Bubbles self-separation digital detection method and system for extracellular vesicle miRNA quantification

Abstract

The application discloses a bubble self-layering digital detection method and system for extracellular vesicle miRNA quantification, and belongs to the field of biomedical engineering and molecular diagnosis technology. The technical problem to be solved is to provide a simple and sensitive tumor extracellular vesicle miRNA detection method capable of simultaneously detecting multiple miRNAs. The scheme is to incubate the sample to be detected with an immunocapture bubble coupled with an anti-EpCAM antibody to enrich tumor extracellular vesicles, obtain a sample to be detected by in-situ ultrasonic lysis, and then construct an enzyme cutting system by combining a fluorescence / DNA double-coding magnetic bead conjugate, add multifunctional click glass microbubbles to capture unreacted magnetic beads, and then collect the magnetic beads at the bottom after standing self-layering. After fluorescence imaging, the image is input into an AI automatic fluorescence counting decoding module for counting to obtain the result. The method is used for detecting miR-21 and miR-155 in tumor extracellular vesicles, and is suitable for non-invasive early diagnosis of tumors.

Description

Technical Field

[0001] This invention belongs to the field of biomedical engineering and molecular diagnostics technology, specifically relating to a method and system for the digital detection of bubble self-stratification for the quantification of extracellular vesicle miRNA. Background Technology

[0002] Extracellular vesicles (EVs) containing microRNAs (miRNAs) have been shown to be closely related to tumorigenesis and development, making them a promising class of biopsy biopsy markers. However, specifically enriching tumor-derived EVs from complex biological samples (such as plasma) and accurately quantifying their internal miRNAs remains technically challenging. Existing EV isolation methods based on ultracentrifugation or kits are inefficient and struggle to differentiate EV subpopulations. Subsequent miRNA detection often relies on quantitative reverse transcription polymerase chain reaction (RT-qPCR) or next-generation sequencing, which typically require large sample volumes and total RNA extraction steps. These methods are also susceptible to interference from highly abundant free miRNAs in the sample, leading to biased results. To circumvent such interference, researchers have attempted to directly introduce detection probes into EVs for in-situ detection, but these methods are limited by membrane penetration efficiency, resulting in sensitivity that fails to meet clinical needs. In recent years, digital detection technologies based on micropores or microdroplets have achieved improved sensitivity at the single-molecule level through physical compartmentalization. However, these methods often require precise microfluidic control or multi-step surface capture operations, resulting in significant interference from empty compartments and low sampling efficiency, thus limiting the number of target molecules that can be counted. Furthermore, most existing technologies still rely on external force fields such as centrifugation or magnetic separation for signal separation, making the process cumbersome. The final results depend on manual counting or analysis using commercial software, making it difficult to avoid subjective bias and undercounting or miscounting due to adhering particles. Therefore, achieving highly specific enrichment, efficient signal conversion, and automated, unbiased quantification of specific miRNAs in tumor-derived EVs without relying on complex equipment and cumbersome operations remains a pressing challenge in this field. Summary of the Invention

[0003] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0004] Another objective of this invention is to provide a bubble-based self-stratification digital detection method for quantifying extracellular vesicle miRNA. This method utilizes immune-captured bubbles to achieve highly efficient and specific enrichment of tumor-derived extracellular vesicles. Combined with in-situ ultrasonic lysis to release target miRNA, and through double-stranded specific nuclease-mediated cyclic signal amplification and fluorescent / DNA dual-encoding magnetic bead conjugates, it achieves simultaneous detection of miR-21 and miR-155 dual targets. Furthermore, by leveraging multifunctional clickable glass microbubbles for clickable chemical capture of unreacted magnetic beads and self-buoyancy-driven self-stratification, background signal interference is effectively eliminated. Finally, an AI-automated fluorescence counting and decoding module completes the unbiased counting and signal output of the dual-target magnetic beads. This provides a simple, non-complex, highly sensitive, and clinically applicable ultrasensitive digital detection method for tumor extracellular vesicle miRNA in plasma samples.

[0005] To achieve these objectives and other advantages of the present invention, a bubble self-stratification digital detection method for quantifying extracellular vesicle miRNA is provided, comprising the following steps:

[0006] S1: Provide the sample to be tested, add immunocapture bubbles, incubate for 10-60 min, then centrifuge at 500-2000 rpm for 0.5-1.5 min to remove the supernatant, wash 3-5 times with phosphate buffer to obtain immunocapture bubbles enriched with tumor cell extracellular vesicles; the immunocapture bubbles are hollow glass microbubbles with anti-epithelial cell adhesion molecule (EpCAM) antibody coupled to the surface;

[0007] S2: Immunocapture bubbles enriched with tumor cell extracellular vesicles are subjected to in situ ultrasonic lysis to release miRNAs within the vesicles. After centrifugation at 500-2000 rpm for 3-10 min, the supernatant is collected to obtain the miRNA sample to be tested.

[0008] S3: The miRNA sample to be tested is mixed with fluorescent / DNA dual-coding magnetic bead conjugates, double-stranded specific nuclease (DSN), DSN reaction buffer, RNase inhibitor, and diethyl pyrocarbonate (DEPC) treated water to form an enzyme digestion system. The mixture is incubated for digestion. After incubation, DSN termination solution is added to terminate the reaction. The magnetic beads are collected by magnetic separation and washed 3-5 times with phosphate-buffered saline (PBS). The fluorescent / DNA dual-coding magnetic bead conjugates include FITC-MB-DNA21-N3 conjugate and Cy3-MB-DNA155-N3 conjugate, which are used to detect miR-21 and miR-155, respectively.

[0009] S4: The magnetic beads collected in step S3 are resuspended in a multifunctional click glass microbubble solution and incubated to allow the unreacted fluorescent / DNA dual-encoding magnetic bead conjugates to bind to the surface of the multifunctional click glass microbubble through a click chemical reaction of azide and dibenzocyclooctylene; the multifunctional click glass microbubble is a hollow glass microbubble with dibenzocyclooctylene groups grafted onto its surface.

[0010] S5: Allow the liquid to stand so that the multifunctional click glass microbubbles, which incorporate unreacted magnetic beads, float to the surface and collect the magnetic beads that have settled at the bottom.

[0011] S6: Perform fluorescence imaging on the collected magnetic beads, acquire fluorescence images of the FITC channel and Cy3 channel respectively, input the images into the AI ​​automated fluorescence counting and decoding module for counting, and obtain the detection results of miR-21 and miR-155.

[0012] Preferably, the preparation of the immune capture bubble includes the following steps:

[0013] Weigh 0.1-1 g of hollow glass microbubbles after hydroxylation activation treatment, add 5-15 mL of anhydrous ethanol solution containing 0.1-1 mL of (3-(2,3-epoxypropoxy)propyl)triethoxysilane (GPTES) to carry out silanization modification reaction for 10-20 h, centrifuge at 3000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with ultrapure water, and vacuum dry at 35-40 ℃ to obtain epoxy hollow glass microbubbles;

[0014] Weigh 0.1-1 g of epoxy hollow glass microbubbles, add 1-10 mL of phosphate buffer, add an anti-EpCAM antibody solution with a concentration of 80-120 μg / mL, mix and react for 0.5-1.5 h, centrifuge at 1000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with phosphate buffer, block with bovine serum albumin, wash again and dry to obtain immunocapture bubbles with surface-coupled anti-EpCAM antibody.

[0015] Preferably, the preparation of fluorescent / DNA dual-coding magnetic bead conjugates includes the following steps:

[0016] Streptavidin-conjugated magnetic beads were mixed with fluorescein isothiocyanate-succinimide ester (FITC-NHS) and Cy3-succinimide ester (Cy3-NHS), respectively, and rotated at room temperature in the dark for 30-90 min to perform fluorescent labeling. The supernatant was removed by magnetic separation, and the beads were washed 3-5 times with 0.01 g / 100 mL Tween 20 phosphate buffer (PBST) to obtain FITC-labeled magnetic beads and Cy3-labeled magnetic beads, respectively.

[0017] Biotin-modified clickable DNA probe Biotin-DNA21-N3 was mixed with FITC-labeled magnetic beads and immobilized by the affinity interaction between biotin and streptavidin to obtain FITC-MB-DNA21-N3 conjugate.

[0018] Biotin-modified clickable DNA probe Biotin-DNA155-N3 was mixed with Cy3-labeled magnetic beads and immobilized by the affinity interaction between biotin and streptavidin to obtain Cy3-MB-DNA155-N3 conjugate.

[0019] Preferably, in step S2, the parameters for in-situ ultrasonic lysis are: frequency 10-30 kHz, each ultrasonic session lasting 5-10 s, followed by a pause of 5-20 s, and repeated 4-6 times.

[0020] Preferably, in step S3, the volume of the enzyme digestion system is 20 μL, comprising: 6 μL of equimolarly mixed fluorescent / DNA dual-coding magnetic bead conjugate, 10 μL of the miRNA sample to be tested, 1 μL of 0.5U DSN enzyme, 1 μL of 10×DSN reaction buffer, 0.5 μL of 20 U RNase inhibitor, and 1.5 μL of DEPC-treated water; the incubation temperature is 40-60 ℃, and the time is 0.5-1.5 h.

[0021] Preferably, the preparation of multifunctional clickable glass microbubbles includes the following steps:

[0022] Hydroxylation activation treatment was performed on the microbubbles of insulating glass to obtain activated insulating glass microbubbles;

[0023] 100-1000 mg of activated hollow glass microbubbles were added to 10-50 mL of polyethyleneimine aqueous solution with a mass-to-volume ratio of 1-5 mg / mL. The mixture was rotated and reacted at room temperature for 10-20 min to form an aminated coating on the surface by electrostatic adsorption. After centrifugation at 3000-5000 rpm for 0.5-2 min, washing with ultrapure water 3-5 times, and vacuum drying at 30-40 °C, aminated hollow glass microbubbles were obtained.

