Holographic immunodetection method of small target detection algorithm based on slice reasoning assistance

Abstract

The invention discloses a holographic immunodetection method of a small target detection algorithm based on slice reasoning assistance, and belongs to the technical field of immunoassay detection. According to the invention, a lensless holographic imaging technology and a small target deep learning detection model are combined, and the limitations of a traditional immunodetection method in the aspects of operation complexity, equipment portability and detection sensitivity are overcome. According to the portable holographic imaging microscope designed by the invention, signal reading under an ultra-large visual field is realized, and the shackle between the lens magnification factor and the visual field size in a traditional microscope is overcome. Meanwhile, a slice reasoning auxiliary module is designed in the target detection process, a lightweight target detection algorithm is matched, and the problem that errors are large in large-view small-target detection is solved; the detection efficiency of the system is greatly improved while accurate signal reading is carried out, the application scene is expanded, antibiotic residues can be rapidly and sensitively detected, and a new research direction is developed for high-sensitivity field convenient detection of trace target objects in portable equipment and complex samples.

Description

TECHNICAL FIELD

[0001] The present application relates to a holographic immune detection method based on slice inference assisted small target detection algorithm, belonging to the technical field of immune analysis detection. BACKGROUND

[0002] Immune analysis technology, as a highly specific and sensitive detection method, plays a crucial role in food safety, environmental monitoring, and clinical diagnosis. Its core principle relies on the specific binding between antigens and antibodies, achieving qualitative or quantitative analysis of target substances through signal amplification and transduction. Traditional immune detection methods, such as enzyme-linked immunosorbent assay (ELISA), have been widely used in clinical laboratories and scientific research due to their high sensitivity and strong specificity. However, ELISA methods have significant limitations: the operation steps are complicated, usually requiring multiple incubation and washing steps; the detection time is relatively long, often taking several hours to obtain results; in addition, this method relies on expensive equipment such as an enzyme label instrument, which limits its application in on-site rapid detection scenarios. Developing rapid, portable, and low-cost detection technology has become a key research direction.

[0003] In recent years, with the rapid development of micro-nano technology, microsphere-based immune sensing technology has gradually attracted attention due to its unique advantages. As signal probes, microspheres have high specific surface area and good biocompatibility, enabling efficient capture of target molecules and detection through optical, magnetic, or electrochemical signal transduction mechanisms. Compared with traditional methods, microsphere immune sensing technology has high sensitivity, fast response speed, and easy integration, making it show broad application prospects in portable detection devices. However, the detection of microsphere signals still faces technical challenges. Traditional microscope imaging technology is limited by the fixed relationship between the field of view and the magnification, which cannot simultaneously achieve high magnification and large field of view imaging. Especially under low concentration conditions, the sparse distribution of microspheres will result in significant sampling errors due to the limited field of view of traditional microscopes, thereby reducing the accuracy and reliability of detection.

[0004] To overcome the limitations of traditional microscopes, lensless holographic imaging technology has emerged as a new imaging method. This technology records the amplitude and phase information of the light field, generates high-resolution holographic reconstruction images using digital holographic reconstruction algorithms, and realizes the unification of large field of view and magnification. Compared with traditional optical microscopes, lensless holographic imaging systems do not require complex lens groups, have simple structure, low cost, and strong portability, and are very suitable for on-site detection, mobile medical care, and other application scenarios. In the field of immune detection, the application of lensless holographic imaging technology can significantly expand the detection area and reduce errors caused by insufficient sampling, thereby improving the sensitivity and stability of detection.

[0005] There is a phenomenon of missing detection of small targets in large field of view in the target detection of holographic images. Conventional deep learning models have obvious deficiencies in processing holographic images with large field of view and small target detection. Due to the limited receptive field of the model and multiple downsampling operations, the feature information of small targets is easily lost, resulting in an increase in the missing detection rate and the false detection rate. In addition, these models usually have a huge amount of calculation, which is difficult to run efficiently on portable devices with limited resources, which is contrary to the demand for low power consumption and real-time performance in on-site detection. SUMMARY

[0006] To solve the above problems, the present application provides a holographic immune detection method based on slice reasoning assisted small target detection algorithm, to realize high sensitivity and rapid detection of trace target objects. This method combines lens-free holographic imaging technology and innovative small target deep learning detection model, overcoming the limitations of traditional immune detection methods in terms of operation complexity, device portability and detection sensitivity. Specifically, the present application designs a slice reasoning assisted small target detection model, which automatically divides the large field of view holographic image into multiple sub-images, and the designed small target detection model significantly reduces the computational complexity, thereby adapting to portable devices. At the same time, the model introduces attention mechanism and multi-scale feature fusion technology, enhancing the extraction ability of small target features, effectively reducing information loss and improving detection accuracy. In addition, this method takes full advantage of the large field of view of lens-free holographic imaging, combined with the optimized algorithm, overcoming the sampling error problem of traditional methods under low concentration conditions. Experimental verification shows that this method performs excellently in trace target object detection, not only improving the sensitivity and accuracy of detection, but also having good real-time performance and portability, providing an innovative solution for new generation detection technology in food safety, environmental monitoring and clinical diagnosis.

