A signal separation method and system for multi-channel fluorescence imaging

Abstract

The invention provides a signal separation method and system for multi-channel fluorescence imaging, which relate to the field of biomedical technology. The method comprises the following steps: a preliminary separation step: performing feature extraction and multi-level feature filtering on a multi-channel fluorescence image through a neural network model, performing preliminary separation on signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence image, and obtaining a single-channel separation signal; a secondary separation step: identifying the single-channel separation signal through an automatic encoder, and re-separating the fluorescence channel of each color to obtain a secondary separation signal; a signal optimization step: automatically adjusting the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signal, and re-separating the area where the signal overlap is greater than a preset threshold to obtain a final separation signal; thereby greatly improving the processing efficiency of signal separation, and improving the accuracy, reliability and overall efficiency of imaging.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and in particular to a signal separation method and system for multi-channel fluorescence imaging. Background Art

[0002] Multichannel fluorescence imaging is widely used in biomedical research, which allows the simultaneous detection and analysis of multiple fluorescent markers in the same experiment. The core of this technology is to use multiple fluorescence channels to capture different wavelengths of light, each channel corresponding to a specific fluorescent marker.

[0003] The following are some key technical points of multi-channel fluorescence imaging technology: (1) Multi-marker detection technology: By using different fluorescent dyes or probes, researchers can label multiple different biological molecules or cell types in one sample, thereby simultaneously observing their behaviors and interactions and visualizing them in cell or tissue samples. Fluorescence imaging can not only provide detailed information on intracellular structure and function, but also play an important role in the dynamic observation of biological processes. It is widely used in research fields such as cell biology and molecular biology.

[0004] (2) High-throughput imaging technology: Multi-channel fluorescence imaging technology can achieve high-throughput imaging, which is particularly important for biomedical research that requires analysis of a large number of samples.

[0005] (3) Signal separation technology: Since multiple fluorescent markers usually have overlapping emission spectra, multi-channel fluorescence imaging technology requires precise algorithms and equipment to separate these signals to ensure that the signal of each marker can be accurately detected.

[0006] (4) Super-resolution imaging technology: Multi-channel fluorescence imaging technology can be combined with super-resolution imaging technology (such as STED, SIM, PALM / STORM) to obtain higher-resolution images; high-resolution microscopy technology can be used to observe biological samples at the nanoscale. This imaging method allows scientists to capture the microstructure of biological samples and reveal their complex biological characteristics, thereby providing a more accurate basis for disease research and drug development.

[0007] Current multi-channel fluorescence imaging technology faces the following problems in application: First, signal overlap is a common challenge. In multi-channel imaging, different fluorescent markers may emit signals in similar wavelength ranges, causing these signals to overlap in the image. This overlap not only makes it difficult to accurately separate the individual signals, but may also lead to misinterpretation of the results, thus affecting subsequent biological analysis and the reliability of experimental results. The severity of the signal overlap problem requires researchers to develop more precise separation techniques to ensure that the true performance of each fluorescent signal is correctly evaluated.

[0008] In addition, existing signal separation algorithms are also insufficient in terms of processing efficiency. Many traditional separation algorithms are computationally complex and take a long time to process, making it difficult to meet the needs of high-throughput imaging in practical applications. This low efficiency not only limits the imaging speed, but may also affect the overall progress of the experiment, especially in the case of large-scale data processing, where extended processing time can significantly reduce the efficiency of the research. Therefore, it is urgent to develop more efficient signal separation algorithms to increase processing speed and ensure that multi-channel fluorescence imaging technology can play a role in fast and dynamic biological experiments.

[0009] Therefore, how to improve the processing efficiency of signal separation is particularly prominent in the current multi-channel fluorescence imaging technology. Solving this problem will help improve the accuracy, reliability and overall efficiency of imaging, thereby promoting the in-depth development of life science research. Summary of the invention

[0010] In view of this, in order to overcome the traditional deficiencies, the purpose of the present invention is to propose a signal separation method for multi-channel fluorescence imaging, which can greatly improve the efficiency of signal separation.

