receiving a first sequence of images of one or more organelles in a fluid chamber of a microfluidic device. The method can include executing a machine learning model using the first sequence of images to determine that the first sequence does not indicate any transient event in the organelles. The machine learning model can be trained to detect human age-based transient organelle events. The method can include determining not to adjust operation of the microfluidic device based on the determination that the first sequence does not indicate any transient events. The method can include receiving a second sequence of images of the organelles. The method can include executing the machine learning model using the second sequence to determine, based at least on differences from the first sequence, that the second sequence indicates a transient event. The method can include automatically triggering a solution exchange in the chamber responsive to detecting the transient event.

[0207] In some implementations, the method can include receiving a third sequence of images after the solution exchange and executing the machine learning model using the third sequence to determine whether organelle dynamics changed as a result of the solution exchange. In some implementations, the method can include quantifying the mean distance traveled by each organelle using image analysis and comparing the distances to a reference dataset to identify abnormal mobility. In some implementations, the method can include identifying actin aggregates containing organelles using a trained classifier and triggering the introduction of an actin depolymerizing agent into the chamber if immobilization is detected. In some implementations, the method can include segmenting individual organelles using a deep learning model to generate per-instance masks and classifying each as mitochondria, lysosome, or endosome to verify that class-specific event signatures are absent. In some implementations, the method can include generating time-linked trajectories for each segmented organelle and applying temporal analysis to identify fission, fusion, or lysis signatures while determining when no event pattern exceeds a detection threshold.

[0208] At least one other aspect relates to a non-transitory computer-readable medium storing executable instructions. When executed, the instructions can cause one or more processors to-57-4919-3750-8870.1Atty Dkt.: 114198-0890 perform a method of automated organelle transient event detection. The method can include receiving a first sequence of images of organelles in a chamber, executing a machine learning model using the first sequence to determine that no transient event occurs, and determining not to adjust operation of the microfluidic device. The method can include receiving a second sequence of images, executing the machine learning model using the second sequence to identify, based at least on differences between the sequences, a transient organelle event, and automatically triggering a solution exchange within the chamber responsive to the detected event. In some implementations, the method can include receiving a third sequence of images after the solution exchange and executing the machine learning model using the third sequence to assess whether the solution exchange caused changes in organelle dynamics.

[0209] FIGS. 15A and 15B are block diagrams depicting embodiments of computing devices that can be used in connection with the methods and systems described herein.

[0210] Having discussed specific embodiments of the present solution, it may be helpful to describe aspects of the operating environment as well as associated system components (e.g., hardware elements) in connection with the methods and systems described herein.

[0211] The systems and methods discussed herein may be deployed as and / or executed on any type and form of computing device, such as a computer, network device or appliance capable of communicating on any type and form of network and performing the operations described herein. FIGS. 15A and 15B depict block diagrams of a computing device 1500 useful for practicing an embodiment of the systems and methods described herein. As shown in FIGS. 15A and 15B, each computing device 1500 includes a central processing unit 1521, and a main memory unit 1522. As shown in FIG. 15 A, a computing device 1500 may include a storage device 1528, an installation device 1516, a network interface 1518, an I / O controller 1523, display devices 1524a-1524n, a keyboard 1526 and a pointing device 1527, such as a mouse. The storage device 1528 may include, without limitation, an operating system and / or software. As shown in FIG. 15B, each computing device 1500 may also include additional optional elements, such as a memory port 1503, abridge 1570, one or more input / output devices 1530a- 1530n (generally referred to using reference numeral 1530), and a cache memory 1540 in communication with the central processing unit 1521.-58-4919-3750-8870.1Atty Dkt.: 114198-0890

[0212] The central processing unit 1521 is any logic circuitry that responds to and processes instructions fetched from the main memory unit 1522. In many embodiments, the central processing unit 1521 is provided by a microprocessor unit, such as: those manufactured by Intel Corporation of Mountain View, California; those manufactured by International Business Machines of White Plains, New York; or those manufactured by Advanced Micro Devices of Sunnyvale, California. The computing device 1500 may be based on any of these processors, or any other processor capable of operating as described herein.