[0007] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0008] The present application provides a holographic immune detection method based on slice reasoning assisted small target detection algorithm, comprising:

[0009] Step 1: Mix the target object to be detected with the pre-prepared detection antibody modified polystyrene microspheres and complete antigen modified magnetic nanoparticles, and perform competitive immune reaction;

[0010] Step 2: Separate the supernatant after reaction by magnetic separation, and collect the supernatant containing free polystyrene microspheres;

[0011] Step 3: Perform holographic imaging of the polystyrene microspheres in the supernatant obtained in step 2 using a portable lens-free holographic imaging microscope to obtain an initial holographic image;

[0012] Step 4: holographically reconstructing the initial holographic image using a holographic reconstruction algorithm based on angular spectrum iteration, and iteratively restoring the image to obtain a reconstructed image;

[0013] Step 5: constructing a target detection dataset from the reconstructed image and training it using a lightweight small target detection model assisted by slice reasoning. The lightweight small target detection model includes a slice reasoning auxiliary module and a lightweight target detection module. The slice reasoning auxiliary module is used to automatically slice the reconstructed image into multiple sub-images. The lightweight target detection module is used to detect and count the polystyrene microspheres in the sub-images, and finally calculates and outputs the total amount of polystyrene microspheres by automatically reassembling the spliced ​​images.

[0014] Step 6: Calculate the concentration information of the target to be detected using the linear relationship between the number of the polystyrene microspheres and the concentration of the target to be detected.

[0015] In one embodiment of the present invention, the slicing reasoning auxiliary module in the lightweight small target detection model automatically slices the reconstructed image into 512×512 pixel sub-images to reduce the receptive field area during the target detection process, increase the recognition accuracy of small microsphere targets, and calculate and output the total amount of polystyrene microspheres by automatically splicing and recombining the sub-images.

[0016] In one embodiment of the present invention, the lightweight target detection module uses the MobileNet_Block module to extract feature information, and combines the Squeeze-and-Excitation attention mechanism module to enhance attention to small target features, while performing multi-scale feature fusion through the Spatial Pyramid Pooling-Fast module.

[0017] In one embodiment of the present invention, the holographic reconstruction algorithm adopts an angular spectrum iterative reconstruction algorithm, which simulates the propagation of the light field in space by decomposing the light field through Fourier transform and inverse Fourier transform, and adds constraints during the iterative process to achieve high-quality holographic image reconstruction, wherein the number of iterations is 100, the diffraction distance is 0.8 mm, and the angular spectrum theory is used to calculate the light field propagation and refine the reconstructed image.

[0018] In one embodiment of the present invention, the portable lensless holographic imaging microscope includes a coherent light source, an optical slit, a CMOS image sensor and a 3D printed dark field housing, wherein the wavelength of the coherent light source is 532 nm, the pixel size of the CMOS image sensor is 1.85 μm, and the resolution is 4032×3036.

[0019] In one embodiment of the present invention, the training parameters of the lightweight small target detection model include: 200 training iterations, a learning rate of 0.01, a batch size of 48, and an AdamW optimizer.

[0020] In one embodiment of the present invention, the polystyrene microspheres are carboxylated polystyrene microspheres with a diameter of 6 μm, and the surface of the microspheres is modified with a detection antibody for the target object to be detected.

[0021] In one embodiment of the present invention, the magnetic nanoparticles are carboxylated magnetic nanoparticles with a diameter of 150 nm and are surface-modified with the complete antigen of the target to be detected.

[0022] In one embodiment of the present invention, the target to be detected is a small molecule antibiotic, including at least one of chloramphenicol, neomycin, and clarithromycin. The linear detection range of the holographic immunoassay is 50 pg / mL to 100 ng / mL, and the linear regression coefficient R 2 It is 0.986.

[0023] In one embodiment of the present invention, the holographic immunoassay method takes 20-30 minutes, and the competitive immune reaction is carried out under the following conditions: 20-40° C. for at least 10 minutes.