[0011] Based on the above objectives, in a first aspect, the present invention provides a signal separation method for multi-channel fluorescence imaging, comprising the following steps: Preliminary separation step: feature extraction and multi-level feature filtering are performed on the multi-channel fluorescence image through a neural network model, and signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence image are preliminarily separated to obtain single-channel separation signals; Secondary separation step: identifying the single-channel separation signal through an automatic encoder, and re-separating the fluorescence channel of each color to obtain a secondary separation signal; Signal optimization step: automatically adjusting the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, re-separating the area where the signal overlap is greater than a preset threshold, and obtaining the final separation signal.

[0012] As a preferred embodiment of the present invention, before the preliminary separation, any one or more of the following methods for preventing signal crosstalk between channels are also included: a. Choose different excitation light sources according to the characteristics of different fluorescent dyes; b. Set different excitation light intensities and exposure times for different fluorescent dyes; c. Configure different multi-band filters for each channel to improve the wavelength discrimination of the fluorescence signal; d. Excite different fluorescent dyes in sequence by channel-by-channel imaging; e. Set appropriate delayed exposure after each excitation.

[0013] As a preferred embodiment of the present invention, in the preliminary separation step, before the multi-channel fluorescence image is input into the neural network model, any one or more of the following pre-processing is also included: Filtering: Apply median filtering or Gaussian filtering to preprocess each channel image to remove background noise; Homogenization: Homogenize the brightness of each channel image; Enhancement processing: Enhance the contrast of each channel image to make the intensity of different fluorescence signals consistent.

[0014] As a preferred embodiment of the present invention, in the preliminary separation step, the image features are extracted from the multi-channel fluorescence image and then input into the neural network model, and the extracted image features are subjected to any of the following dimensionality reduction processes in the feature extraction stage: PCA processing: Apply principal component analysis to reduce the dimension of the extracted image features, remove redundant information, and retain the main features; ICA processing: Apply independent component analysis to separate independent source signals in the image; NMF processing: Use non-negative matrix factorization to decompose image data into the product of two non-negative matrices to extract the subject or feature information in the image.

[0015] The PCA process further comprises the following steps: Performing standardization processing on multi-channel fluorescence images to obtain standardized image data; Calculating a covariance matrix of the image data; Perform eigenvalue decomposition on the covariance matrix and extract the principal components; The image data is projected onto the principal component to obtain a dimensionally reduced fluorescence signal.

[0016] The ICA process further comprises the following steps: Constructing the observation matrix of multi-channel fluorescence images; Use fast ICA algorithm to decompose the observation matrix; Independent components are extracted from the decomposed data to obtain dimensionally reduced fluorescence signals.

[0017] The NMF process further comprises the following steps: Represent the multi-channel image data as a non-negative matrix V; Decompose into basis matrix W and coefficient matrix H; By minimizing the error function, the basis matrix W and the coefficient matrix H are iteratively updated to obtain the dimensionally reduced fluorescence signal.

[0018] As a preferred embodiment of the present invention, in the secondary separation step, the autoencoder compresses the single-channel separation signal to a low-dimensional representation and reconstructs it, and learns the main features of the image through image opening and closing operations; wherein the opening operation is an operation of first corroding and then dilating, which is used to eliminate small white noise, disconnect the connection between objects, and separate close objects; the closing operation is an operation of first dilating and then corroding, which is used to close small black noise, fill small holes inside objects, and connect close objects.

[0019] As a preferred embodiment of the present invention, in the signal optimization step, the overlapping degree of the secondary separated signals is detected by performing contour detection through signal characteristics, and judging whether the signals overlap based on the roundness, aspect ratio, major axis to minor axis ratio, morphological matching, and number of convex hulls of the signal contour.

[0020] Another aspect of the present invention also provides a signal separation system for multi-channel fluorescence imaging, comprising: Neural network model: The neural network model is used to perform feature extraction and multi-level feature filtering on multi-channel fluorescence images, and the signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence images are preliminarily separated to obtain single-channel separation signals; Automatic encoder: the single-channel separation signal is recognized by the automatic encoder, and each color of the fluorescent channel is separated again to obtain a secondary separation signal; Signal monitoring module: automatically adjusts the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, and re-separates the area where the signal overlap is greater than a preset threshold to obtain the final separation signal.

[0021] Compared with the prior art, the signal separation method of multi-channel fluorescence imaging proposed in the present invention has the following beneficial effects: 1. The present invention combines neural network technology and automatic encoding technology. It performs preliminary separation through a neural network model, and can automatically identify and separate the signal characteristics of fluorescence channels of different colors or different wavelengths in a multi-channel fluorescence image. The automatic encoder further performs secondary separation on the fluorescence channel of each color, which greatly improves the efficiency of signal separation while ensuring the purity and accuracy of the signal.