[0024] 10-1000 mg of aminated hollow glass microbubbles were added to 0.5-20 mL of phosphate buffer, followed by 100-300 μL of a 10-30 mM dibenzocyclooctyne-N-hydroxysuccinimide ester solution. The mixture was rotated and reacted at room temperature for 0.5-1.5 h to graft dibenzocyclooctyne groups via a covalent reaction between the surface amino group and the active ester. After centrifugation at 3000-5000 rpm for 0.5-2 min, washing with ultrapure water 3-5 times, and vacuum drying at 30-40 °C, multifunctional click glass microbubbles with bioorthogonal functions were obtained.

[0025] Preferably, the multifunctional clickable glass microbubbles have a particle size of 10-30 μm and a density of 0.5-0.7 g / cm³. 3 The floating time in phosphate buffer should not exceed 5 minutes.

[0026] Preferably, the AI-automated fluorescence counting decoding module incorporates image classification algorithms, grayscale algorithms, adaptive binarization algorithms, aspect ratio threshold-based image segmentation algorithms, and HSV parameter adaptive adjustment algorithms to automatically classify, preprocess, segment, and count the input FITC and Cy3 channel fluorescence images, and output the dual-target counting results.

[0027] Preferably, the AI-automated fluorescence counting decoding module performs the following processing flow:

[0028] 1) For the acquired FITC and Cy3 channel fluorescence images, the phase correlation algorithm based on fast Fourier transform is used to calculate the subpixel translation offset between the two images, and the images are corrected according to the offset to make the two channel images accurately aligned, ensuring that the magnetic beads in the same field of view have consistent coordinates in the two channels.

[0029] 2) Convert the registered FITC and Cy3 images from the RGB color space to the HSV color space respectively, and extract the hue (H), saturation (S), and lightness (V) components; according to the pre-calibrated characteristic hue range of FITC and Cy3 fluorescent dyes, perform hue matching on each pixel to generate a binary mask, and initially screen out the pixel areas that may be magnetic beads in order to eliminate noise interference from non-target colors in the background.

[0030] 3) Within the mask area after hue screening, the lightness component V is segmented locally using the Bernson method based on the mean and standard deviation of the pixel neighborhood to obtain a binary image of the magnetic bead candidate area, in order to adapt to uneven illumination and changes in background fluorescence.

[0031] 4) Perform morphological opening operations to remove small noise and closing operations to connect neighboring regions on the binary image in sequence. Then, use morphological reconstruction algorithms to fill the holes inside the magnetic beads caused by uneven fluorescence to obtain the complete magnetic bead connected region.

[0032] 5) Perform distance transformation on the connected components, calculate the Euclidean distance from each pixel to the background, and generate a distance map; search for local maxima on the distance map as seed points, and use the watershed algorithm to segment the adhered magnetic beads to obtain the independent regions of individual magnetic beads, thus solving the counting bias caused by magnetic bead aggregation.

[0033] 6) Calculate the area, perimeter, roundness, and aspect ratio of each independent region. Set a threshold based on the actual particle size of the magnetic bead and the imaging magnification. Remove regions that do not conform to the roundness of the magnetic bead or have abnormal size, and filter out long strip regions that are still stuck together due to incomplete segmentation.

[0034] 7) Count the number of filtered magnetic beads in the FITC channel and Cy3 channel respectively to obtain the counting results of miR-21 and miR-155; at the same time, if a magnetic bead is detected in both channels, it is determined to be the same target based on its coordinates and is regarded as cross-color interference and removed to ensure the specificity and accuracy of dual-target counting.

[0035] In fluorescence microscopy, subpixel shift between dual-channel images, background variations caused by uneven illumination, holes formed by uneven fluorescence distribution of magnetic beads, magnetic bead aggregation and adhesion, and cross-color interference between different channels can all severely affect the accuracy of counting. To address these issues, this method first employs a phase correlation algorithm based on Fast Fourier Transform to accurately register the FITC and Cy3 channel images, ensuring that the coordinates of magnetic beads within the same field of view are consistent across both channels. Subsequently, the registered images are converted to the HSV color space, and a binary mask is generated using pre-calibrated FITC and Cy3 characteristic hue ranges to effectively eliminate noise interference from non-target colors in the background. Within the mask region, the luminance component V is segmented using a locally adaptive threshold based on the pixel neighborhood mean and standard deviation to adapt to uneven illumination and background fluorescence variations, accurately extracting candidate regions for magnetic beads. Furthermore, morphological opening operations are used to remove minor noise, closing operations connect neighboring regions, and a morphological reconstruction algorithm is applied to fill the holes inside the magnetic beads caused by uneven fluorescence, obtaining complete connected regions. For adhered magnetic beads, a distance map is generated using distance transformation. Local maxima are searched on the distance map as seed points, and a watershed algorithm is used to segment the adhered region into individual magnetic beads, avoiding undercounting caused by aggregation. Subsequently, thresholds are set based on the actual particle size, roundness, and aspect ratio of the magnetic beads to remove false positive targets that do not meet the characteristics of magnetic beads and elongated regions left by incomplete segmentation. Finally, the number of filtered magnetic beads in both channels is counted separately, and crosstalk interference appearing in both channels simultaneously is removed based on coordinate correlation, thus obtaining accurate counts of miR-21 and miR-155. This series of image processing steps is interconnected, eliminating various interference factors such as offset, noise, adhesion, and crosstalk at the source, achieving high-precision and high-reliability automated counting of dual-target magnetic beads, providing accurate data support for the quantitative analysis of miRNAs in clinical samples.

[0036] A bubble self-stratification digital detection system for quantifying extracellular vesicle miRNAs, comprising:

[0037] Immunocapture bubbles, which are hollow glass microbubbles with anti-EpCAM antibodies coupled to their surface, are used to specifically enrich tumor cell extracellular vesicles.

[0038] Fluorescent / DNA dual-coding magnetic bead conjugates, including FITC-MB-DNA21-N3 conjugate and Cy3-MB-DNA155-N3 conjugate, are used to detect miR-21 and miR-155, respectively;

[0039] Multifunctional clickable glass microbubbles are hollow glass microbubbles with dibenzocyclooctylene groups grafted onto their surface. They are used to capture unreacted fluorescent / DNA dual-encoded magnetic bead conjugates through an azide-dibenzocyclooctylene clickable chemical reaction and achieve self-separation by floating up on their own buoyancy.

[0040] Double-stranded specific nucleases and their reaction buffers are used for cyclic cleavage of DNA probes on fluorescent / DNA dual-coding magnetic bead conjugates in the presence of target miRNA;

[0041] The AI-automated fluorescence counting decoding module is configured to perform the following image processing flow:

[0042] 1) Perform phase correlation registration based on fast Fourier transform on the acquired FITC and Cy3 channel fluorescence images to eliminate sub-pixel offset between channels;

[0043] 2) Convert the registered image to the HSV color space, generate a binary mask based on the characteristic hue range of the pre-calibrated FITC and Cy3 fluorescent dyes, and eliminate background noise;

[0044] 3) Local adaptive thresholding is applied to the brightness component V within the mask region to obtain candidate regions for magnetic beads;

[0045] 4) Morphological reconstruction and hole filling are performed on the candidate regions to obtain complete magnetic bead connected domains;

[0046] 5) Perform distance transformation on the connected components and use the watershed algorithm to segment the sticky magnetic beads to obtain independent regions for each individual magnetic bead;

[0047] 6) Shape and size filtering is performed based on the particle size, roundness, and aspect ratio of the magnetic beads to eliminate false positive targets;

[0048] 7) Count the number of filtered magnetic beads in the dual channels, and remove cross-color interference based on coordinate correlation, and output the counting results of miR-21 and miR-155.

[0049] The present invention has at least the following beneficial effects:

[0050] 1. The multifunctional clickable glass microbubble designed in this invention possesses both bioorthogonal reaction activity and self-suspension properties. Through a catalyst-free clickable chemical reaction between DBCO and azide groups, it can specifically capture unreacted azide-modified encoded magnetic beads and utilize its low-density hollow structure to achieve rapid buoyancy, completing the self-separation of reacted and unreacted magnetic beads. This process does not require the intervention of external force fields such as centrifugation or magnetic fields, effectively eliminating background interference from unreacted magnetic beads, achieving unbiased counting of target signals, and overcoming the technical bottlenecks of severe hollow compartment interference and low sampling efficiency in traditional digital detection methods.

[0051] 2. The immunocapture bubbles constructed in this invention, modified with anti-EpCAM antibodies on their surface, achieve highly efficient and specific enrichment of tumor-derived extracellular vesicles, with a capture efficiency of up to 68.77%, significantly higher than traditional ultracentrifugation (<20%) and the ExoQuick kit (<50%). This effectively eliminates interference from non-tumor EVs and the sample matrix, improving the specificity and accuracy of the detection. Furthermore, combined with in-situ ultrasonic lysis technology, the cumbersome steps of RNA extraction and purification are eliminated, significantly simplifying the detection process.

[0052] 3. The fluorescent / DNA dual-coding magnetic bead conjugate designed in this invention utilizes the dual encoding of fluorescent signals and DNA sequences to achieve simultaneous detection of two targets, miR-21 and miR-155. Combined with the cyclic cleavage characteristics of double-stranded specific nucleases, signal amplification can be achieved with only a single enzymatic reaction, eliminating the need for target nucleic acid pre-amplification and avoiding amplification bias. This method achieves a detection sensitivity of up to 5 fM, an EV detection limit as low as the single-particle level, and single-base resolution, accurately distinguishing homologous sequences with single-base mismatches.

[0053] 4. The AI-automated fluorescence counting and decoding module developed in this invention integrates image classification, background removal, target segmentation, and counting functions. Its counting accuracy for dual-coded magnetic beads is highly consistent with manual interpretation, significantly outperforming commercial ImageJ software. This module eliminates the subjective error of manual counting, achieves high-throughput automated decoding of dual-target signals, lowers the operational threshold, and is suitable for rapid detection of large batches of clinical samples.

[0054] 5. The integrated detection system constructed in this invention achieves a fully integrated detection process from EV enrichment, in-situ lysis, enzymatic reaction, self-stratification separation to AI-automated reading. The system eliminates reliance on precision microfluidic equipment and high-resolution imaging systems, requiring only a conventional fluorescence microscope for detection. It is simple to operate, has a short detection cycle, and can be directly adapted to clinical plasma samples. In the clinical diagnosis of lung and prostate cancer, the system achieved an AUC value of 1.00 and a diagnostic accuracy of 100%, providing a reliable technical platform for non-invasive early diagnosis and efficacy monitoring of tumors, and has promising prospects for clinical translation.