[0024] The beneficial effects of the present invention are:

[0025] The present invention utilizes lensless holographic imaging technology combined with a slice reasoning-assisted small target detection algorithm for microsphere signal probe quantification, which can quickly and sensitively detect antibiotic residues.

[0026] The present invention innovatively designs a portable holographic imaging microscope to replace the traditional bulky microscope as the signal readout device of the microsphere probe. The holographic imaging microscope has the advantages of a wide field of view, low cost and good portability.

[0027] The slice reasoning-assisted small target detection algorithm proposed in the present invention can accurately identify and quantify micron-scale microsphere signal probes in the wide field of view of the holographic imaging microscope. At the same time, the small target detection algorithm can also achieve rapid detection, solving the recognition pain point of large errors in small target detection in a large field of view.

[0028] The detection system designed in the present invention has high sensitivity, strong specificity, simple operation and high detection efficiency, and has broad application prospects in the fields of food safety testing, environmental monitoring and in vitro monitoring.

[0029] The present application realizes high-precision detection of micro targets under a super large field of view, and breaks through the physical limitation of the optical system by combining the lensless holographic imaging technology and the adaptive image slicing algorithm, thereby solving the problem of missing detection of small targets under a large field of view.

[0030] The present application realizes the deployment of a lightweight deep learning model on a portable device, and significantly reduces the algorithm calculation amount by introducing an innovative slice reasoning auxiliary module and a lightweight network architecture design, so that the system can stably run on a portable device with limited resources.

[0031] The present application realizes accurate identification of small targets under complex background, and effectively improves the detection accuracy of micro targets under complex background by introducing an attention mechanism and a multi-scale feature fusion technology.

[0032] The present application realizes compatible detection of multiple sample types, and makes the system compatible with the detection requirements of nucleic acids, proteins and other target substances by optimizing the algorithm architecture and feature extraction method. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 It is a flowchart of the holographic immune detection method based on the slice reasoning auxiliary small target detection algorithm of the present application. Wherein: (a) is the working flowchart of the antibiotic detection based on competitive immune reaction; (b) is the architecture diagram of the lensless holographic imaging hardware and software system.

[0035] Figure 2 It is a flowchart of the small target detection algorithm based on the angular spectrum iterative reconstruction algorithm and slice reasoning assistance of the present application.

[0036] Figure 3 It is an effect display based on the angular spectrum iterative reconstruction (ASIR) algorithm of the present application.

[0037] Figure 4This is a comparison chart between the training results of the slice inference assisted small target detection algorithm (SIALSO) of the present invention and the conventional lightweight target detection algorithm YOLOV5s; among them: (a) is the training loss curve; (b) is the object detection accuracy curve; (c) is the mean average precision mAP@0.5; (d) is the verification loss curve; (e) is the object detection recall curve; (f) is a comparison chart between the number of SIALSO and YOLOV5s model parameters.

[0038] Figure 5 3 is a comparison chart between the concentration and number of polystyrene microspheres in holographic imaging and microscopic imaging in an embodiment of the present invention.

[0039] Figure 6 This is a consistency comparison between manual counting and the small target detection algorithm assisted by slice reasoning in an embodiment of the present invention, and a target loss comparison diagram with and without slice reasoning assistance.

[0040] Figure 7 This is a comparison diagram of the original image, the image reconstructed by the ASIR algorithm, and the SIALSO detection result at different concentrations in an embodiment of the present invention.

[0041] Figure 8 This is a diagram showing the optimized conditions for the addition of antibody-modified polystyrene microspheres and fully antigen-modified magnetic nanoparticles in a competitive immune reaction according to an embodiment of the present invention.

[0042] Figure 9 This is a diagram showing the optimization of conditions for competitive immune response time in an embodiment of the present invention.

[0043] Figure 10 This is a comparison diagram between the detection of different concentrations of chloramphenicol and the number of signal probes in the holographic image in an embodiment of the present invention.

[0044] Figure 11 This is a diagram showing the corresponding relationship between different concentrations of chloramphenicol and the number of signal probes detected based on the SIALSO detection method in an embodiment of the present invention.

[0045] Figure 12 3 is a comparison chart between the detection range and quantification limit of the SIALSO detection method and the ELISA detection method in the embodiment of the present invention.

[0046] Figure 13 Graph showing the specific response experimental results for different targets based on the SIALSO detection method in an embodiment of the present invention.