[0213] Main memory unit 1522 may be one or more memory chips capable of storing data and allowing any storage location to be directly accessed by the microprocessor 1521, such as any type or variant of Static random access memory (SRAM), Dynamic random access memory (DRAM), Ferroelectric RAM (FRAM), NAND Flash, NOR Flash and Solid State Drives (SSD). The main memory 1522 may be based on any of the above described memory chips, or any other available memory chips capable of operating as described herein. In the embodiment shown in FIG. 15 A, the processor 1521 communicates with main memory 1522 via a system bus 1580 (described in more detail below). FIG. 15B depicts an embodiment of a computing device 1500 in which the processor communicates directly with main memory 1522 via a memory port 1503. For example, in FIG. 15B the main memory 1522 may be DRDRAM.

[0214] FIG. 15B depicts an embodiment in which the main processor 1521 communicates directly with cache memory 1540 via a secondary bus, sometimes referred to as a backside bus. In other embodiments, the main processor 1521 communicates with cache memory 1540 using the system bus 1580. Cache memory 1540 typically has a faster response time than main memory 1522 and is provided by, for example, SRAM, BSRAM, or EDRAM. In the embodiment shown in FIG. 15B, the processor 1521 communicates with various I / O devices 1530 via a local system bus 1580. Various buses may be used to connect the central processing unit 1521 to any of the VO devices 1530, for example, a VESA VL bus, an ISA bus, an EISA bus, a MicroChannel Architecture (MCA) bus, a PCI bus, a PCI-X bus, a PCI-Express bus, or a NuBus. For embodiments in which the I / O device is a video display 1524, the processor 1521 may use an Advanced Graphics Port (AGP) to communicate with the display 1524. FIG. 15B depicts an embodiment of a computer 1500 in which the main processor 1521 may communicate directly with I / O device 1530b, for example via HYPERTRANSPORT,-59-4919-3750-8870.1Atty Dkt.: 114198-0890RAPIDIO, or INFINIBAND communications technology. FIG. 15B also depicts an embodiment in which local busses and direct communication are mixed: the processor 1521 communicates with I / O device 1530a using a local interconnect bus while communicating with I / O device 1530b directly.

[0215] A wide variety of I / O devices 1530a-1530n may be present in the computing device 1500. Input devices include keyboards, mice, trackpads, trackballs, microphones, dials, touch pads, touch screens, and drawing tablets. Output devices include video displays, speakers, inkjet printers, laser printers, projectors and dye-sublimation printers. The VO devices may be controlled by an VO controller 1523 as shown in FIG. 15 A. The I / O controller may control one or more I / O devices such as a keyboard 1526 and a pointing device 1527, e.g., a mouse or optical pen. Furthermore, an I / O device may also provide storage and / or an installation device 1516 for the computing device 1500. In still other embodiments, the computing device 1500 may provide USB connections (not shown) to receive handheld USB storage devices such as the USB Flash Drive line of devices manufactured by Twintech Industry, Inc., of Los Alamitos, California.

[0216] Referring again to FIG. 15 A, the computing device 1500 may support any suitable installation device 1516, such as a disk drive, a CD-ROM drive, a CD-R / RW drive, a DVD- ROM drive, a flash memory drive, tape drives of various formats, USB device, hard-drive, a network interface, or any other device suitable for installing software and programs. The computing device 1500 may further include a storage device, such as one or more hard disk drives or redundant arrays of independent disks, for storing an operating system and other related software, and for storing application software programs such as any program or software 1520 for implementing (e.g., configured and / or designed for) the systems and methods described herein. Optionally, any of the installation devices 1516 could also be used as the storage device. Additionally, the operating system and the software can be run from a bootable medium.