[0055] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0056] Figure 1 The flowcharts for detecting extracellular vesicle miRNAs in tumor cells and for AI-automated counting are shown below; where a is a schematic diagram of the process for detecting miRNAs in cancer-derived EVs; and b is a flowchart of the AI-automated counting process.

[0057] Figure 2 Fluorescence characterization images of the hollow glass microbubbles of the present invention, namely HB, HB-PEI, and HB-DBCO (scale bar is 200 μm).

[0058] Figure 3 The image shows the nitrogen peak X-ray photoelectron spectroscopy (XPS) results of HB, HB-PEI, and HB-DBCO of the present invention.

[0059] Figure 4 Figure 1 shows the results of a fluid dynamics simulation of hollow glass microbubbles; where a is the relative velocity distribution around the simulated hollow glass microbubbles in an aqueous solution; and b is the pressure distribution around the simulated hollow glass microbubbles floating in water.

[0060] Figure 5 Figure 1 shows the characterization results of the fluorescent / DNA dual-coding magnetic bead conjugates. Specifically, a is the fluorescence micrograph of FITC-MB-DNA21; b is the fluorescence micrograph of cDNA21-Cy3, complementary to FITC-MB-DNA21; c is the fusion image of FITC-MB-DNA21 and cDNA21-Cy3; d is the fluorescence micrograph of Cy3-MB-DNA155; e is the fluorescence image of cDNA155-Cy5, complementary to Cy3-MB-DNA155; f is the fusion image of Cy3-MB-DNA155 and cDNA155-Cy5; g is the fluorescence intensity distribution along the white dashed line in figure c; h is the fluorescence frequency scatter plot in figure c; i is the fluorescence intensity distribution along the white dashed line in figure f; and j is the fluorescence frequency scatter plot in figure f.

[0061] Figure 6 The figure shows the characterization results of HB-DBCO combined with dual-coded fluorescent magnetic beads; where a is a scanning electron microscope image (SEM) of HB-DBCO linked with dual-coded fluorescent magnetic beads, and b is an EDS elemental analysis result of the product of HB-DBCO combined with dual-coded fluorescent magnetic beads.

[0062] Figure 7The graph shows the motion behavior and kinetics detection results of HB-DBCO after binding to magnetic beads; where a is the force analysis diagram of HB-DBCO after binding to FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3 in solution, b is the time-delayed recording of the motion behavior of HB-MB21 and HB-MB155 in PBS solution, and c is the kinetic measurement results of HB, HB-MB21 and HB-MB155.

[0063] Figure 8 The image shows the fluorescence characterization of IHB in the immune capture bubble; the scale bar is 200 μm.

[0064] Figure 9 Scanning electron microscope (SEM) images of tumor extracellular vesicles captured by IHB; where a is an SEM image of PC-3 extracellular vesicles captured by IHB; b is a magnified view of image a.

[0065] Figure 10 Figure 1 shows the sequence specificity validation results of the detection method; where a) is the specificity detection result of digital assay in distinguishing target miR-21 from non-target sequences (SM, DM, miR-155, miR-1246, miR-21-3p, miR-451a, miR-122, and NC); and b) is the specificity detection result of digital assay in distinguishing target miR-155 from non-target sequences (SM, DM, miR-21, miR-1246, miR-21-3p, miR-451a, miR-122, and NC).

[0066] Figure 11 This is a graph showing the verification results of the anti-interference performance of the present invention;

[0067] Figure 12 This is a graph showing the sensitivity verification results of the detection method; where a represents the control group; b represents the concentration of 10... 3 Fluorescence images of EVs / mL; c represents a concentration of 10. 4 Fluorescence images of EVs / mL; d represents a concentration of 10. 5 Fluorescence image of EVs / mL; e represents a concentration of 10. 6 Fluorescence images of EVs / mL; f represents a concentration of 10. 7 Fluorescence image of EVs / mL; g represents a concentration of 10. 8 Fluorescence image of EVs / mL; h is the standard curve for simultaneous detection of miR-21 and miR-155 in extracellular vesicles by the sensor of this invention; y miR-21 =0.779logC EV -1.554(R 2 =0.998), and y miR-155=0.609logC EV -1.204(R 2 =0.994), and LODs were 254 and 274 particles / mL, respectively;

[0068] Figure 13 Figures show the correlation between EV detection in clinical plasma samples; Figure a is a schematic diagram of the detection process; Figure b is a heatmap of biomarker expression in the corresponding samples; Figure c is a violin image of the miR-21 results detected in 91 clinical EV samples using this method; Figure d is a violin image of the miR-155 results detected in 91 clinical EV samples using this method; and Figure e is a violin image of the SUM results detected in 91 clinical EV samples using this method. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0070] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0071] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0072] like Figure 1 As shown in a, the ultrasensitive digital detection system for extracellular vesicle miRNA of tumor cells in this invention mainly consists of immune capture bubbles, fluorescent / DNA dual-coding magnetic bead conjugates, and multifunctional clickable glass microbubbles; as shown in a diagram. Figure 1 As shown in b, the AI ​​automated counting process of the present invention includes dual-channel image registration, HSV color space conversion and hue filtering, local adaptive threshold segmentation, morphological reconstruction and hole filling, watershed segmentation, shape and size filtering, and cross-color interference removal steps.

[0073] Immunocapture bubbles are used to specifically capture tumor-derived extracellular vesicles from test samples. After capture, the vesicles are lysed in situ by ultrasound to release internal miRNAs. The released miRNAs are mixed with fluorescent / DNA dual-encoded magnetic bead conjugates and double-stranded specific nucleases, resulting in target recognition and signal amplification reactions. After the reaction, unreacted encoded magnetic beads in the system bind to the surface of multifunctional click glass microbubbles through click chemistry and spontaneously float to the liquid surface with the help of the bubble's buoyancy, achieving physical separation from the reaction products. Finally, the reaction magnetic beads that settle at the bottom are collected for fluorescence imaging, and the images are input into an AI automated fluorescence counting and decoding module to achieve automated counting of dual targets miR-21 and miR-155.

[0074] The preparation methods, performance verification, and overall detection methods of the above components will be described in detail below through specific embodiments.

[0075] Example 1:

[0076] Hydroxylation activation of hollow glass microbubbles: Hollow glass microbubbles (HB) were immersed in freshly prepared piranha solution (concentrated sulfuric acid: 30% hydrogen peroxide = 3:1, note the strong corrosiveness) at room temperature for 1 h to complete surface hydroxylation activation and impurity removal; then HB was repeatedly rinsed with ultrapure water until the washing solution was neutral, glass fragments and organic residues were separated and removed using a separatory funnel, and vacuum dried at room temperature for 24 h to obtain purified and activated HB.

[0077] Aminolation modification of HB: Weigh 500 mg of activated HB, add 10 mL of 2 mg / mL PEI (molecular weight 25 kDa) aqueous solution, and react on a rotary mixer at room temperature for 15 min; after the reaction is complete, centrifuge at 4000 rpm for 1 min, discard the lower layer solution, wash 5 times with ultrapure water, and dry under vacuum at 37 ℃ to obtain amino-modified HB-PEI.

[0078] Preparation of click bubbles: Weigh 100 mg HB-PEI, add 2 mL PBS solution, and then add 200 μL of 20 mM dibenzocyclooctyne-N-hydroxysuccinimide (DBCO-NHS) ester solution. Mix and react at room temperature for 1 h. After the reaction is complete, centrifuge at 4000 rpm for 1 min, discard the lower layer solution, wash 5 times with ultrapure water, and vacuum dry at 37 ℃ to obtain self-suspended click glass microbubbles (HB-DBCO) with bioorthogonal function, i.e., the multifunctional click glass microbubbles of this invention. Store in a sealed container at 4 ℃ for later use.

[0079] Fluorescence characterization: Take 200 μg of HB, HB-PEI, and HB-DBCO, add 1 μL of 1 mM Cy3 azide (N3-Cy3) and 199 μL of PBS, react at room temperature in the dark for 10 min, wash three times with 0.01 g / 100 mL Tween 20 phosphate buffer (PBST, containing 0.01 g Tween 20 per 100 mL of phosphate buffer), and observe under a fluorescence microscope. Figure 2 As shown, the HB-DBCO surface exhibits strong fluorescence, while HB and HB-PEI show no obvious fluorescence, proving that the DBCO group was successfully grafted and the click bubble preparation was successful.

[0080] Elemental characterization: Surface elemental analysis of HB, HB-PEI, and HB-DBCO was performed using X-ray photoelectron spectroscopy (XPS), such as... Figure 3The results show that no nitrogen peak appeared in HB, while HB-PEI and HB-DBCO showed obvious N 1s characteristic peaks at 400 eV, providing direct chemical evidence for the successful preparation of click bubbles.

[0081] Physical properties of multifunctional clickable glass microbubbles: The motion behavior of the multifunctional clickable glass microbubbles prepared in this invention in solution was studied by fluid dynamics simulation, and the results are as follows. Figure 4 Place. Figure 4 The results in a and 4b indicate that the fluid velocity is higher near the bubble surface, and there is a pressure difference between the bottom and top of the bubble. This pressure difference is one of the driving forces for the bubble's spontaneous buoyancy. This physical property provides a theoretical basis for the present invention to achieve self-stratification using buoyancy.

[0082] Example 2: Preparation and characterization of fluorescent / DNA dual-coding magnetic bead conjugates:

[0083] Fluorescent labeling of magnetic beads: 20 μL of 10 mg / mL streptavidin (SA) conjugated magnetic beads (produced by Suzhou Zhiyi Microsphere Technology Co., Ltd.) were divided into two groups. 2 μL of 1 mg / mL FITC-NHS and 2 μL of 1 mg / mL Lcy3-NHS fluorescent dye were added to each group, and the mixture was rotated at room temperature in the dark for 60 min. The supernatant was removed by magnetic separation, and the mixture was washed three times with 0.01 g / 100 mL PBST to obtain FITC-MB-SA and Cy3-MB-SA, respectively. The FITC-MB-SA and Cy3-MB-SA were resuspended in PBS and stored at 4 °C.