[0022] 2. Before the initial separation, the present invention provides a variety of ways to prevent signal crosstalk between channels, such as selecting different excitation light sources, setting different excitation light intensities and exposure times, configuring multi-band filters, imaging channel by channel, and setting delayed exposure, etc. These measures effectively reduce signal interference between channels and improve the clarity of signal separation.

[0023] 3. The filtering, homogenization and enhancement processes in the preprocessing step of the present invention can remove background noise, balance image brightness, enhance image contrast, and provide high-quality image data for subsequent signal separation.

[0024] 4. In the preliminary separation step using a neural network model, the present invention introduces dimensionality reduction processing methods such as PCA, ICA and NMF, which can remove redundant information, retain main features, separate independent source signals or extract image themes, thereby improving the accuracy and efficiency of signal separation.

[0025] 5. In the signal optimization step, by detecting the overlap of the secondary separation signals and automatically adjusting the separation algorithm parameters of the autoencoder, the areas with high overlap are re-separated to ensure the accuracy and integrity of the final separation signal. In addition, the separation efficiency is also improved by adopting the local optimization method.

[0026] In summary, the signal separation method of multi-channel fluorescence imaging of the present invention realizes efficient, accurate and adaptive signal separation by combining advanced neural network models, autoencoders and a variety of preprocessing and optimization strategies. It has broad application prospects and important practical value. It is not only suitable for fluorescence signal separation of biomedical images, but also can be widely used in other fields that require multi-channel fluorescence imaging signal separation, such as materials science, environmental monitoring, etc.

[0027] These and other aspects of the present application will be more concise and understandable in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these drawings without paying creative work.

[0029] In the figure: Figure 1 4 is a simplified flowchart of a signal separation method for multi-channel fluorescence imaging according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0031] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0032] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are intended to distinguish two non-identical entities or non-identical parameters with the same name. It can be seen that "first" and "second" are only for the convenience of expression and should not be understood as limitations on the embodiments of the present invention. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, other steps or units inherent to a process, method, system, product or device that includes a series of steps or units.

[0033] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0035] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0036] Traditional separation algorithms are computationally complex and take a long time to process, making it difficult to meet the needs of high-throughput imaging in practical applications. This inefficiency not only limits the imaging speed, but may also affect the overall progress of the experiment, especially in the case of large-scale data processing, where the extended processing time can significantly reduce the efficiency of the research.

[0037] To overcome the low efficiency of traditional separation algorithms, see Figure 1As shown, an embodiment of the present invention provides a signal separation method for multi-channel fluorescence imaging, which adopts a three-step method, which not only ensures the quality of signal separation, but also greatly improves the signal separation efficiency, ensuring that multi-channel fluorescence imaging technology can play a role in rapid and dynamic biological experiments.

[0038] The specific steps include: Preliminary separation step: feature extraction and multi-level feature filtering are performed on the multi-channel fluorescence image through a neural network model, and signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence image are preliminarily separated to obtain single-channel separation signals; Secondary separation step: identifying the single-channel separation signal through an automatic encoder, and re-separating the fluorescence channel of each color to obtain a secondary separation signal; Signal optimization step: automatically adjusting the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, re-separating the area where the signal overlap is greater than a preset threshold, and obtaining the final separation signal.

[0039] In this embodiment, the single-channel separation signal includes a red channel signal, a green channel signal, and a blue channel signal.

[0040] Before the initial separation, any one or more of the following methods are also included to prevent signal crosstalk between channels: a. Choose different excitation light sources according to the characteristics of different fluorescent dyes; each fluorescent dye has its own specific absorption and emission spectrum. By choosing an excitation light source that matches the dye absorption spectrum, the excitation efficiency of a specific dye can be maximized while reducing the excitation of other dyes, thereby reducing crosstalk.

[0041] b. Set different excitation light intensities and exposure times for different fluorescent dyes; by adjusting the intensity of the excitation light and the exposure time, the intensity of the fluorescent signal in each channel can be controlled, making the signal clearer in the image. This method can help balance the brightness differences of different dyes and reduce overexposure or underexposure.