[0217] Furthermore, the computing device 1500 may include a network interface 1518 to interface to a network through a variety of connections including, but not limited to, standard telephone lines, LAN or WAN links (e.g., 802.11, Tl, T3, 156kb, X.25, SNA, DECNET), broadband connections (e.g., ISDN, Frame Relay, ATM, Gigabit Ethernet, Ethernet-over- SONET), wireless connections, or some combination of any or all of the above. Connections -60-4919-3750-8870.1Atty Dkt.: 114198-0890 can be established using a variety of communication protocols (e.g., TCP / IP, IPX, SPX, NetBIOS, Ethernet, ARCNET, SONET, SDH, Fiber Distributed Data Interface (FDDI), RS232, IEEE 802.11, IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, IEEE 802.1 In, IEEE 802.1 lac, IEEE 802. Had, CDMA, GSM, WiMax and direct asynchronous connections). In one embodiment, the computing device 1500 communicates with other computing devices 1500’ via any type and / or form of gateway or tunneling protocol such as Secure Socket Layer (SSL) or Transport Layer Security (TLS). The network interface 1518 may include a built-in network adapter, network interface card, PCMCIA network card, card bus network adapter, wireless network adapter, USB network adapter, modem or any other device suitable for interfacing the computing device 1500 to any type of network capable of communication and performing the operations described herein.

[0218] In some implementations, the computing device 1500 may include or be connected to one or more display devices 1524a-1524n. As such, any of the I / O devices 1530a-1530n and / or the I / O controller 1523 may include any type and / or form of suitable hardware, software, or combination of hardware and software to support, enable or provide for the connection and use of the display device(s) 1524a-1524n by the computing device 1500. For example, the computing device 1500 may include any type and / or form of video adapter, video card, driver, and / or library to interface, communicate, connect or otherwise use the display device(s) 1524a- 1524n. In one embodiment, a video adapter may include multiple connectors to interface to the display device(s) 1524a-1524n. In other embodiments, the computing device 1500 may include multiple video adapters, with each video adapter connected to the display device(s) 1524a-1524n. In some implementations, any portion of the operating system of the computing device 1500 may be configured for using multiple displays 1524a-1524n. One ordinarily skilled in the art will recognize and appreciate the various ways and embodiments that a computing device 1500 may be configured to have one or more display devices 1524a-1524n.

[0219] In further embodiments, an I / O device 1530 may be a bridge between the system bus 1580 and an external communication bus, such as a USB bus, an Apple Desktop Bus, an RS- 232 serial connection, a SCSI bus, a FireWire bus, a FireWire 500 bus, an Ethernet bus, an AppleTalk bus, a Gigabit Ethernet bus, an Asynchronous Transfer Mode bus, a FibreChannel bus, a Serial Attached small computer system interface bus, a USB connection, or a HDMI bus.-61-4919-3750-8870.1Atty Dkt.: 114198-0890

[0220] A computing device 1500 of the sort depicted in FIGS. 15 A and 15B may operate under the control of an operating system, which control scheduling of tasks and access to system resources. The computing device 1500 can be running any operating system, such as any of the versions of the MICROSOFT WINDOWS operating systems, the different releases of the Unix and Linux operating systems, any version of the MAC OS for Macintosh computers, any embedded operating system, any real-time operating system, any open source operating system, any proprietary operating system, any operating systems for mobile computing devices, or any other operating system capable of running on the computing device and performing the operations described herein. Typical operating systems include, but are not limited to, Android, produced by Google Inc.; WINDOWS 7 and 8, produced by Microsoft Corporation of Redmond, Washington; MAC OS, produced by Apple Computer of Cupertino, California; WebOS, produced by Research In Motion (RIM); OS / 2, produced by International Business Machines of Armonk, New York; and Linux, a freely-available operating system distributed by Caldera Corp, of Salt Lake City, Utah, or any type and / or form of a Unix operating system, among others.

[0221] The computer system 1500 can be any workstation, telephone, desktop computer, laptop or notebook computer, server, handheld computer, mobile telephone or other portable telecommunications device, media playing device, a gaming system, mobile computing device, or any other type and / or form of computing, telecommunications or media device that is capable of communication. The computer system 1500 has sufficient processor power and memory capacity to perform the operations described herein.