[0084] Immobilization of clickable DNA probes: Add 100 μL of 10 μM Biotin-DNA21-N3 probe (e.g., SEQ ID NO.1: / 5Biotin / ttttttcaacatcagtctgataagctattttt / 3N3 / ) to FITC-MB-SA, and add 100 μL of 10 μM Biotin-DNA155-N3 probe (e.g., SEQ ID NO.2: / 5Biotin / tttttacccctatcacgattagcattaattttt / 3N3 / ) to Cy3-MB-SA. Make up the volume to 1 mL with PBS and rotate the reaction at room temperature for 30 min. Remove the supernatant by magnetic separation, wash three times with 0.01 g / 100 mL PBST to obtain FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3, which are the fluorescent / DNA dual-coding magnetic bead conjugates. Resuspend them in PBS and store at 4 °C in the dark for later use.

[0085] Characterization and verification of FITC-MB-DNA21 and Cy3-MB-DNA155: The two prepared dual-coding magnetic bead conjugates were reacted with their corresponding complementary sequences Cy3-cDNA21 (e.g., SEQ ID NO.3: / 5Cy3 / tagcttatcagactgatgttga) and Cy5-cDNA155 (e.g., SEQ ID NO.4: / 5Cy5 / ttaatgctaatcgtgataggggt) at room temperature in the dark for 30 min. After washing three times with 0.01 g / 100 mL PBST, the mixtures were observed under a fluorescence microscope. Figure 5 The results showed that multi-channel fluorescence imaging, signal co-localization analysis, and fluorescence intensity statistics comprehensively verified the successful preparation and sequence recognition specificity of the dual-coding magnetic bead conjugates. Figure 5 a and 5b are fluorescence micrographs of FITC-MB-DNA21 and its reaction with the complementary sequence cDNA21-Cy3, respectively. Figure 5 c shows the fluorescence fusion image of the two, which clearly demonstrates the high spatial overlap between the FITC fluorescence coding signal of the magnetic bead itself and the Cy3 fluorescence signal of the complementary sequence. Correspondingly, Figure 5 Images d and 5e are fluorescence micrographs of Cy3-MB-DNA155 and its reaction with the complementary sequence cDNA155-Cy5, respectively. Figure 5 f is the fluorescence fusion image of the two, which also shows a high degree of overlap between the fluorescence coding signal of the magnetic bead itself and the fluorescence signal of the complementary sequence. Based on this, Figure 5 g and 5i are respectively Figure 5 In the fluorescence intensity distribution images shown by the white dashed lines in c and 5f, it can be clearly seen that the intensity peaks of the two fluorescence channels are completely synchronized in position without any obvious shift. Figure 5 h and 5j are respectively Figure 5 The scatter plots of fluorescence frequencies corresponding to c and 5f further verified from a statistical perspective that the signals of the two fluorescence channels are significantly positively correlated. The above results together prove that the clickable DNA probe has been successfully immobilized on the surface of the corresponding fluorescently encoded magnetic beads. Both fluorescent / DNA dual-encoding magnetic bead conjugates have been successfully prepared and can only specifically bind to the corresponding complementary sequences, exhibiting excellent sequence recognition specificity.

[0086] 1 mg HB-DBCO was reacted with excess FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3 at room temperature for 30 min. After washing three times with 0.01 g / 100 mL PBST, the product was dried and stored at 4 °C. Figure 6 The characterization results of the combination of HB-DBCO and dual-encoded fluorescent magnetic beads provide direct dual verification of the click chemistry capture capability of multifunctional clickable glass microbubbles. Figure 6 Image a is a scanning electron microscope (SEM) image of the product of HB-DBCO combined with dual-coding fluorescent magnetic beads. It can be clearly observed that a large number of spherical magnetic beads MB particles are densely attached to the surface of the hollow glass microbubble HB. The microscopic morphology directly confirms that HB-DBCO can effectively bind fluorescent / DNA dual-coding magnetic beads through click chemical reaction. Figure 6 b represents the elemental analysis results of the energy-dispersive X-ray spectroscopy (EDS) corresponding to the bonding product. The detection results not only detected the characteristic silicon (Si) element of hollow glass microbubbles, but also clearly detected the iron (Fe) and phosphate (P) elements unique to magnetic beads. From the elemental composition level, this further confirms the successful bonding of HB and MB. The two characterization results corroborate each other, fully demonstrating that the multifunctional clickable glass microbubbles prepared by this invention have excellent magnetic bead trapping and bonding capabilities, providing core support for the subsequent self-layering separation of unreacted magnetic beads.

[0087] The motility of HB-MB21 (the product of the reaction of FITC-MB-DNA21-N3 with HB-DBCO) and HB-MB155 (the product of the reaction of Cy3-MB-DNA155-N3 with HB-DBCO) in PBS solution was recorded over time. Figure 7 The motion behavior and dynamics of HB-DBCO combined with magnetic beads were tested, comprehensively verifying the buoyancy and self-stratification ability of the multifunctional clickable glass microbubble combined with magnetic beads. Figure 7 Figure a shows the force analysis diagram of HB-DBCO combined with FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3 in solution. It clarifies that the microbubbles are subjected to upward buoyancy, downward gravity and viscous resistance in the liquid phase. It confirms that the HB-MB complex after being combined with magnetic beads still satisfies the condition that the buoyancy is greater than the net external force, and explains the intrinsic mechanism of its spontaneous floating from the perspective of mechanical principles. Figure 7 b is a time-delayed recording of the motion behavior of HB-MB21 and HB-MB155 in PBS solution, which intuitively presents the floating process of microbubbles after binding with magnetic beads in a static state. It can be clearly observed that most of the microbubbles bound with magnetic beads can float to the upper layer of the liquid surface within 5 minutes, directly verifying their practical application effect of rapid floating. Figure 7Figure c shows the floating dynamics measurement results of HB, HB-MB21, and HB-MB155, quantitatively demonstrating the change in floating efficiency of the three types of microbubbles over time. The results show that HB-MB21 and HB-MB155, after being combined with magnetic beads, exhibit enhanced floating ability compared to the blank HB due to changes in surface properties. Quantitative detection showed that after standing for 5 minutes, the floating efficiencies of HB, HB-MB21, and HB-MB155 reached 74.43%, 87.53%, and 86.40%, respectively. This quantitatively confirms that even when combined with unreacted magnetic beads, the multifunctional microbubbles still possess excellent and efficient floating performance, providing reliable data support for the stable realization of self-stratification separation in the detection system of this invention.

[0088] Example 3: Preparation and characterization of immune capture bubbles:

[0089] Silanization modification of HB: Weigh 0.5 g of the activated HB from Example 1, add 10 mL of anhydrous ethanol solution containing 0.5 mL of GPTES, and mix by rotation at room temperature overnight; after the reaction is complete, centrifuge at 4000 rpm for 1 min, discard the lower layer solution, wash 3 times with ultrapure water, and dry under vacuum at 37 ℃ to obtain epoxy-modified HB-GPTES.

[0090] Antibody conjugation: Weigh 0.5 g HB-GPTES, add 5 mL PBS solution, then add 0.5 mL 100 μg / mL anti-EpCAM antibody, and mix by rotation at room temperature for 1 h. After the reaction is complete, centrifuge at 4000 rpm for 1 min, discard the lower layer solution, wash 3 times with PBS, block with 2 g / 100 mL BSA solution (containing 2 g bovine serum albumin per 100 mL solution) at room temperature for 30 min, wash and dry again to obtain immunocapture bubble IHB, and store at 4 ℃ for later use.

[0091] Characterization and Validation:

[0092] Characterization of EpCAM antibody modification: After reacting HB-GPTES with EpCAM antibody, the reaction was followed by reaction with a secondary antibody labeled with Cy3 fluorescent dye (secondary antibody-Cy3). The results were then observed under a fluorescence microscope. Figure 8 As shown, the EpCAM group exhibits strong fluorescence on its surface, while the group without EpCAM shows no obvious fluorescence, proving that the EpCAM antibody modification was successful.

[0093] EV capture performance validation: After incubating immune capture bubbles (IHB) with EpCAM-positive EVs derived from PC-3 cells, as shown... Figure 9 As shown. Figure 9The scanning electron microscopy (SEM) characterization results of the IHB used to capture tumor extracellular vesicles visually validate the highly efficient and specific capture ability of the IHB prepared in this invention for EpCAM-positive tumor EVs. Figure 9 Image a is a panoramic scanning electron microscope image of IHB after incubation with EpCAM-positive EVs derived from PC-3 cells. The complete basal structure of the immune-capturing bubble IHB can be clearly observed, and a large number of vesicular particles are widely attached to its surface. The overall field of view intuitively presents the capture effect of IHB on tumor EVs. Figure 9 b is Figure 9 The magnified image of the corresponding region clearly shows that the particles bound to the IHB surface have the typical spherical vesicle morphology of tumor extracellular vesicles, confirming that these bound substances are the target captured EpCAM-positive tumor EVs, further corroborating the capture results from a microscopic detail level. The mutual corroboration between the two SEM images, from panoramic to local, fully demonstrates that the immune capture bubble IHB constructed in this invention can efficiently and specifically capture EpCAM-positive tumor EVs, providing direct morphological experimental evidence for the tumor-derived EV-specific enrichment step in the detection system.

[0094] Capture efficiency test: Using nanoparticle tracking analysis (NTA), the capture efficiency of immune capture bubble (IHB) for tumor EVs reached 68.77%, which is significantly better than ultracentrifugation and commercial ExoQuick kit.

[0095] Example 4: Construction and Verification of an AI-Automated Fluorescence Counting and Decoding System

[0096] System Architecture: An AI-automated fluorescence counting and decoding module (AI Counter) was developed using Python. This module incorporates an image processing workflow to achieve high-precision, automated counting and dual-target signal decoding of fluorescence images in the FITC channel (miR-21) and Cy3 channel (miR-155). Specifically, it includes the following processing steps:

[0097] 1) For the FITC channel and Cy3 channel fluorescence images acquired by the inverted fluorescence microscope, the subpixel translation offset between the two images is calculated using a phase correlation algorithm based on fast Fourier transform. The images are then corrected according to the offset to ensure that the two channel images are precisely aligned and that the magnetic beads in the same field of view have consistent coordinates in the two channels.