[0042] c. Configure different multi-band filters for each channel to improve the wavelength differentiation of the fluorescence signal; using filters specific to each channel can filter out the fluorescence signals of other channels and only allow light of a specific wavelength to pass through. This can improve the specificity of the signal and reduce cross-interference between different channels.

[0043] d. Take a channel-by-channel imaging approach to sequentially excite different fluorescent dyes; by exciting different fluorescent dyes one by one and capturing images separately, signal overlap that may occur when multiple dyes are excited at the same time can be avoided. This method ensures that the signal of each channel is captured separately, thereby reducing crosstalk.

[0044] e. Set appropriate delayed exposure after each excitation; Setting a short delay time after excitation can give the fluorescent dye enough time to return to the ground state and reduce the impact of the excitation light on subsequent imaging. This helps to reduce background noise caused by the excitation light and improve image quality.

[0045] They can be used alone or in combination to adapt to different experimental conditions and needs. Through these measures, the quality of multi-channel fluorescence imaging can be significantly improved, providing more accurate data for subsequent image analysis and data processing.

[0046] In the preliminary separation step, before the multi-channel fluorescence image is input into the neural network model, any one or more of the following preprocessing is also included: Filtering: Apply median filtering or Gaussian filtering to preprocess each channel image to remove background noise; median filtering is used to remove salt-and-pepper noise in the image. Median filtering replaces each pixel value in the image with the median of its neighboring pixel values, which is very effective in maintaining image edge features; Gaussian filtering is used to reduce Gaussian noise (i.e. random noise) in the image. Gaussian filtering replaces each pixel value in the image with the weighted average of its neighboring pixel values, with the weight determined by the Gaussian distribution. This method can smooth the image.

[0047] Homogenization: Homogenize the brightness of each channel image; Homogenization, also known as normalization, refers to adjusting the brightness distribution of an image so that the brightness level of the image is more uniform across the entire image. Homogenization helps reduce uneven brightness problems caused by uneven illumination or differences in sample preparation.

[0048] Enhancement: Enhance the contrast of each channel image to make the intensity of different fluorescent signals consistent; contrast enhancement can be achieved by adjusting the contrast stretching of the image, histogram equalization, or using other contrast enhancement techniques. Contrast enhancement helps to improve the visibility of weak signals in the image, making it easier for the model to distinguish different fluorescent channels.

[0049] These preprocessing steps can be applied individually or in combination to suit specific image characteristics and analysis needs. The purpose of preprocessing is to improve image quality, reduce noise and inhomogeneity, and thus provide cleaner and more consistent input data for the neural network model, thereby improving the performance of the model and the accuracy of the separation results.

[0050] In this embodiment, the preliminary separation step further includes the following subdivision steps: Multi-channel fluorescence image acquisition: First, it is necessary to acquire multi-channel fluorescence images of the sample through a fluorescence microscope. These fluorescence images usually contain fluorescence signals of different colors or wavelengths, and each color or wavelength corresponds to a different marker or structure in the sample.

[0051] Feature extraction: A neural network model is used to extract features from the fluorescence image. Feature extraction is the process of identifying key information in an image (such as the edge, texture, shape, etc. of a cell), which is crucial for subsequent image analysis.

[0052] Multi-level feature filtering: After feature extraction, multi-level feature filtering is performed. This step may involve screening the extracted features to retain the features that are most helpful in distinguishing different fluorescent channels.

[0053] Preliminary separation: Through the above steps, the neural network model is able to perform preliminary separation of the signal characteristics of fluorescence channels of different colors or wavelengths. This means that the model can recognize and distinguish the signals of the red, green, and blue channels.

[0054] Single channel separated signals: Ultimately, this process will produce single channel separated signals, that is, signals corresponding to the red, green and blue channels respectively. These separated signals can be used for further analysis, such as quantitative analysis, morphological study of cells or tissues, etc.

[0055] In this embodiment, the neural network model adopts a CNN model, and by training the CNN model, specific signal features in a multi-channel fluorescence image are identified. The CNN performs multi-level filtering and analysis on the image through a feature extraction process, thereby separating fluorescence signals of different wavelengths. It can learn the spatial hierarchical features of the signal, effectively distinguish the fluorescence intensity and spatial distribution under different targets or experimental conditions, and automatically separate the fluorescence signal features of different channels.