[0047] Figure 14 This is a comparison of chloramphenicol in real samples (fish samples) using heat map distribution comparison based on the SIALSO detection method and the ELISA detection method in the embodiments of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0049] Example 1

[0050] like Figures 1-14 As shown, this embodiment provides a holographic immune detection method based on a small target detection algorithm assisted by slice reasoning, including:

[0051] Step 1: The target substance to be detected is mixed with pre-prepared detection antibody-modified polystyrene microspheres and fully antigen-modified magnetic nanoparticles to conduct a competitive immune reaction;

[0052] Step 2: Separate the supernatant after the reaction by magnetic separation to collect the supernatant containing free polystyrene microspheres;

[0053] Step 3: performing holographic imaging on the polystyrene microspheres in the supernatant obtained in step 2 using a portable lensless holographic imaging microscope to obtain an initial holographic image;

[0054] Step 4: Using an angular spectrum iteration-based holographic reconstruction algorithm (ASIR algorithm) to holographically reconstruct the initial holographic image, and iteratively restore to obtain a reconstructed image;

[0055] Step 5: construct a target detection dataset using the reconstructed image and train it using a slice reasoning-assisted lightweight small target detection model (SIALSO model). The lightweight small target detection model includes a slice reasoning-assisted module and a lightweight target detection module. The slice reasoning-assisted module is used to automatically slice the reconstructed image into multiple sub-images. The lightweight target detection module is used to detect and count the polystyrene microspheres in the sub-images, and finally calculate and output the total amount of polystyrene microspheres by automatically reassembling the spliced ​​images.

[0056] Step 6: Calculate the concentration information of the target to be detected using the linear relationship between the number of the polystyrene microspheres and the concentration of the target to be detected.

[0057] Optionally, the holographic reconstruction algorithm adopts an angular spectrum iterative reconstruction algorithm, which decomposes the light field through Fourier transform and inverse Fourier transform to simulate the propagation of the light field in space, and adds constraints during the iterative process to achieve high-quality holographic image reconstruction, where the number of iterations is 100 and the diffraction distance is 0.8 mm.

[0058] Optionally, the slicing reasoning auxiliary module in the lightweight small target detection model automatically slices the reconstructed image into 512×512 pixel sub-images to reduce the receptive field area during the target detection process, increase the recognition accuracy of small microsphere targets, and calculate and output the total amount of polystyrene microspheres by automatically splicing and recombining the sub-images.

[0059] Optionally, the lightweight target detection module uses the MobileNet_Block module to extract feature information, and combines the Squeeze-and-Excitation attention mechanism module to enhance attention to small target features, while performing multi-scale feature fusion through the SpatialPyramid Pooling-Fast module.

[0060] Optionally, the portable lensless holographic imaging microscope includes a coherent light source, an optical slit, a CMOS image sensor and a 3D printed dark field housing, wherein the wavelength of the coherent light source is 532 nm, the pixel size of the CMOS image sensor is 1.85 μm, and the resolution is 4032×3036.

[0061] Optionally, the number of iterations of the angular spectrum iterative reconstruction algorithm is 100, the diffraction distance is 0.8 mm, and the angular spectrum theory is used to calculate the light field propagation and refine the reconstructed image.

[0062] Optionally, the training parameters of the lightweight small target detection model include: 200 training iterations, a learning rate of 0.01, a batch size of 48, using the AdamW optimizer, and training on a hardware device group built with an NVIDIA GeForce RTX 3090Ti GPU and an Inteli9-12900K CPU.

[0063] Optionally, the lightweight small object detection model dataset contains 1,600 reconstructed images. Of these, 1,280 images (512×512) are designated for training and validation, while 320 images (2048×2048) are reserved for prediction. The dataset is divided into training, validation, and prediction in a 6:2:2 ratio.

[0064] Optionally, the portable lensless holographic imaging microscope includes a coherent light source, an optical slit, a CMOS image sensor and a 3D printed dark field housing, wherein the wavelength of the coherent light source is 532 nm and the power is 3.5 W, the slit diameter of the optical slit is 200 μm, the CMOS image sensor model is SONY IMX226, the pixel size is 1.85 μm, the target surface size is 1 / 1.7 inch, and the resolution is 4032×3036.

[0065] Optionally, the polystyrene microspheres are carboxylated polystyrene microspheres with a diameter of 6 μm and a surface modified with a detection antibody for the target to be detected.

[0066] Optionally, the magnetic nanoparticles are carboxylated magnetic nanoparticles with a diameter of 150 nm and are surface-modified with complete antigens of the target object to be detected.