[0098] 2) Convert the registered FITC and Cy3 images from the RGB color space to the HSV color space respectively, and extract the hue (H), saturation (S), and lightness (V) components; according to the pre-calibrated characteristic hue range of FITC and Cy3 fluorescent dyes, perform hue matching on each pixel to generate a binary mask, and initially screen out the pixel areas that may be magnetic beads in order to eliminate noise interference from non-target colors in the background.

[0099] 3) Within the mask area after hue screening, the lightness component V is segmented locally using the Bernson method based on the mean and standard deviation of the pixel neighborhood to obtain a binary image of the magnetic bead candidate area, in order to adapt to uneven illumination and changes in background fluorescence.

[0100] Specifically, a Bernson improved algorithm based on dynamic windowing and pixel robust estimation is used for local adaptive threshold segmentation of the brightness component V. This algorithm addresses the problem of gray-level differences between the edges and centers of magnetic beads and interference from local sporadic fluorescence spectra in fluorescence microscopy, and achieves accurate segmentation through the following steps:

[0101] First, construct an initial window of size w×w centered on the current pixel (w is set to 1.5 times the average diameter of the magnetic bead based on the bead size, ensuring that the window contains both the magnetic bead and the background area).

[0102] Secondly, the top 5% and bottom 5% of pixels with the highest grayscale values ​​are removed from the window (i.e., truncated mean processing) to eliminate the interference of fluorescent noise or dark spots inside the magnetic beads on the extreme value calculation.

[0103] Then, calculate the mean gray value μ and standard deviation σ of the remaining pixels after truncation, and set the local threshold T. local Set to μ-α·σ (α is an adjustable coefficient, ranging from 0.2 to 0.5), this formula is based on robust statistical theory and uses the standard deviation to dynamically reflect the degree of local gray-level fluctuation. When the window contains the edge of the magnetic bead, the gray-level fluctuation increases (σ increases), and the threshold is adaptively reduced to preserve edge details. When the window is located in a uniform background area, σ approaches 0, and the threshold approaches the mean, avoiding background missegmentation.

[0104] Finally, the grayscale value I of the center pixel is compared with T. local The comparison completes the binarization;

[0105] This improved algorithm effectively solves the inherent problems of the classic Bernson algorithm, which is sensitive to noise due to its reliance on a single maximum and minimum value and is prone to breakage or expansion at the edge of the magnetic bead.

[0106] 4) Perform morphological opening operations to remove small noise and closing operations to connect neighboring regions on the binary image in sequence. Then, use morphological reconstruction algorithms to fill the holes inside the magnetic beads caused by uneven fluorescence to obtain the complete magnetic bead connected region.

[0107] The specific steps for filling the pores inside the magnetic beads caused by uneven fluorescence using a seed-label-based morphological reconstruction algorithm are as follows:

[0108] First, Euclidean distance transformation is performed on the binary image after opening and closing operations to calculate the distance from each foreground pixel to the nearest background pixel, generating a distance map D(x,y). The distance value reflects the "depth" of the pixel inside the magnetic bead - the distance value is large in the central region of the magnetic bead, and the distance value is small near the edge and the hole.

[0109] Then, local maxima are searched on the distance map, i.e., points whose pixel distance value is greater than the distance values ​​of all pixels in its 8-neighborhood, and these maxima are extracted as candidate seeds; to eliminate spurious maxima caused by noise, a distance threshold T is set. d =0.4×R avg , where R avg The average radius of the magnetic beads (pre-calibrated based on the bead size) is used, retaining only distance values ​​greater than T. d The local maxima are used as effective seed points to ensure that the seed points are located in the solid core region of the magnetic bead rather than near the edge or holes.

[0110] Subsequently, the binary image formed by these seed points is used as the marker image, and the original binary image (after opening and closing operations) is used as the mask image for morphological reconstruction: the marker is repeatedly expanded by 3×3 structuring elements, and the intersection with the mask is taken after each expansion until the image no longer changes; during the reconstruction process, the marker points grow outward from the seed position, but are limited by the contour of the mask. Eventually, each seed will expand to the entire connected region of the magnetic bead in which it is located, while the holes inside the magnetic bead (areas that are the background in the mask but surrounded by the foreground) will be filled by the foreground pixels flowing in from all sides during the marker expansion process, thus eliminating the holes;

[0111] The algorithm uses distance transformation to automatically locate the reliable core region of each magnetic bead as the starting point for reconstruction, avoiding the tedious manual selection of markers, and ensures the stability of seed points through threshold screening; morphological reconstruction strictly maintains the original shape of the magnetic bead, only filling the internal voids without changing the magnetic bead boundary, providing a complete and void-free connected domain for subsequent adhesion segmentation.

[0112] 5) Perform distance transformation on the connected components, calculate the Euclidean distance from each pixel to the background, and generate a distance map; search for local maxima on the distance map as seed points, and use the watershed algorithm to segment the adhered magnetic beads to obtain the independent regions of individual magnetic beads, thus solving the counting bias caused by magnetic bead aggregation.

[0113] 6) Calculate the area, perimeter, roundness, and aspect ratio of each independent region. Set a threshold based on the actual particle size of the magnetic bead and the imaging magnification. Remove regions that do not conform to the roundness of the magnetic bead or have abnormal size, and filter out long strip regions that are still stuck together due to incomplete segmentation.

[0114] More specifically, to address the issue of magnetic beads having a uniform actual particle size (10-30 μm) but potentially leaving behind adhered areas (elongated, dumbbell-shaped) and interference from impurities after imaging due to incomplete segmentation, the following steps are used for precise filtering:

[0115] First, based on microscope calibration parameters (such as pixels per micrometer p) and the physical particle size range of the magnetic beads [d] min ,d max ], calculate the effective range of pixel area [A min A max ]=[π·(d min / 2) 2 ·p 2 , π·(d max / 2) 2 ·p 2 Simultaneously calculate the pixel range of the equivalent circle diameter [D]. min D max ]=[d min ·p,d max ·p]. This absolute scale threshold is adaptively bound to imaging conditions, eliminating biases at different magnifications or resolutions;

[0116] Then, for each connected component Ω obtained after partitioning, the following geometric features are calculated:

[0117] Area A = |Ω|;

[0118] Perimeter P (calculated using eight-neighbor chain code);

[0119] Find the longest side L and the shortest side W of the smallest bounding rectangle, and obtain the aspect ratio R = L / W;

[0120] The convex hull area (calculated using the Graham scan algorithm);

[0121] Solidity = A / HullArea;

[0122] Circularity = 4πA / P 2 ;

[0123] Next, a two-level screening strategy is designed:

[0124] Level 1 (Fundamental Physical Constraints): If or If any of these are found, they are directly identified as non-target areas (impurities or large pieces of adhesion) and removed.

[0125] Level 2 (Fine Shape Recognition): For regions that have passed through Level 1, calculate their convex hull depth. Specifically: extract the contour point set, calculate the shortest distance from each point on the contour to the convex hull, and obtain a distance sequence; set a distance threshold T. dep =0.15×D min (Take 15% of the smallest particle size as the indentation tolerance), statistical distance greater than T dep Number of contour points N dep If N dep If R > 0 and these points are continuously distributed and span an angle range exceeding 90°, then the area is determined to have obvious depressions and belongs to unsegmented adhered magnetic beads, and is therefore rejected. At the same time, if R > 1.8 or Solidity < 0.85 or Circularity < 0.6, it is also considered to have abnormal shape (elongated shape or over-segmented residue) and is rejected. This multi-parameter joint threshold is set based on the prior knowledge that magnetic beads are approximately circular. The depression depth analysis can sensitively capture the depression features of dumbbell-shaped adhered areas, making up for the shortcomings of a single circularity index.

[0126] This screening scheme transforms the physical size prior into a pixel-domain adaptive threshold and combines it with convex hull and concavity detection to distinguish individual magnetic beads from aggregates from a geometrical perspective. This ensures size compatibility and accurately eliminates false positive targets caused by incomplete segmentation.

[0127] 7) Count the number of filtered magnetic beads in the FITC channel and Cy3 channel respectively to obtain the counting results of miR-21 and miR-155; at the same time, if a magnetic bead is detected in both channels, it is determined to be the same target based on its coordinates and is regarded as cross-color interference and removed to ensure the specificity and accuracy of dual-target counting.

[0128] This module integrates functions such as image classification, grayscale conversion, adaptive binarization, image segmentation, and adaptive adjustment of HSV parameters. It can achieve automated and unbiased counting of dual-coded magnetic beads and decoding of dual-target signals. The counting accuracy is highly consistent with that of manual counting and is significantly better than commercial software.

[0129] In fluorescence microscopy imaging analysis, subpixel shift between dual-channel images, background variations caused by uneven illumination, holes formed by uneven fluorescence distribution of the magnetic beads themselves, magnetic bead aggregation and adhesion, and cross-color interference between different channels can all severely affect the accuracy of counting. To solve these problems, this module first uses a phase correlation algorithm to accurately register the FITC and Cy3 channel images; then, it uses hue prior information in the HSV color space to generate a mask, effectively eliminating background noise; next, it obtains complete magnetic bead connected regions through local adaptive thresholding, morphological reconstruction, and hole filling; for adhered magnetic beads, it uses distance transform and watershed algorithms for segmentation to avoid missed counts caused by aggregation; finally, it filters based on the shape and size characteristics of the magnetic beads and removes cross-color interference, thereby achieving high-precision and high-reliability automated counting of miR-21 and miR-155.