[0056] In this embodiment, in the preliminary separation step, the image features are extracted from the multi-channel fluorescence image and then input into the neural network model, and the extracted image features are subjected to any of the following dimensionality reduction processes in the feature extraction stage: PCA processing (Principal Component Analysis): The principal component analysis method is applied to reduce the dimensionality of the extracted image features, remove redundant information, and retain the main features. PCA is a statistical method used to transform a set of possibly correlated variables into a set of linearly uncorrelated variables, called principal components, through an orthogonal transformation. These principal components are sorted by the size of the variance, with the first principal component having the largest variance and capturing the most variability in the data. By selecting the first few principal components, redundant information in the data can be removed while retaining most of the important information. Using PCA to reduce the dimensionality of the feature space can also retain the important features of the image as much as possible.

[0057] ICA processing (Independent Component Analysis): Independent component analysis is applied to separate independent source signals in an image. ICA is a computational method used to decompose a multivariate signal or data set into statistically independent non-Gaussian signal sources. Unlike PCA, ICA assumes that the data is a mixture of independent components, and its goal is to recover these independent components. Therefore, ICA processing can be used to separate independent source signals in an image. In this embodiment, ICA processing is used to separate independent signals of different fluorescent markers in a multi-channel fluorescence image.

[0058] NMF processing (Non-negative matrix factorization): Non-negative matrix factorization is used to decompose image data into the product of two non-negative matrices to extract the subject or feature information in the image. NMF is a decomposition technique that decomposes a non-negative matrix into the product of two non-negative matrices. This decomposition method is particularly useful in image processing because it can force the decomposed matrix elements to remain non-negative, which is meaningful when processing image data because pixel values ​​and eigenvalues ​​are usually non-negative. NMF can be used to extract subject or feature information in an image. In this embodiment, specific fluorescence features are extracted in a multi-channel fluorescence image through NMF processing.

[0059] Dimensionality reduction can not only reduce computational complexity, but also help neural network models generalize better and reduce the risk of overfitting. Through these methods, key features can be effectively extracted from multi-channel fluorescence images, laying a solid foundation for subsequent image separation and analysis.

[0060] (1) The PCA processing further comprises the following steps: Standardization: Standardize the multi-channel fluorescence image to obtain standardized image data; standardization. The purpose of this step is to convert the image data into a form with zero mean and unit variance. Standardization can eliminate the effects of brightness and contrast differences between different channels, making the PCA results more accurate.

[0061] Compute the covariance matrix: Calculate the covariance matrix of the image data; the covariance matrix is ​​a square matrix whose elements represent the covariance between image data of different channels. The calculation formula of the covariance matrix is: ; in, is the i-th sample, is the sample mean, and N is the sample size.

[0062] Eigenvalue decomposition and extraction of principal components: Perform eigenvalue decomposition on the covariance matrix and extract the principal components; The purpose of eigenvalue decomposition is to find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix. The eigenvalue represents the variance of the data in the direction of the eigenvector, that is, the variability of the data in that direction. The eigenvector represents the distribution direction of the data in that direction. Various numerical methods can be used for eigenvalue decomposition, such as power iteration method, QR algorithm, etc. Then, from the results of eigenvalue decomposition, select the first few largest eigenvalues ​​and their corresponding eigenvectors. These eigenvectors are the principal components, which represent the most important directions of variability in the data. The number of principal components selected depends on the specific application requirements and the characteristics of the data.

[0063] Principal component projection: The image data is projected onto the principal components to obtain a dimensionally reduced fluorescence signal. The standardized image data is projected onto the extracted principal components. The purpose of this step is to transform the original data into a new space composed of principal components, thereby achieving dimensionality reduction.

[0064] The projection formula is: Z = X·W; Among them, X is the standardized image data, W is the matrix composed of principal components, and Z is the data after dimensionality reduction.

[0065] (2) The ICA process further comprises the following steps: Construct the observation matrix of a multi-channel fluorescence image: First, the data of the multi-channel fluorescence image needs to be organized into an observation matrix. Each row of this matrix represents a pixel position, and each column represents the fluorescence intensity value of a different channel. In this way, the fluorescence signal of each pixel in different channels is integrated into a matrix, laying the foundation for subsequent ICA processing.