[0067] Optionally, the target substance to be detected is chloramphenicol, the linear detection range of the method is 50 pg / mL to 100 ng / mL, and the linear regression coefficient R 2 It is 0.986.

[0068] Optionally, the detection method takes 20-30 minutes, of which the competitive immune reaction time is 15 minutes.

[0069] Optionally, the detection antibody-modified polystyrene microspheres are added to the reaction system at a concentration of 2 mg / mL.

[0070] Optionally, the antigen-modified magnetic nanoparticles are added to the reaction system at a concentration of 500 μg / mL.

[0071] Optionally, the pre-prepared detection antibody-modified polystyrene microspheres and antigen-modified magnetic nanoparticles are stored at 4° C. after preparation.

[0072] Optionally, the target to be detected is a small molecule antibiotic, including at least one of chloramphenicol, neomycin, and clarithromycin.

[0073] Optionally, the competitive immune reaction reaction conditions are: 20-40° C. for at least 10 min, optionally 37° C. for 15 min.

[0074] Example 2

[0075] This embodiment provides an immune detection method based on a small target detection algorithm assisted by slice reasoning, see Figure 1 , this embodiment is used to detect the concentration of chloramphenicol in the sample to be tested.

[0076] First, the sources of the main reagents used in this example are introduced:

[0077] Carboxyl functionalized polystyrene microspheres (PSM-COOH, 6 μm) were purchased from Bangs Laboratories, Inc (USA).

[0078] Carboxylated magnetic nanoparticles (MNPs 150 -COOH) was purchased from Ocean Nano-Tech (USA).

[0079] 1-Ethyl-3-(3-dimethylaminopropyl)-carbodiimide hydrochloride (EDC), N-hydroxysulfosuccinimide sodium salt (Sulfo-NHS), phosphate buffered saline (PBS) and 2-(N-morpholino)ethanesulfonic acid hydrate (MES) were purchased from Aladdin Company (Shanghai).

[0080] Bovine serum albumin (BSA) was purchased from Amresco (USA).

[0081] Chloramphenicol (CAP), neomycin, and clarithromycin were purchased from Sigma Aldrich (USA).

[0082] Chloramphenicol antibody (CAP-Ab) and BSA-chloramphenicol antigen (BSA-CAP-Ag) were purchased from Sangon Biotechnology Co., Ltd. (Shanghai).

[0083] The experimental water was deionized using a Millipore water purification system (USA). All chemicals were of analytical grade and did not require further purification before use.

[0084] Preparation method of the relevant reagents used in this example:

[0085] PBS buffer (10 mM, pH = 7.4): Take 8.00 g NaCl, 0.20 g KCl, 0.20 g KH2PO4 and 2.90 g Na2HPO4·12H2O in a 1000 mL volumetric flask, make up to volume and shake well.

[0086] MES buffer (0.1 M, pH 6.0): Dissolve 21.325 g of MES in deionized water and dilute to 1000 mL to form Solution A. Dissolve 4 g of NaOH in deionized water and dilute to 1000 mL to form Solution B. Mix 1000 mL of Solution A and 400 mL of Solution B and shake well.

[0087] PBST, MEST solution: Add 0.5 mL of Tween-20 to 1000 mL of prepared PBS or MES buffer and shake well.

[0088] The immunoassay method of this embodiment mainly includes the following contents:

[0089] (1) Preparation of PS-CAP-Ab probe at room temperature

[0090] PS-6μm microspheres (2 mg) functionalized with surface carboxyl groups were washed twice with MES buffer and resuspended in MES. EDC (5 mg / mL, 30 mL) and NHS (5 mg / mL, 15 mL) (prepared fresh) were then added and incubated at room temperature for 15 minutes. After the incubation reaction, the microspheres were washed using a centrifuge (5000 rpm) and resuspended in PBS. CAP-Ab (3.5 mg / mL, 100 μL) was then added to the microsphere solution and the coupling reaction was allowed to proceed for 3 hours at room temperature. After the coupling step, nonspecific binding sites on the microsphere surface were blocked with blocking buffer, and the coupled microspheres were rinsed three times with PBST and stored below 4°C until use.

[0091] (2) Preparation of MNP-BSA-CAP probe at room temperature

[0092] Take surface carboxyl functionalized MNP-150nm particles (500μg), wash twice with MES buffer, and resuspend in MES after washing. Subsequently, EDC (5mg / mL, 30mL) and NHS (5mg / mL, 15mL) (prepared on demand) were added and incubated at room temperature for 15 minutes. After the reaction is complete, the particles are washed by magnetic separation and resuspended in PBS. BSA-CAP (50μg) is then added and the coupling reaction is carried out at room temperature for 3 hours. After the reaction is complete, the reaction mixture is purified by magnetic separation, and the nonspecific binding sites on the surface of the particles are blocked with BSA (1%, 100μL) blocking solution for 30 minutes. After the blocking step is complete, the particles are rinsed three times with PBST and stored at 4°C for future use.