[0130] Performance Verification: The FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3 dual-encoding magnetic bead conjugates prepared in Example 2 were selected. After slide preparation, representative fluorescence images covering low, medium, and high density gradients were randomly collected from three independent preparation batches using an inverted fluorescence microscope, totaling 15 images. The AI-automated fluorescence counting and decoding module (AI Counter) constructed in this invention, manual counting, and ImageJ software were used to count the magnetic beads in the same image. The results showed that the counting results of the AI ​​Counter were highly consistent with those of manual counting (Pearson correlation coefficient R). 2 =0.998), with an average counting deviation of less than 5%, and a counting accuracy significantly better than ImageJ software (ImageJ vs. manual counting R). 2 (Only 0.85). This module, through built-in algorithms such as phase correlation registration, HSV tone filtering, local adaptive threshold segmentation, morphological reconstruction, watershed segmentation, and shape filtering, can effectively eliminate counting errors caused by image offset, background noise, magnetic bead adhesion, and crosstalk interference, and achieve unbiased and automated decoding of miR-21 and miR-155 dual target signals.

[0131] Example 5: A self-stratified digital detection method for extracellular vesicle miRNA quantification

[0132] Based on the multifunctional clickable glass microbubbles (HB-DBCO), fluorescent / DNA dual-encoding magnetic bead conjugates (FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3), immune capture bubbles (IHB), and AI-automated fluorescence counting and decoding module (AI Counter) prepared above, an ultrasensitive digital detection method for tumor EV miRNA was established. The specific steps are as follows:

[0133] 1. Tumor EV capture and in situ lysis:

[0134] Take the sample to be tested (cell culture supernatant or human plasma), add the immunocapture bubble IHB prepared according to Example 3, and incubate at room temperature for 40 min (preferably 30-60 min) to specifically capture EpCAM-positive tumor EVs in the sample; after incubation, centrifuge at 1000 rpm for 1 min, discard the supernatant, and wash 3 times with phosphate-buffered saline (PBS) to obtain IHB enriched with tumor EVs; use an ultrasonic cell disruptor to perform in situ ultrasonic lysis of the IHB enriched with EVs to release miRNA inside the vesicles. The lysis parameters are: frequency 20 kHz, 6 s of sonication per cycle, 10 s pause, 5 cycles; after lysis, centrifuge at 1000×g for 5 min, and take the supernatant as the miRNA sample to be tested.

[0135] 2. DSN enzyme-mediated cyclic signal amplification reaction:

[0136] Prepare a 20 μL enzyme digestion reaction system, including: 6 μL of equimolarly mixed dual-coding magnetic bead conjugates (FITC-MB-DNA21-N3 and Cy3-MB-DNA155-N3), 10 μL of the miRNA sample to be tested, 1 μL of 0.5 U double-stranded specific nuclease (DSN), 1 μL of 10×DSN reaction buffer, 0.5 μL of 20 U RNase inhibitor, and 1.5 μL of DEPC-treated water; after mixing, incubate at 40 ℃ for 50 min to complete the target miRNA-mediated DNA probe cyclic digestion reaction; immediately after incubation, add DSN termination solution (2 μL of 50 mM EDTA) to terminate the reaction, collect the magnetic beads by magnetic separation, and wash three times with 0.01 g / 100 mL Tween 20 phosphate buffer (PBST).

[0137] 3. Click on chemical reactions and self-layering separation:

[0138] The collected magnetic beads were resuspended in 60 μL of 5 μg / μL HB-DBCO solution (prepared according to Example 1) and incubated at room temperature for 30 min by rotation. The unreacted magnetic beads were specifically captured by a catalyst-free click chemistry reaction between the dibenzocyclooctylene (DBCO) groups grafted onto the HB-DBCO surface and the azide (N3) groups on the surface of the unreacted magnetic beads. After incubation and standing, the HB-DBCO, with its low-density hollow structure (density 0.5-0.7 g / cm³), effectively combined with the unreacted magnetic beads. 3 The magnetic beads quickly float to the surface of the liquid, while the magnetic beads after the reaction (whose surface DNA probes have been cut and lost their azide groups) settle to the bottom of the tube, completing the self-separation. Collect the magnetic beads that have settled at the bottom of the tube and dilute them with PBS to a suitable concentration for later use.

[0139] 4. Fluorescence Imaging and AI-Automated Counting:

[0140] The processed magnetic bead suspension was dropped onto a glass slide, and fluorescence images of the FITC channel (corresponding to miR-21) and Cy3 channel (corresponding to miR-155) were acquired using an inverted fluorescence microscope. The acquired images were then imported into the AI ​​automated fluorescence counting and decoding module (AI Counter). This module has a built-in image processing algorithm that sequentially performs dual-channel image registration, HSV color space conversion and hue filtering, local adaptive threshold segmentation, morphological reconstruction and hole filling, watershed segmentation, shape and size filtering, and crosstalk interference removal. Finally, it outputs the automated counting results of miR-21 and miR-155 and the dual-target signal decoding data.

[0141] Example 6: Analytical performance verification of the detection system:

[0142] Specificity test: Target miRNAs (miR-21, miR-155) at a concentration of 10 pM, as well as homologous single-base mismatch sequences (SM), double-base mismatch sequences (DM), non-target miRNAs (miR-451a, miR-1246, etc.) and random sequences (NC) were selected as controls and detected according to the detection method described in Example 5 of this invention to examine the sequence recognition specificity of this method. Figure 10 To verify the sequence specificity of the detection method of this invention, the recognition specificity and sequence resolution capability of the method for the target miRNA were comprehensively evaluated. Figure 10 a) shows the specificity of the detection method in distinguishing the target miR-21 from various non-target sequences. The results indicate that only the target miR-21 can generate a significant magnetic bead counting signal, while homologous single-base mismatch sequences (SM), double-base mismatch sequences (DM), other non-target miRNAs (miR-155, miR-1246, miR-21-3p, miR-451a, miR-122), and random negative control sequences (NC) do not produce significant counting signals. Correspondingly, Figure 10 b shows the specific detection results of the detection method in distinguishing the target miR-155 from various non-target sequences. Similarly, only the target miR-155 can produce a significant positive count signal. Single-base mismatch sequences, double-base mismatch sequences, non-target miRNAs, and random negative sequences do not produce obvious effective signals. The two sets of parallel validation results jointly show that the detection method of the present invention can only react specifically with perfectly matched target miRNAs. It can accurately distinguish homologous sequences with only single-base differences, has excellent single-base resolution, and can effectively avoid false positive results caused by cross-reaction of homologous sequences. It provides a reliable guarantee of specificity for the accurate quantitative detection of target miRNAs in complex clinical biological samples.

[0143] The sequence of miR-21 is shown in SEQ ID NO.5;

[0144] The sequence of miR-155 is shown in SEQ ID NO.6;

[0145] The sequence of miR-21-3p is shown in SEQ ID NO.7;

[0146] The sequence of miR-1246 is shown in SEQ ID NO.8;

[0147] The sequence of miR-451a is shown in SEQ ID NO.9;

[0148] The sequence of miR-122 is shown in SEQ ID NO.10;

[0149] The sequence of SM-21 is shown in SEQ ID NO.11;

[0150] The sequence of DM-21 is shown in SEQ ID NO.12;

[0151] The sequence of SM-155 is shown in SEQ ID NO.13;

[0152] The sequence of DM-155 is shown in SEQ ID NO.14;

[0153] The sequence of NC is shown in SEQ ID NO.15.

[0154] Anti-interference ability test: To evaluate the anti-interference ability of the detection method of the present invention in complex biological sample matrices, fetal bovine serum with extracellular vesicles removed (EV-free FBS) was used as a complex matrix model. Samples containing 5 g / 100 mL and 10 g / 100 mL EV-free FBS were prepared, respectively, and mixed with the same concentration of tumor EV lysis buffer as detection samples; samples of the same concentration prepared with phosphate-buffered saline (PBS) were used as controls. The detection was performed according to the method of Example 5 of the present invention, and the results are as follows. Figure 11 As shown, in complex matrices containing 5 g / 100 mL and 10 g / 100 mL of EV-depleted FBS, the counting signals of miR-21 and miR-155 showed no significant difference compared to the PBS control group (P>0.05), indicating that proteins, lipids, and other biomolecules in serum did not significantly interfere with the detection system. This result demonstrates that the method of the present invention has excellent resistance to matrix interference and can be directly applied to the detection of complex clinical samples such as human plasma without the need for complex pretreatment steps.

[0155] Sensitivity testing: To evaluate the analytical sensitivity of the detection method of this invention, gradient concentrations of PC-3 cell-derived tumor extracellular vesicles (EVs), with a concentration range of 10, were used. 3 ~10 8 Using particles / mL as the detection model, the expression levels of miR-21 and miR-155 were detected respectively according to the method of Example 5 of the present invention. Figure 12 To verify the sensitivity of the detection method of this invention, parallel detection of tumor EV samples at gradient concentrations was conducted to comprehensively evaluate the detection sensitivity and quantitative performance of the method. Figure 12 a is the fluorescence microscopy image of the control group, where there is almost no effective positive fluorescent magnetic bead signal in the field of view, providing a stable negative baseline for the detection system; Figure 12 b to Figure 12 g represents EV concentration 10 3 Particle count / mL, 10 4 Particle count / mL, 10 5 Particle count / mL, 10 6 Particle count / mL, 10 7 Particle count / mL, 10 8 The fluorescence microscopy images corresponding to the particle count / mL sample allow for a direct observation that as the EV concentration gradient increases, the number of positive fluorescent magnetic beads in the field of view shows a significant increasing trend, from a small amount of discrete fluorescence signal at low concentrations to a dense and uniform fluorescence signal at high concentrations, directly demonstrating the positive correlation between the detection signal and the EV concentration. Figure 12 h represents the standard curve for the simultaneous detection of miR-21 and miR-155 in extracellular vesicles using this method. The results show that with increasing EV concentration, the counting signals of both miR-21 and miR-155 exhibit a good logarithmic linear relationship, and the corresponding standard curve equations are y miR-21 =0.779logC EV -1.554(R 2 =0.998) and y miR-155 =0.609logC EV -1.204(R 2 =0.994), and calculations show that the limits of detection (LODs) for miR-21 and miR-155 are as low as 254 particles / mL and 274 particles / mL, respectively, reaching the level of single-particle EV detection, which is significantly better than traditional RT-qPCR and commercial kit methods. The above-mentioned intuitive results from fluorescence microscopy imaging and complete validation data from the quantitative standard curve collectively demonstrate that the detection method of this invention possesses extremely high detection sensitivity and excellent quantitative linearity, meeting the precise quantification needs of low-abundance EV miRNAs in clinical samples.