[0066] Use the fast ICA algorithm to decompose the observation matrix: The FastICA algorithm is an efficient method for independent component analysis. It estimates the independent components in the observed data through an iterative process. First, the data needs to be preprocessed, including demeaning and whitening, to improve the stability and efficiency of the algorithm. Demeaning is to subtract the mean of the data so that the mean of the data is zero; whitening is to transform the data into a new space through the eigenvalue decomposition of the covariance matrix, so that the covariance matrix of the data is the unit matrix, thereby simplifying the subsequent ICA processing.

[0067] Extract independent components from the decomposed data to obtain the dimension-reduced fluorescence signal: After completing the data decomposition, the FastICA algorithm extracts independent components, which represent independent signal sources in the original observation matrix. In multi-channel fluorescence image processing, these independent components can be regarded as independent signals of different fluorescent dyes, thus achieving the purpose of separating each independent signal from the mixed signal.

[0068] (3) The NMF process further comprises the following steps: Represent multi-channel image data as a non-negative matrix V: In the non-negative matrix V, each element represents the fluorescence intensity of a pixel, and all values ​​are non-negative.

[0069] Decomposed into basis matrix W and coefficient matrix H: The goal of NMF is to decompose the non-negative matrix V into the product of two non-negative matrices W and H, that is, V ≈ WH; where W is a non-negative feature matrix, each column represents a feature; H is a non-negative coefficient matrix, each row represents a sample.

[0070] By minimizing the error function, the basis matrix W and the coefficient matrix H are iteratively updated to obtain the reduced-dimensional fluorescence signal: NMF finds the best W and H by minimizing an error function. The commonly used error function is the Frobenius norm, which minimizes the difference between V and WH. The iterative optimization algorithm uses a multiplication update rule or a gradient descent method: the multiplication update rule is to gradually optimize the objective function by iteratively updating the elements of W and H. The gradient descent method is mainly used to solve the local minimum of the loss function.

[0071] In the secondary separation step, the autoencoder compresses the single-channel separation signal into a low-dimensional representation and reconstructs it, and learns the main features of the image through image opening and closing operations.

[0072] An autoencoder consists of two parts: an encoder and a decoder. The encoder compresses the input image into a low-dimensional hidden representation, while the decoder reconstructs this low-dimensional representation back to the original image. During training, the autoencoder learns how to efficiently represent image data by minimizing the difference between the input and reconstructed image.

[0073] In the encoding stage, the autoencoder maps the image data into a low-dimensional space through a series of neuron layers. This low-dimensional space captures the most important features of the input data while ignoring unimportant details such as noise and background interference.

[0074] In the decoding phase, the autoencoder maps the low-dimensional representation back to a high-dimensional space to reconstruct the original image. Ideally, the reconstructed image should be very similar to the original image, but with noise and unnecessary background information removed.

[0075] In this embodiment, on the basis of the autoencoder, morphological operations, including opening and closing operations, are further applied to enhance image features and reduce noise. Among them, the opening operation (MORPH_OPEN) is an operation of first corroding and then dilating, which is used to eliminate small white noise, disconnect the connection between objects, and separate close objects; the closing operation (MORPH_CLOSE) is an operation of first dilating and then corroding, which is used to close small black noise, fill small holes inside objects, and connect close objects. The combination of the autoencoder and the morphological operation can significantly improve the clarity and separation accuracy of the target signal.

[0076] By monitoring the output of the convolutional neural network (CNN) and the autoencoder, the system can automatically detect areas where the signal overlaps severely or the separation effect is poor. In this embodiment, in the signal optimization step, the overlap degree of the secondary separation signal is detected by performing contour detection through signal characteristics, and judging whether the signal overlaps based on the roundness, aspect ratio, major axis to minor axis ratio, morphological matching, and number of convex hulls of the signal contour.

[0077] In the signal optimization step, the overlap degree of the secondary separation signals can be detected by contour detection through various signal characteristics. The following are some key signal characteristics and methods for determining whether the signals overlap: Contour roundness: Roundness refers to the smoothness of the signal contour. In image processing, a rounded contour usually indicates a clearer signal, while an irregular contour may indicate overlap or interference between signals. Roundness can be evaluated by calculating the change in the curvature of the contour.