[0093] (3) Chloramphenicol testing process

[0094] Before the test, we conducted an experiment to optimize the ratio of MNP-BSA-CAP and PS-CAP-Ab addition and the competition reaction time. The experimental results are shown in the figure. Figure 8 、 Figure 9 As shown, subsequent detection experiments were performed based on the results of the optimized conditions. A chloramphenicol standard solution (10 mg / mL, solvent: methanol) was gradiently diluted with PBS to prepare chloramphenicol solutions with concentrations ranging from 0 to 10 μg / mL. Pre-prepared chloramphenicol antibody-modified polystyrene microspheres (PSM-CAP-Ab, 2 mg / mL, 4 μL) and chloramphenicol complete antigen-modified magnetic particles (MNP-BSA-CAP, 500 μg / mL, 10 μL) were mixed with 150 μL of the diluted chloramphenicol standard solutions of varying concentrations. The reaction was rotated at room temperature for 15 minutes to allow the competitive reaction to proceed. After the reaction, the supernatant was collected by magnetic separation and then imaged using a lensless holographic imaging device. The resulting images were used for further data analysis.

[0095] The gradient concentration of chloramphenicol detection corresponds to Figure 10 As shown in the figure, the corresponding relationship between the number of polystyrene microspheres under different chloramphenicol concentration gradients in lensless holographic imaging is shown. The results show that there is a gradient corresponding relationship between the two. Secondly, a linear regression relationship between the two was established, as shown in the figure. Figure 11 As shown in the figure, the linear range of the detection method is 50pg / mL to 100ng / mL, and the linear regression coefficient R 2 Compared with the conventional gold standard ELISA method, Figure 12 As shown, it shows a lower limit of quantification and a wider linear detection range. Finally, in order to verify the specificity of the detection method, specific detection experiments of different targets were carried out. The experimental results are shown in Figure 13 As shown, the detection method only responds to the target chloramphenicol and does not respond to other non-target substances. There is no significant difference between the results and the blank control, which shows that the detection method has good specificity.

[0096] (4) Construction of holographic reconstruction algorithm based on angular spectrum iterative reconstruction

[0097] The holographic reconstruction algorithm is based on the angular spectrum iterative reconstruction (ASIR) algorithm, the algorithm architecture is as follows Figure 2 As shown in (a) in the figure, the algorithm uses angular spectrum theory to calculate the propagation of light field in space. The angular spectrum theory uses Fourier transform and inverse Fourier transform to decompose the light field and restore the real image. Through multiple iterations, the reconstructed image is refined to make it close to the actual light field distribution. The ASIR reconstruction algorithm has a good effect on hologram reconstruction. Figure 3 As shown, in this embodiment, the number of iterations is set to 100, the wavelength is 532 nm, the pixel size is 1.85 μm, and the diffraction distance is 0.8 mm.

[0098] (5) Construction of small target detection algorithm dataset based on slice reasoning assistance

[0099] Images reconstructed using the ASIR reconstruction algorithm are used to construct the SIALSO model dataset, which consists of 1,600 reconstructed images. Of these, 1,280 images (512×512) are designated for training and validation, while 320 images (2048×2048) are reserved for prediction. The dataset is divided into training, validation, and prediction sets in a 6:2:2 ratio. To enhance model robustness, the training and validation sets undergo image augmentation, including the addition of Gaussian noise, sharpening, and flipping.

[0100] (6) Construction of small target detection algorithm based on slice reasoning assistance