[0222] In some implementations, the computing device 1500 may have different processors, operating systems, and input devices consistent with the device. For example, in one embodiment, the computing device 1500 is a smart phone, mobile device, tablet or personal digital assistant. In still other embodiments, the computing device 1500 is an Android-based mobile device, an iPhone smart phone manufactured by Apple Computer of Cupertino, California, or a Blackberry or WebOS-based handheld device or smart phone, such as the devices manufactured by Research In Motion Limited. Moreover, the computing device 600 can be any workstation, desktop computer, laptop or notebook computer, server, handheld computer, mobile telephone, any other computer, or other form of computing or-62-4919-3750-8870.1Atty Dkt.: 114198-0890 telecommunications device that is capable of communication and that has sufficient processor power and memory capacity to perform the operations described herein.

[0223] Although examples of communications systems described above may include devices operating according to an 802.11 standard, it should be understood that embodiments of the systems and methods described can operate according to other standards and use wireless communications devices other than devices configured as devices and APs. For example, multiple-unit communication interfaces associated with cellular networks, satellite communications, vehicle communication networks, and other non-802.11 wireless networks can utilize the systems and methods described herein to achieve improved overall capacity and / or link quality without departing from the scope of the systems and methods described herein.

[0224] It should be noted that certain passages of this disclosure may reference terms such as “first” and “second” in connection with devices, mode of operation, transmit chains, antennas, etc., for purposes of identifying or differentiating one from another or from others. These terms are not intended to merely relate entities (e.g., a first device and a second device) temporally or according to a sequence, although in some cases, these entities may include such a relationship. Nor do these terms limit the number of possible entities (e.g., devices) that may operate within a system or environment.

[0225] It should be understood that the systems described above may provide multiple ones of any or each of those components and these components may be provided on either a standalone machine or, in some implementations, on multiple machines in a distributed system. In addition, the systems and methods described above may be provided as one or more computer- readable programs or executable instructions embodied on or in one or more articles of manufacture. The article of manufacture may be a floppy disk, a hard disk, a CD-ROM, a flash memory card, a PROM, a RAM, a ROM, or a magnetic tape. In general, the computer-readable programs may be implemented in any programming language, such as LISP, PERL, C, C++, C#, PROLOG, or in any byte code language such as JAVA. The software programs or executable instructions may be stored on or in one or more articles of manufacture as object code.

[0226] Equivalents-63-4919-3750-8870.1Atty Dkt.: 114198-0890

[0227] It is to be understood that while the invention has been described in conjunction with the above embodiments, that the foregoing description and examples are intended to illustrate and not limit the scope of the invention. Other aspects, advantages and modifications within the scope of the invention is apparent to those skilled in the art to which the invention pertains.

[0228] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. All nucleotide sequences provided herein are presented in the 5' to 3' direction.

[0229] The inventions illustratively described herein may suitably be practiced in the absence of any element or elements, limitation or limitations, not specifically disclosed herein. Thus, for example, the terms “comprising”, “including,” containing”, etc. shall be read expansively and without limitation. Additionally, the terms and expressions employed herein have been used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions of excluding any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention claimed.

[0230] Thus, it should be understood that although the present invention has been specifically disclosed by preferred embodiments and optional features, modification, improvement and variation of the inventions embodied therein herein disclosed may be resorted to by those skilled in the art, and that such modifications, improvements and variations are considered to be within the scope of this invention. The materials, methods, and examples provided here are representative of preferred embodiments, are exemplary, and are not intended as limitations on the scope of the invention.

[0231] The invention has been described broadly and generically herein. Each of the narrower species and subgeneric groupings falling within the generic disclosure also form part of the invention. This includes the generic description of the invention with a proviso or negative limitation removing any subject matter from the genus, regardless of whether or not the excised material is specifically recited herein.

[0232] In addition, where features or aspects of the invention are described in terms of Markush groups, those skilled in the art will recognize that the invention is also thereby described in terms of any individual member or subgroup of members of the Markush group.-64-4919-3750-8870.1Atty Dkt.: 114198-0890

[0233] All publications, patent applications, patents, and other references mentioned herein are expressly incorporated by reference in their entirety, to the same extent as if each were incorporated by reference individually. In case of conflict, the present specification, including definitions, will control.