[0156] Example 7: Validation of tumor EV miRNA detection in clinical plasma samples

[0157] 1. Clinical Sample Collection and Ethical Statement:

[0158] This study, approved by the ethics committee of the cancer hospital, included 91 clinical plasma samples, including: 14 healthy controls (HC), 10 patients with benign prostatic lesions (BPL), 29 patients with prostate cancer (PCa), 12 patients with benign lung lesions (BLC), and 26 patients with lung cancer (LuC). All participants signed informed consent forms, and sample collection and processing were strictly carried out in accordance with standard operating procedures.

[0159] 2. Clinical sample testing:

[0160] Using the ultrasensitive digital detection method for tumor EV miRNA described in Example 5 of this invention, EVmiR-21 and EV miR-155 in all 91 clinical plasma samples were simultaneously detected. Three replicate wells were set for each sample, and the average value was taken as the final count result.

[0161] 3. Test Results and Statistical Analysis:

[0162] Figure 13 To validate the detection method of this invention for EV miRNA detection in clinical plasma samples, the application performance, target biomarker efficacy, and tumor diagnostic value of this method in real clinical biological matrices were systematically evaluated. The results of each subfigure are described below:

[0163] Figure 13 Figure 'a' shows the detection process for clinical plasma samples, which fully presents the entire clinical application path of this method, from loading clinical plasma samples, tumor EV-specific enrichment and in situ lysis, target miRNA signal amplification reaction, to the final AI-automated fluorescence counting and result output. It intuitively demonstrates the core advantages of this method: simple operation, no need for complex sample pretreatment, and adaptability to routine clinical testing scenarios.

[0164] Figure 13 b shows a heatmap of miR-21 and miR-155 dual-target expression in 91 enrolled clinical samples. This allows for a clear stratification of samples among healthy controls, patients with benign lesions, and patients with malignant tumors. Among them, malignant tumor samples from patients with prostate cancer (PCa) and lung cancer (LuC) showed significant high expression of both targets, forming a clear expression boundary with healthy controls and benign lesion samples, which intuitively reflects the potential of dual targets in tumor identification.

[0165] Figure 13 c. Figure 13d represents violin plots showing the results of miR-21 and miR-155 detection in 91 clinical EV samples, visually presenting the differences in the expression distribution of the two targets among different groups. It can be observed that the overall target expression level in the malignant tumor group is significantly higher than that in the benign lesion group and the healthy control group, while the expression distribution in the benign lesion group and the healthy control group is highly similar. Figure 13 e is a violin plot of the SUM values ​​of miR-21 and miR-155 in 91 clinical EV samples. The results show that the combined detection of the two targets can further widen the expression difference between the malignant tumor group and the control group, significantly improve the discrimination between different groups, and fully demonstrate the performance advantages of the combined detection of the two targets.

[0166] Based on the above visualization results, further systematic statistical analysis was conducted to verify them as follows:

[0167] 1. Statistical validation of target expression differences: The results showed that the expression levels of miR-21 and miR-155 in plasma EV of cancer patients (PCa group and LuC group) were significantly higher than those of the corresponding benign lesion groups (BPL group and BLC group) and healthy control group (HC group) (P<0.0001), while there was no significant difference between the benign lesion group and the healthy control group. Statistically, this confirms that EV miR-21 and miR-155 can serve as reliable potential markers for distinguishing between malignant tumors and benign diseases.

[0168] 2. Clinical diagnostic efficacy evaluation: Receiver operating characteristic (ROC) curve analysis was conducted based on the combined detection data of the two biomarkers. The results showed that the area under the curve (AUC) for distinguishing between lung cancer patients (LuC) and patients with benign lung lesions (BLC) reached 1.00, and the AUC for distinguishing between prostate cancer patients (PCa) and patients with benign prostatic lesions (BPL) also reached 1.00, with a diagnostic accuracy of 100%. The diagnostic efficacy was significantly better than that of a single biomarker, confirming that the combined detection of two targets can significantly improve the accuracy and reliability of tumor diagnosis.

[0169] 3. Validation of Tumor Type Identification Ability: Principal Component Analysis (PCA) was used to reduce the dimensionality of the detection data of three groups of samples: lung cancer, prostate cancer, and benign controls. The dimensionality-reduced principal components were used as features to input a Support Vector Machine (SVM) classifier for modeling. After leave-one-out cross-validation, the results showed that the three groups of samples exhibited a clear clustering distribution in the PCA score map. The SVM classification accuracy reached 100%, which can achieve accurate classification of different tumor types and benign controls, further validating the clinical application potential of the detection system of this invention in tumor type identification.

[0170] In summary, the clinical sample validation results demonstrate that the tumor EV miRNA ultrasensitive digital detection system constructed in this invention can achieve efficient enrichment of tumor-derived EVs and accurate quantification of dual targets miR-21 / miR-155 in complex plasma matrices. It can effectively distinguish between patients with malignant tumors, individuals with benign lesions, and healthy controls. The diagnostic accuracy of the dual-target combined detection reaches 100%, and the AUC value reaches 1.00. It has excellent clinical application value and translational prospects in non-invasive early diagnosis of tumors, efficacy monitoring, and recurrence early warning.

[0171] It should be understood that the miR-21 and miR-155 detection sequences specifically used in the embodiments of the present invention are merely illustrative examples and are not intended to limit the present invention. Those skilled in the art can replace the miRNA sequences with the corresponding sequences of other miRNAs using the method described in this invention, depending on the actual detection target, and can still achieve ultrasensitive digital detection.

[0172] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for the self-stratification digital detection of extracellular vesicle miRNA, characterized in that, Includes the following steps: S1: Provide the sample to be tested, add immunocapture bubbles, incubate for 10-60 min, then centrifuge at 500-2000 rpm for 0.5-1.5 min to remove the supernatant, wash 3-5 times with phosphate buffer to obtain immunocapture bubbles enriched with tumor cell extracellular vesicles; the immunocapture bubbles are hollow glass microbubbles with anti-EpCAM antibodies coupled to their surface; The preparation of immune capture bubbles includes the following steps: Weigh 0.1-1 g of hydroxylated activated hollow glass microbubbles, add 5-15 mL of anhydrous ethanol solution containing 0.1-1 mL of (3-(2,3-epoxypropoxy)propyl)triethoxysilane, and carry out silanization modification reaction for 10-20 h. After centrifugation at 3000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with ultrapure water, and vacuum dry at 35-40 ℃ to obtain epoxy hollow glass microbubbles. Weigh 0.1-1 g of epoxy hollow glass microbubbles, add 1-10 mL of phosphate buffer, add an anti-EpCAM antibody solution with a concentration of 80-120 μg / mL, mix and react for 0.5-1.5 h, centrifuge at 1000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with phosphate buffer, block with bovine serum albumin, wash again and dry to obtain immunocapture bubbles with surface-coupled anti-EpCAM antibody; S2: Immunocapture bubbles enriched with tumor cell extracellular vesicles are subjected to in situ ultrasonic lysis to release miRNAs within the vesicles. After centrifugation at 500-2000 rpm for 3-10 min, the supernatant is collected to obtain the miRNA sample to be tested. S3: The miRNA sample to be tested is mixed with fluorescent-DNA dual-coding magnetic bead conjugates, double-stranded specific nuclease, double-stranded specific nuclease reaction buffer, RNase inhibitor, and DEPC-treated water to form an enzyme digestion system. Enzyme digestion is performed by incubation. After incubation, double-stranded specific nuclease stop solution is added to terminate the reaction. Magnetic beads are collected by magnetic separation and washed 3-5 times with PBS. The fluorescent-DNA dual-coding magnetic bead conjugates include FITC-MB-DNA21-N3 conjugate and Cy3-MB-DNA155-N3 conjugate, which are used to detect miR-21 and miR-155, respectively. The preparation of fluorescent-DNA dual-coding magnetic bead conjugates includes the following steps: Streptavidin-conjugated magnetic beads were mixed with fluorescein isothiocyanate-succinimide ester and Cy3-succinimide ester, respectively, and rotated at room temperature in the dark for 30-90 min to perform fluorescent labeling. The supernatant was removed by magnetic separation, and the beads were washed 3-5 times with 0.01 g / 100 mL Tween 20 phosphate buffer to obtain FITC-labeled magnetic beads and Cy3-labeled magnetic beads, respectively. Biotin-modified clickable DNA probe Biotin-DNA21-N3 was mixed with FITC-labeled magnetic beads and immobilized by the affinity interaction between biotin and streptavidin to obtain the FITC-MB-DNA21-N3 conjugate; wherein the sequence of Biotin-DNA21-N3 is shown in SEQ ID NO.1; Biotin-modified clickable DNA probe Biotin-DNA155-N3 was mixed with Cy3-labeled magnetic beads and immobilized by the affinity interaction of biotin and streptavidin to obtain Cy3-MB-DNA155-N3 conjugate; wherein the sequence of Biotin-DNA155-N3 is shown in SEQ ID NO.2; S4: The magnetic beads collected in step S3 are resuspended in a multifunctional click glass microbubble solution and incubated to allow the unreacted fluorescent-DNA dual-encoding magnetic bead conjugate to bind to the surface of the multifunctional click glass microbubble through a click chemical reaction of azide and dibenzocyclooctylene; the multifunctional click glass microbubble is a hollow glass microbubble with dibenzocyclooctylene groups grafted onto its surface. The preparation of multifunctional clickable glass microbubbles includes the following steps: Hydroxylation activation treatment was performed on the microbubbles of insulating glass to obtain activated insulating glass microbubbles; 100-1000 mg of activated hollow glass microbubbles were added to 10-50 mL of polyethyleneimine aqueous solution with a mass-to-volume ratio of 1-5 mg / mL. The mixture was rotated and reacted at room temperature for 10-20 min to form an aminated coating on the surface by electrostatic adsorption. After centrifugation at 3000-5000 rpm for 0.5-2 min, the mixture was washed 3-5 times with ultrapure water and dried under vacuum at 30-40 °C to obtain aminated hollow glass microbubbles. 10-1000 mg of aminated hollow glass microbubbles were added to 0.5-20 mL of phosphate buffer, followed by 100-300 μL of a 10-30 mM dibenzocyclooctyne-N-hydroxysuccinimide ester solution. The mixture was rotated and reacted at room temperature for 0.5-1.5 h to graft dibenzocyclooctyne groups via a covalent reaction between the surface amino group and the active ester. After centrifugation at 3000-5000 rpm for 0.5-2 min, washing with ultrapure water 3-5 times, and vacuum drying at 30-40 °C, multifunctional click glass microbubbles with bioorthogonal function were obtained. S5: Allow the liquid to stand so that the multifunctional click glass microbubbles, which incorporate unreacted magnetic beads, float to the surface and collect the magnetic beads that have settled at the bottom. S6: Perform fluorescence imaging on the collected magnetic beads, acquire fluorescence images of the FITC channel and Cy3 channel respectively, input the images into the AI ​​automated fluorescence counting and decoding module for counting, and obtain the detection results of miR-21 and miR-155.