[0101] Lensless holographic imaging has the advantage of large field of view imaging. Figure 5The number of polystyrene microspheres imaged by holographic imaging and conventional microscopy is shown, further demonstrating the advantages of holographic imaging and the necessity of slice reasoning when detecting small targets. The slice reasoning-assisted small target detection model includes a slice reasoning auxiliary module and a lightweight target detection module. See the algorithm architecture for details. Figure 2 In (b), the slice reasoning auxiliary module automatically slices the large holographic image into 512×512 sub-images. After using the lightweight target detection module to quickly and accurately detect the small target microspheres, the slice reasoning auxiliary module automatically reconstructs the sliced ​​sub-images and outputs the total number of small target microspheres. The comparative experimental results of whether the slice reasoning auxiliary module is used to detect small targets are shown as follows. Figure 6 The results demonstrate the accuracy and practical application of the slice inference-assisted model. The training model for the small object detection model first extracts image feature information using MobileNet_Block, where depthwise separable convolution reduces computational complexity. The Squeeze-and-Excitation (SE) attention mechanism enhances the model's focus on important features, particularly those of small objects. Furthermore, the Spatial Pyramid Pooling-Fast (SPPF) module fuses multi-scale contextual information from the image to improve object perception at different scales. The UP_Block upsampling module restores image resolution during the decoding phase, enhancing the model's recognition capabilities. The DF_Block downsampling module further compresses features to capture deeper feature information, helping to identify microspheres under different conditions. The prediction model applies the slice inference-assisted module based on the trained model, and after inference, outputs a reassembled image result. This automated process not only improves model accuracy but also reduces computational complexity, enhancing the detection accuracy and efficiency of small objects in high-resolution images.

[0102] The small object detection model was trained for 200 epochs with a learning rate of 0.01, a batch size of 48, and the AdamW optimizer. The model architecture is shown in Figure 2 The small object detection model was implemented using Python 3.8 and Pytorch 3.0.1 architecture and trained on a microcomputer built with an NVIDIA GeForce RTX 3090Ti GPU and an Intel i9-12900K CPU. The model training results are shown in Figure 4 As shown in Figure 2, compared with the commonly used lightweight target detection model YOLOV5s, the evaluation indicators (training loss, accuracy, recall rate, average precision) of the SIALSO model training show better detection results than YOLOV5s. At the same time, the number of model parameters of the SIALSO model and the YOLOV5s model are compared, as shown in Figure 2. Figure 4As shown in (f), the SIALSO model has 29% fewer parameters than YOLOV5s, which proves the lightweightness of the model. The SIALSO model is applied to the reconstruction and detection of holographic images at different concentrations. Figure 7 As shown, the accuracy of its reconstruction and detection is demonstrated.

[0103] Example 3: Detection of the chloramphenicol content in aquatic product samples.

[0104] Aquatic product (sea bass) samples were purchased randomly from supermarkets.

[0105] Pretreatment: Sample pretreatment was carried out in accordance with the national standard test method (GB 31658.2-2021). The specific treatment was to crush the fish sample with a grinder. Chloramphenicol solution of unknown concentration was randomly added to the minced fish, mixed thoroughly and stored at 4°C overnight. Ultrasonic extraction was performed using a mixture of ethyl acetate and acetonitrile (1:1, v / v), with two extraction cycles of 30 minutes each. The obtained supernatant was evaporated under a stream of nitrogen and resuspended in PBS solution for subsequent detection experiments.

[0106] The immunoassay method described in Example 1 was used to detect the chloramphenicol content in the pretreated aquatic product samples. The test results were as follows: Figure 14 As shown, among the 10 aquatic product samples, chloramphenicol was detected in three samples, and the same results were detected in parallel samples, which was the same as the gold standard method ELISA method for detecting chloramphenicol sample size and the samples were consistent.

[0107] In summary, the present invention provides a holographic immunoassay method based on a slice-reasoning-assisted small target detection algorithm, achieving high-sensitivity and rapid detection of trace targets. This method combines lensless holographic imaging technology with an innovative small target deep learning detection model, overcoming the limitations of traditional immunoassay methods in terms of operational complexity, equipment portability, and detection sensitivity. Specifically, the present invention designs a slice-reasoning-assisted small target detection model. By dividing a large-field-of-view holographic image into multiple sub-images, the designed small target detection model significantly reduces computational complexity, thereby adapting to the resource constraints of portable devices. Simultaneously, the model introduces an attention mechanism and multi-scale feature fusion technology to enhance the ability to extract small target features, effectively reduce information loss, and improve detection accuracy. Furthermore, this method fully utilizes the large field of view advantage of lensless holographic imaging and, combined with an optimized algorithm, overcomes the sampling error problem of traditional methods under low-concentration conditions. Experimental verification shows that this method demonstrates excellent performance in trace target detection, not only improving detection sensitivity and accuracy, but also possessing good real-time and portability, providing an innovative solution for the next generation of detection technologies in fields such as food safety, environmental monitoring, and clinical diagnosis.