[0078] Aspect Ratio: The aspect ratio is the ratio of the major axis to the minor axis of the signal profile, which can reflect the shape characteristics of the signal. In some cases, abnormal aspect ratios may indicate signal overlap or morphological changes.

[0079] Long axis to short axis ratio: Similar to the aspect ratio, the major axis to minor axis ratio is also an important parameter to describe the shape of the signal. It can help identify whether the signal is deformed due to overlap.

[0080] Morphological matching: Morphological matching is a method to evaluate whether signals overlap by comparing the similarity between the actual contours of the signals and the expected contours.

[0081] Convex hull number: The number of convex hulls refers to the number of convex hulls of the signal contour. The convex hull is the smallest convex shape that can contain the signal contour. In signal processing, changes in the number of convex hulls may indicate signal overlap or anomalies.

[0082] By combining these features and methods, the overlap of the secondary separation signals can be effectively detected and evaluated. This comprehensive evaluation method can improve the accuracy of signal separation and help optimize the separation effect.

[0083] Another aspect of the present invention also provides a signal separation system for multi-channel fluorescence imaging, comprising: Neural network model: The neural network model is used to perform feature extraction and multi-level feature filtering on multi-channel fluorescence images, and the signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence images are preliminarily separated to obtain single-channel separation signals; Automatic encoder: the single-channel separation signal is recognized by the automatic encoder, and each color of the fluorescent channel is separated again to obtain a secondary separation signal; Signal monitoring module: automatically adjusts the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, and re-separates the area where the signal overlap is greater than a preset threshold to obtain the final separation signal.

[0084] The present invention is particularly suitable for separating fluorescence signals from biomedical images, and can process complex biomedical image data efficiently and accurately. In practical applications, appropriate algorithms can be selected for combination and optimization according to the characteristics of specific tasks and data sets to achieve the best processing effect. Especially in the feature extraction stage of CNN, the addition of PCA, ICA and NMF can significantly improve the accuracy of feature extraction and signal separation.

[0085] Through the detailed analysis of the above steps, this technical solution realizes the efficient and accurate separation of multi-channel fluorescence imaging signals, providing strong support for research in biomedicine and other fields.

[0086] The system of the present invention includes at least one processor and at least one memory, and may also include: an input system and an output system. The processor, memory, input system and output system may be connected via a bus or other means, and the input system may receive input digital or character information and generate signal input of the method of the present invention. The output system may include display devices such as a display screen.

[0087] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the method of the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the method of the present invention, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0088] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of the computer device. In the present embodiment, the processor is used to run the program code stored in the memory or process data. The processors of multiple computer devices of the computer device of the present embodiment execute various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, i.e., implementing the steps of the above method embodiment.

[0089] It should be understood that, in the absence of mutual conflict, all embodiments, features and advantages described above for the method according to the present invention are also applicable to the method and storage medium of the present invention.

[0090] It will also be appreciated by those skilled in the art that various exemplary logic blocks, modules, circuits and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, a general description has been given to the functions of various schematic components, blocks, modules, circuits and steps. Whether this function is implemented as software or hardware depends on specific applications and the design constraints imposed on the entire system. Those skilled in the art can implement the function in various ways for each specific application, but this implementation decision should not be interpreted as causing a departure from the disclosed scope of the embodiments of the present invention.

[0091] The above are exemplary embodiments disclosed in the present invention, but it should be noted that various changes and modifications may be made without departing from the scope disclosed in the embodiments of the present invention as defined in the claims. The functions, steps and / or actions of the method claims according to the disclosed embodiments described herein do not need to be performed in any particular order. In addition, although the elements disclosed in the embodiments of the present invention may be described or required in individual form, they may also be understood as multiple unless explicitly limited to the singular.

[0092] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples; under the concept of the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the above embodiments of the present invention, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.

Claims

1. A signal separation method for multi-channel fluorescence imaging, characterized in that: The following steps are involved: Preliminary Separation step: feature extraction and multi-level feature filtering are performed on the multi-channel fluorescence image through a neural network model, and the signal features of the fluorescence channels of different colors or wavelengths in the multi-channel fluorescence image are preliminarily separated to obtain a single-channel separation signal; Secondary separation step: identifying the single-channel separation signal through an automatic encoder, and re-separating the fluorescence channel of each color to obtain a secondary separation signal; Signal optimization step: automatically adjusting the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, re-separating the area where the signal overlap is greater than a preset threshold, and obtaining the final separation signal.