[0234] Other aspects are set forth within the following claims.4919-3750-8870.1

Claims

1. Atty Dkt.: 114198-0890WHAT IS CLAIMED IS:

1. A microfluidic system for automated organelle transient event detection, the microfluidic device comprising: a chamber configured to transport one or more organelles in a fluid; an image capture device oriented toward a face of the chamber, the image capture device to generate images of the one or more organelles in the chamber over time; and a controller comprising one or more processors configured by executable instructions stored in memory to, when the one or more organelles are located in the fluid in the chamber: receive, from the image capture device, a first sequence of images of the one or more organelles in the chamber; execute a machine learning model using the first sequence of images to determine the first sequence of images does not indicate any transient events in the one or more organelles, the machine learning model trained to detect human age-based transient events in organelles; determine not to adjust operation of the microfluidic device based on the determination that the first sequence of images does not indicate any transient events in the one or more organelles; subsequent to receipt of the first sequence of images, receive a second sequence of images of the one or more organelles in the chamber; execute the machine learning model using the second sequence of images to determine, at least based on differences between the second sequence of images, the second sequence of images indicates a transient event in the one or more organelles; and automatically trigger a solution exchange within the chamber responsive to the determination that the second sequence of images indicates the transient event in the one or more organelles.

2. The microfluidic system of claim 1, wherein the controller is further configured to: receive a third sequence of images of the organelles in the chamber after the solution exchange; and-66-4919-3750-8870.1Atty Dkt.: 114198-0890 execute the machine learning model using the third sequence of images to determine whether the solution exchange altered organelle dynamics.

3. The microfluidic system of claim 1, wherein the controller is configured to execute the machine learning model to determine the first sequence of images does not indicate any transient events in the one or more organelles by: quantifying the mean distance traveled by each organelle in the chamber using image analysis; and comparing the quantified distances to a reference dataset to identify abnormal organelle mobility.

4. The microfluidic system of claim 1, wherein the controller is configured to automatically trigger a solution exchange within the chamber responsive to the determination that the second sequence of images indicates the transient event in the one or more organelles by: identifying actin aggregates containing organelles in the images using a trained classifier; and triggering the introduction of an actin depolymerizing agent into the chamber if immobilization is detected.

5. The microfluidic system of claim 1, wherein the controller is configured to segment and classify organelles by: generating per-instance masks for individual organelles using a deep learning model; and classifying each instance as mitochondria, lysosome, or endosome and verifying that class-specific event signatures are absent.

6. The microfluidic system of claim 1, wherein the controller is configured to track organelle movement and detect transient events by: generating time-linked trajectories for each segmented organelle across the first sequence; and-67-4919-3750-8870.1Atty Dkt.: 114198-0890 applying temporal pattern analysis to detect fission, fusion, or lysis signatures and concluding non-occurrence when no event-pattern exceeds a detection threshold.

7. The microfluidic system of claim 1, wherein the controller is further configured to: detect age-related organelle morphological features, including fragmentation and enlargement; and record the frequency of detected features for statistical aggregation across multiple sequences of images.

8. The microfluidic system of claim 1, wherein the controller is further configured to: automatically quantify inter-organelle interactions, including colocalization events over time; and generate a session report summarizing interaction rates and spatial overlap distributions.

9. The microfluidic system of claim 1, wherein the controller is further configured to: compare event frequencies and mobility profiles between samples grouped by donor age or disease state; and store comparative analytics and cohort labels in a database for longitudinal evaluation.

10. The microfluidic system of claim 1, wherein the controller is further configured to: trigger chemical fixation of the organelles within one second of detecting a transient event; and initiate a correlative light-and-electron microscopy workflow for the fixed sample.

11. The microfluidic system of claim 1, wherein the controller is configured to distinguish between transient events in different organelle types by: differentiating event likelihoods across organelle classes including mitochondria, lysosomes, and endoplasmic reticulum; and assigning a confidence score to the determination for each class based on model output probabilities.-68-4919-3750-8870.1Atty Dkt.: 114198-089012. The microfluidic system of claim 1, wherein the controller is further configured to: monitor imaging conditions for phototoxicity or photobleaching indicators during acquisition; and adapt imaging parameters in real time to minimize phototoxic effects while maintaining analysis fidelity.