2. The method for quantifying extracellular vesicle miRNA using bubble self-stratification digital detection according to claim 1, characterized in that, In step S2, the parameters for in-situ ultrasonic lysis are: frequency 10-30 kHz, each ultrasonic session lasting 5-10 s, followed by a pause of 5-20 s, and repeated 4-6 times.

3. The method for quantifying extracellular vesicle miRNA using bubble self-stratification digital detection according to claim 1, characterized in that, In step S3, the enzyme digestion system has a volume of 20 μL and includes: 6 μL of equimolarly mixed fluorescent-DNA dual-coding magnetic bead conjugate, 10 μL of the miRNA sample to be tested, 1 μL of 0.5U DSN enzyme, 1 μL of 10×DSN reaction buffer, 0.5 μL of 20URNase inhibitor, and 1.5 μL of DEPC-treated water; the incubation temperature is 40-60 ℃, and the time is 0.5-1.5 h.

4. The method for quantifying extracellular vesicle miRNA using bubble self-stratification digital detection according to claim 1, characterized in that, The multifunctional clickable glass microbubbles have a particle size of 10-30 μm and a density of 0.5-0.7 g / cm³. 3 The floating time in phosphate buffer should not exceed 5 minutes.

5. The method for quantifying extracellular vesicle miRNA using bubble self-stratification digital detection according to claim 1, characterized in that, The AI-automated fluorescence counting and decoding module incorporates image classification, grayscale, adaptive binarization, aspect ratio threshold-based image segmentation, and HSV parameter adaptive adjustment algorithms. It is used to automatically classify, preprocess, segment, and count the input FITC and Cy3 channel fluorescence images, and output the counting results of dual targets.

6. The method for quantifying extracellular vesicle miRNA using bubble self-stratification digital detection according to claim 1, characterized in that, The AI-automated fluorescence counting decoding module performs the following processing flow: 1) For the acquired FITC and Cy3 channel fluorescence images, the phase correlation algorithm based on fast Fourier transform is used to calculate the subpixel translation offset between the two images, and the images are corrected according to the offset to make the two channel images accurately aligned, ensuring that the magnetic beads in the same field of view have consistent coordinates in the two channels. 2) Convert the registered FITC and Cy3 images from the RGB color space to the HSV color space respectively, and extract the hue (H), saturation (S), and lightness (V) components; according to the pre-calibrated characteristic hue range of FITC and Cy3 fluorescent dyes, perform hue matching on each pixel to generate a binary mask, and initially screen out the pixel areas that may be magnetic beads in order to eliminate noise interference from non-target colors in the background. 3) Within the mask area after hue screening, the lightness component V is segmented locally using the Bernson method based on the mean and standard deviation of the pixel neighborhood to obtain a binary image of the magnetic bead candidate area, in order to adapt to uneven illumination and changes in background fluorescence. 4) Perform morphological opening operations to remove small noise and closing operations to connect neighboring regions on the binary image in sequence. Then, use morphological reconstruction algorithms to fill the holes inside the magnetic beads caused by uneven fluorescence to obtain the complete magnetic bead connected region. 5) Perform distance transformation on the connected components, calculate the Euclidean distance from each pixel to the background, and generate a distance map; search for local maxima on the distance map as seed points, and use the watershed algorithm to segment the adhered magnetic beads to obtain the independent regions of individual magnetic beads, thus solving the counting bias caused by magnetic bead aggregation. 6) Calculate the area, perimeter, roundness, and aspect ratio of each independent region. Set a threshold based on the actual particle size of the magnetic bead and the imaging magnification. Remove regions that do not conform to the roundness of the magnetic bead or have abnormal size, and filter out long strip regions that are still stuck together due to incomplete segmentation. 7) Count the number of filtered magnetic beads in the FITC channel and Cy3 channel respectively to obtain the counting results of miR-21 and miR-155; Meanwhile, if a magnetic bead is detected in both channels, it is determined to be the same target based on its coordinates and is considered as cross-color interference and is therefore removed, ensuring the specificity and accuracy of dual-target counting.

7. A bubble self-stratification digital detection system for quantifying extracellular vesicle miRNA, characterized in that, include: Immunocapture bubbles, which are hollow glass microbubbles with anti-EpCAM antibodies coupled to their surface, are used to specifically enrich tumor cell extracellular vesicles. The preparation of immune capture bubbles includes the following steps: Weigh 0.1-1 g of hydroxylated activated hollow glass microbubbles, add 5-15 mL of anhydrous ethanol solution containing 0.1-1 mL of (3-(2,3-epoxypropoxy)propyl)triethoxysilane, and carry out silanization modification reaction for 10-20 h. After centrifugation at 3000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with ultrapure water, and vacuum dry at 35-40 ℃ to obtain epoxy hollow glass microbubbles. Weigh 0.1-1 g of epoxy hollow glass microbubbles, add 1-10 mL of phosphate buffer, add an anti-EpCAM antibody solution with a concentration of 80-120 μg / mL, mix and react for 0.5-1.5 h, centrifuge at 1000-5000 rpm for 0.5-1.5 min, discard the lower layer solution, wash 3-5 times with phosphate buffer, block with bovine serum albumin, wash again and dry to obtain immunocapture bubbles with surface-coupled anti-EpCAM antibody; Fluorescent-DNA dual-coding magnetic bead conjugates, comprising FITC-MB-DNA21-N3 conjugate and Cy3-MB-DNA155-N3 conjugate, are used to detect miR-21 and miR-155, respectively; the preparation of the fluorescent-DNA dual-coding magnetic bead conjugates includes the following steps: Streptavidin-conjugated magnetic beads were mixed with fluorescein isothiocyanate-succinimide ester and Cy3-succinimide ester, respectively, and rotated at room temperature in the dark for 30-90 min to perform fluorescent labeling. The supernatant was removed by magnetic separation, and the beads were washed 3-5 times with 0.01 g / 100 mL Tween 20 phosphate buffer to obtain FITC-labeled magnetic beads and Cy3-labeled magnetic beads, respectively. Biotin-modified clickable DNA probe Biotin-DNA21-N3 was mixed with FITC-labeled magnetic beads and immobilized by the affinity interaction between biotin and streptavidin to obtain FITC-MB-DNA21-N3 conjugate. Biotin-modified clickable DNA probe Biotin-DNA155-N3 was mixed with Cy3-labeled magnetic beads and immobilized by the affinity interaction between biotin and streptavidin to obtain Cy3-MB-DNA155-N3 conjugate. Multifunctional clickable glass microbubbles are hollow glass microbubbles with dibenzocyclooctynyl groups grafted onto their surface. These microbubbles are used to capture unreacted fluorescent-DNA dual-coding magnetic bead conjugates via an azide-dibenzocyclooctynyl click chemical reaction, and then float to the surface using their own buoyancy to achieve self-separation. The preparation of multifunctional clickable glass microbubbles includes the following steps: Hydroxylation activation treatment was performed on the microbubbles of insulating glass to obtain activated insulating glass microbubbles; 100-1000 mg of activated hollow glass microbubbles were added to 10-50 mL of polyethyleneimine aqueous solution with a mass-to-volume ratio of 1-5 mg / mL. The mixture was rotated and reacted at room temperature for 10-20 min to form an aminated coating on the surface by electrostatic adsorption. After centrifugation at 3000-5000 rpm for 0.5-2 min, the mixture was washed 3-5 times with ultrapure water and dried under vacuum at 30-40 °C to obtain aminated hollow glass microbubbles. 10-1000 mg of aminated hollow glass microbubbles were added to 0.5-20 mL of phosphate buffer, followed by 100-300 μL of a 10-30 mM dibenzocyclooctyne-N-hydroxysuccinimide ester solution. The mixture was rotated and reacted at room temperature for 0.5-1.5 h to graft dibenzocyclooctyne groups via a covalent reaction between the surface amino group and the active ester. After centrifugation at 3000-5000 rpm for 0.5-2 min, washing with ultrapure water 3-5 times, and vacuum drying at 30-40 °C, multifunctional click glass microbubbles with bioorthogonal function were obtained. Double-stranded specific nucleases and their reaction buffers are used to cyclically cleave DNA probes on fluorescent-DNA dual-coding magnetic bead conjugates in the presence of target miRNA. The AI-automated fluorescence counting decoding module is configured to perform the following image processing flow: 1) Perform phase correlation registration based on fast Fourier transform on the acquired FITC and Cy3 channel fluorescence images to eliminate sub-pixel offset between channels; 2) Convert the registered image to the HSV color space, generate a binary mask based on the characteristic hue range of the pre-calibrated FITC and Cy3 fluorescent dyes, and eliminate background noise; 3) Local adaptive thresholding is applied to the brightness component V within the mask region to obtain candidate regions for magnetic beads; 4) Morphological reconstruction and hole filling are performed on the candidate regions to obtain complete magnetic bead connected domains; 5) Perform distance transformation on the connected components and use the watershed algorithm to segment the sticky magnetic beads to obtain independent regions for each individual magnetic bead; 6) Shape and size filtering is performed based on the particle size, roundness, and aspect ratio of the magnetic beads to eliminate false positive targets; 7) Count the number of filtered magnetic beads in the dual channels, and remove cross-color interference based on coordinate correlation, and output the counting results of miR-21 and miR-155.

Images