2. The signal separation method for multi-channel fluorescence imaging according to claim 1, characterized in that: Before the initial separation, any one or more of the following methods are also included to prevent signal crosstalk between channels: a. Choose different excitation light sources according to the characteristics of different fluorescent dyes; b. Set different excitation light intensities and exposure times for different fluorescent dyes; c. Configure different multi-band filters for each channel to improve the wavelength discrimination of the fluorescence signal; d. Excite different fluorescent dyes in sequence by channel-by-channel imaging; e. Set appropriate delayed exposure after each excitation.

3. The signal separation method for multi-channel fluorescence imaging according to claim 1, characterized in that: In the preliminary separation step, before the multi-channel fluorescence image is input into the neural network model, any one or more of the following preprocessing is also included: Filtering: Apply median filtering or Gaussian filtering to preprocess each channel image to remove background noise; Homogenization: Homogenize the brightness of each channel image; Enhancement processing: Enhance the contrast of each channel image to make the intensity of different fluorescence signals consistent.

4. The signal separation method for multi-channel fluorescence imaging according to claim 1, characterized in that: In the preliminary separation step, the image features are extracted from the multi-channel fluorescence image and then input into the neural network model, and the extracted image features are subjected to any of the following dimensionality reduction processes in the feature extraction stage: PCA processing: Apply principal component analysis to reduce the dimension of the extracted image features, remove redundant information, and retain the main features; ICA processing: Apply independent component analysis to separate independent source signals in the image; NMF processing: Use non-negative matrix factorization to decompose image data into the product of two non-negative matrices to extract the subject or feature information in the image.

5. The signal separation method for multi-channel fluorescence imaging according to claim 4, characterized in that: The PCA process further comprises the following steps: Performing standardization processing on multi-channel fluorescence images to obtain standardized image data; Calculating a covariance matrix of the image data; Perform eigenvalue decomposition on the covariance matrix and extract the principal components; The image data is projected onto the principal component to obtain a dimensionally reduced fluorescence signal.

6. The signal separation method for multi-channel fluorescence imaging according to claim 4, characterized in that: The ICA process further comprises the following steps: Constructing the observation matrix of multi-channel fluorescence images; Use fast ICA algorithm to decompose the observation matrix; Independent components are extracted from the decomposed data to obtain dimensionally reduced fluorescence signals.

7. The signal separation method for multi-channel fluorescence imaging according to claim 4, characterized in that: The NMF process further comprises the following steps: Represent the multi-channel image data as a non-negative matrix V; Decompose into basis matrix W and coefficient matrix H; By minimizing the error function, the basis matrix W and the coefficient matrix H are iteratively updated to obtain the dimensionally reduced fluorescence signal.

8. The signal separation method for multi-channel fluorescence imaging according to claim 1, characterized in that: In the secondary separation step, the autoencoder compresses the single-channel separation signal to a low-dimensional representation and reconstructs it, and learns the main features of the image through image opening and closing operations; wherein the opening operation is an operation of first corroding and then dilating, which is used to eliminate small white noise, disconnect the connection between objects, and separate close objects; the closing operation is an operation of first dilating and then corroding, which is used to close small black noise, fill small holes inside objects, and connect close objects.

9. The signal separation method for multi-channel fluorescence imaging according to claim 8, characterized in that: In the signal optimization step, the overlapping degree of the secondary separation signals is detected by performing contour detection through signal characteristics, and judging whether the signals overlap based on the roundness, aspect ratio, major axis to minor axis ratio, morphological matching, and number of convex hulls of the signal contours.

10. A signal separation system for multi-channel fluorescence imaging, characterized in that: include: Neural network model: The neural network model is used to perform feature extraction and multi-level feature filtering on multi-channel fluorescence images, and the signal features of fluorescence channels of different colors or wavelengths in the multi-channel fluorescence images are preliminarily separated to obtain single-channel separation signals; Automatic encoder: the single-channel separation signal is recognized by the automatic encoder, and each color of the fluorescent channel is separated again to obtain a secondary separation signal; Signal monitoring module: automatically adjusts the parameters of the separation algorithm of the automatic encoder by detecting the overlap degree of the secondary separation signals, and re-separates the area where the signal overlap is greater than a preset threshold to obtain the final separation signal.

Images