[0108] Furthermore, the present invention designs a portable holographic imaging microscope that achieves signal readout in an ultra-large field of view, overcoming the constraints between lens magnification and field of view size in traditional microscopes. Simultaneously, a slice reasoning auxiliary module is designed during target detection, combined with a lightweight target detection algorithm to overcome the problem of large errors in detecting small targets in a large field of view. While accurately reading signals, the system's detection efficiency is greatly improved, expanding its application scenarios and enabling rapid and sensitive detection of antibiotic residues. This invention opens up new research directions for the high-sensitivity, on-site, and convenient detection of trace targets in portable devices and complex samples.

[0109] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0110] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A holographic immune detection method based on a small target detection algorithm assisted by slice reasoning, characterized in that: include: Step 1: The target substance to be detected is mixed with pre-prepared detection antibody-modified polystyrene microspheres and fully antigen-modified magnetic nanoparticles to conduct a competitive immune reaction; Step 2: Separate the supernatant after the reaction by magnetic separation to collect the supernatant containing free polystyrene microspheres; Step 3: performing holographic imaging on the polystyrene microspheres in the supernatant obtained in step 2 using a portable lensless holographic imaging microscope to obtain an initial holographic image; Step 4: holographically reconstructing the initial holographic image using a holographic reconstruction algorithm based on angular spectrum iteration, and iteratively restoring the image to obtain a reconstructed image; Step 5: constructing a target detection dataset from the reconstructed image and training it using a lightweight small target detection model assisted by slice reasoning. The lightweight small target detection model includes a slice reasoning auxiliary module and a lightweight target detection module. The slice reasoning auxiliary module is used to automatically slice the reconstructed image into multiple sub-images. The lightweight target detection module is used to detect and count the polystyrene microspheres in the sub-images, and finally calculates and outputs the total amount of polystyrene microspheres by automatically reassembling the spliced ​​images. Step 6: Calculate the concentration information of the target to be detected using the linear relationship between the number of the polystyrene microspheres and the concentration of the target to be detected.

2. The holographic immunoassay method according to claim 1, characterized in that: The slicing reasoning auxiliary module in the lightweight small target detection model automatically slices the reconstructed image into 512×512 pixel sub-images to reduce the receptive field area during the target detection process, increase the recognition accuracy of small microsphere targets, and calculate and output the total amount of polystyrene microspheres by automatically splicing and recombining the sub-images.

3. The holographic immunoassay method according to claim 2, characterized in that: The lightweight target detection module uses the MobileNet_Block module to extract feature information, and combines it with the Squeeze-and-Excitation attention mechanism module to enhance the focus on small target features, while performing multi-scale feature fusion through the Spatial Pyramid Pooling-Fast module.

4. The holographic immunoassay method according to claim 3, characterized in that: The holographic reconstruction algorithm adopts an angular spectrum iterative reconstruction algorithm. The angular spectrum iterative reconstruction algorithm decomposes the light field through Fourier transform and inverse Fourier transform to simulate the propagation of the light field in space, and adds constraints during the iterative process to achieve high-quality holographic image reconstruction. The number of iterations is 100, the diffraction distance is 0.8mm, and the angular spectrum theory is used to calculate the light field propagation and refine the reconstructed image.

5. The holographic immunoassay method according to claim 1, characterized in that: The portable lensless holographic imaging microscope includes a coherent light source, an optical slit, a CMOS image sensor, and a 3D-printed dark field housing. The wavelength of the coherent light source is 532 nm, the pixel size of the CMOS image sensor is 1.85 μm, and the resolution is 4032×3036.

6. The holographic immunoassay method according to claim 1, characterized in that: The training parameters of the lightweight small target detection model include: 200 training iterations, a learning rate of 0.01, a batch size of 48, and the use of the AdamW optimizer.

7. The holographic immunoassay method according to claim 1, characterized in that: The polystyrene microspheres are carboxylated polystyrene microspheres with a diameter of 6 μm and are surface-modified with detection antibodies for the target object to be detected.

8. The holographic immunoassay method according to claim 1, characterized in that: The magnetic nanoparticles are carboxylated magnetic nanoparticles with a diameter of 150 nm and are surface-modified with the complete antigen of the target object to be detected.

9. The holographic immunoassay method according to claim 1, characterized in that: The target to be detected is a small molecule antibiotic, including at least one of chloramphenicol, neomycin, and clarithromycin. The linear detection range of the holographic immunoassay is 50 pg / mL to 100 ng / mL, and the linear regression coefficient R 2 It is 0.

986.

10. The holographic immunoassay method according to claim 1, characterized in that: The holographic immunoassay method takes 20-30 minutes, and the competitive immune reaction is carried out under the conditions of 20-40° C. for at least 10 minutes.

Images