13. A method for automated organelle transient event detection, the method comprising: receiving, by one or more processors from an image capture device oriented toward a face of a chamber of a microfluidic device configured to transport one or more organelles in a fluid, a first sequence of images of the one or more organelles in the chamber; executing, by the one or more processors, a machine learning model using the first sequence of images to determine the first sequence of images does not indicate any transient event in the one or more organelles, the machine learning model trained to detect human agebased transient events in organelles; determining, by the one or more processors, not to adjust operation of the microfluidic device based on the determination that the first sequence of images does not indicate any transient events in the one or more organelles; subsequent to receiving the first sequence of images, receiving, by the one or more processors, a second sequence of images of the one or more organelles in the chamber; executing, by the one or more processors, the machine learning model using the second sequence of images to determine, at least based on differences between the second sequence of images, the second sequence of images indicates a transient event in the one or more organelles; and automatically triggering, by the one or more processors, a solution exchange within the chamber responsive to the determination that the second sequence of images indicates the transient event in the one or more organelles.-69-4919-3750-8870.1Atty Dkt.: 114198-089014. The method of claim 13, further comprising: receiving, by the one or more processors, a third sequence of images of the one or more organelles in the chamber after the solution exchange; and executing, by the one or more processors, the machine learning model using the third sequence of images to determine whether the solution exchange altered organelle dynamics.

15. The method of claim 13, wherein executing the machine learning model to determine the first sequence of images does not indicate any transient event in the one or more organelles comprises: quantifying, by the one or more processors, the mean distance traveled by each organelle in the chamber using image analysis; and comparing, by the one or more processors, the quantified distances to a reference dataset to identify abnormal organelle mobility.

16. The method of claim 13, wherein automatically triggering a solution exchange within the chamber responsive to the determination that the second sequence of images indicates the transient event in the one or more organelles comprises: identifying, by the one or more processors, actin aggregates containing organelles in the images using a trained classifier; and triggering, by the one or more processors, the introduction of an actin depolymerizing agent into the chamber if immobilization is detected.

17. The method of claim 13, wherein executing the machine learning model to determine the first sequence of images does not indicate any transient event in the one or more organelles comprises: segmenting, by the one or more processors, individual organelles in the images using a deep learning model to generate per-instance masks; and classifying, by the one or more processors, each instance as mitochondria, lysosome, or endosome and verifying that class-specific event signatures are absent.-70-4919-3750-8870.1Atty Dkt.: 114198-089018. The method of claim 13, wherein executing the machine learning model to determine the first sequence of images does not indicate any transient event in the one or more organelles comprises: generating, by the one or more processors, time-linked trajectories for each segmented organelle across the first sequence; and applying, by the one or more processors, temporal pattern analysis to detect fission, fusion, or lysis signatures and concluding non-occurrence when no event-pattern exceeds a detection threshold.

19. A non-transitory computer-readable medium storing executable instructions that, when executed by one or more processors, cause the one or more processors to perform a method for automated organelle transient event detection, the method comprising: receiving, from an image capture device oriented toward a face of a chamber of a microfluidic device configured to transport one or more organelles in a fluid, a first sequence of images of the one or more organelles in the chamber; executing a machine learning model using the first sequence of images to determine the first sequence of images does not indicate any transient event in the one or more organelles, the machine learning model trained to detect human age-based transient events in organelles; determining not to adjust operation of the microfluidic device based on the determination that the first sequence of images does not indicate any transient events in the one or more organelles; subsequent to receiving the first sequence of images, receiving a second sequence of images of the one or more organelles in the chamber; executing the machine learning model using the second sequence of images to determine, at least based on differences between the second sequence of images, the second sequence of images indicates a transient event in the one or more organelles; and automatically triggering a solution exchange within the chamber responsive to the determination that the second sequence of images indicates the transient event in the one or more organelles.-71-4919-3750-8870.1Atty Dkt.: 114198-089020. The non-transitory computer-readable medium of claim 13, wherein the method further comprises: receiving a third sequence of images of the one or more organelles in the chamber after the solution exchange; and executing the machine learning model using the third sequence of images to determine whether the solution exchange altered organelle dynamics.-72-4919-3750-8870.1

Images