Image processing method for dynamic monitoring of cell proliferation

Abstract

The present application relates to the technical field of image processing, and more particularly to an image processing method for dynamic monitoring of cell proliferation. The content includes: obtaining an original microscopic image and performing preprocessing, constructing a structural three-channel tensor based on the preprocessed microscopic image, and introducing a disturbance mechanism driven by a structural response to obtain a disturbance tensor of the preprocessed microscopic image; projecting the disturbance tensor of the preprocessed microscopic image to a joint space-frequency domain to obtain a multi-scale frequency response; superimposing the multi-scale frequency response to obtain a space-frequency joint image response; performing brightness compression and edge smoothing on the space-frequency joint image response to obtain a reconstructed microscopic image, and performing cell tracking and behavior recognition to realize dynamic monitoring of cell proliferation. The problems of inaccurate image processing, easy interruption of cell tracking in complex scenarios, and lack of time sequence dynamic modeling ability in behavior recognition ability in the process of dynamic monitoring of cell proliferation are solved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an image processing method for monitoring the dynamics of cell proliferation. Background Technology

[0002] With the development of life sciences and precision medicine, cell-level dynamic behavior observation and automated analysis have become important tools for studying cell cycle, tumor evolution, drug response, and intervention mechanisms. In this process, microscopic imaging technology, as a core supporting tool, enables continuous imaging of cells in in vitro or in vivo environments, providing a rich data source for subsequent image analysis. However, limited by the optical performance of microscopic imaging systems, illumination conditions, imaging time span, and the complexity of tissue environments, the actual acquired cell image sequences often suffer from uneven brightness, local blurring, low signal-to-noise ratio, structural discontinuities, and inter-frame displacement, severely affecting the accuracy and stability of subsequent image segmentation, cell tracking, and behavior recognition. Existing image enhancement methods often employ histogram equalization, Retinex theory, edge filtering, or frequency domain enhancement, but these methods are often limited to single-frame image processing, making it difficult to simultaneously optimize image quality from both spatial structure and temporal continuity dimensions. Furthermore, they suffer from over-enhancement or artifact generation in edge and texture detail enhancement. In cell tracking, most existing methods rely on simple geometric matching or template-based short-range correspondence strategies, which cannot handle complex movements, cell occlusion, or division, leading to interrupted cell trajectories or target confusion. At the behavior recognition level, existing algorithms generally rely on fixed rules or shallow classifiers, lacking the ability to model the complex evolutionary patterns of cells over time. Summary of the Invention

[0003] This invention provides an image processing method for dynamic monitoring of cell proliferation, which solves the technical problems of inaccurate image processing, easy interruption of cell tracking in complex scenes, and lack of temporal dynamic modeling capability in behavior recognition during the dynamic monitoring of cell proliferation.

[0004] The present invention provides an image processing method for dynamic monitoring of cell proliferation, specifically comprising the following technical solutions:

[0005] An image processing method for monitoring cell proliferation dynamics includes the following steps:

[0006] S1. Acquire the original microscopic image and preprocess it to obtain the preprocessed microscopic image; based on the preprocessed microscopic image, construct the three-channel structural tensor and introduce a perturbation mechanism driven by structural response to construct the perturbation response function and obtain the perturbation tensor of the preprocessed microscopic image; project the perturbation tensor of the preprocessed microscopic image onto the joint space-frequency domain to obtain the multi-scale frequency response.

[0007] S2. A perturbation energy weighting mechanism is introduced to superimpose the multi-scale frequency responses to obtain a spatial-frequency joint image response; brightness compression and edge smoothing are performed on the spatial-frequency joint image response to obtain a reconstructed microscopic image; based on the reconstructed microscopic image, cell tracking and behavior recognition are performed to achieve dynamic monitoring of cell proliferation.

[0008] Preferably, S1 specifically includes:

[0009] The brightness values ​​of the preprocessed microscopic image, as well as the brightness gradients of the preprocessed microscopic image in the horizontal and vertical directions, are explicitly embedded into the three-dimensional tensor space to construct a three-channel structural tensor.

[0010] Preferably, S1 specifically includes:

[0011] In the process of realizing the perturbation mechanism driven by structural response, the perturbation response function is constructed by calculating the second derivatives of the preprocessed microscopic image in the horizontal and vertical directions and introducing a nonlinear enhancement exponent.

[0012] Preferably, S1 specifically includes:

[0013] Based on the perturbation response function, a perturbation energy modulation term is constructed, and combined with the three-channel structural tensor, the perturbation tensor of the preprocessed microscopic image is obtained.

[0014] Preferably, S1 specifically includes:

[0015] A two-dimensional frequency response kernel function is constructed using a Gaussian-sine composite frequency kernel, and then applied to the brightness channel of the perturbation tensor of the preprocessed microscopic image to obtain a multi-scale frequency response. The two-dimensional frequency response kernel function is constructed based on a Gaussian attenuation term and a sine modulation term.

[0016] Preferably, S2 specifically includes:

[0017] In the implementation of the perturbation energy weighting mechanism, based on the perturbation response function, a double integral is applied to quantify the energy accumulation of the overall structural change intensity of the preprocessed microscopic image, thereby obtaining the perturbation energy.

[0018] Preferably, S2 specifically includes:

[0019] Based on the perturbation energy, the fusion weighting factor is calculated; based on the fusion weighting factor, the multi-scale frequency responses are weighted and superimposed to obtain the spatial-frequency joint image response.

[0020] Preferably, S2 specifically includes:

[0021] By calculating the gradient magnitude of the spatial-frequency joint image response and introducing a gradient penalty coefficient and a nonlinear enhancement exponent of a power function, brightness compression and edge smoothing are performed on the spatial-frequency joint image response to obtain the reconstructed microscopic image.

[0022] The beneficial effects of the technical solution of the present invention are:

[0023] 1. By constructing a Gaussian-sine composite frequency kernel, the structural frequency response can be extracted from multiple scales and directions, encompassing not only the overall brightness information of the low-frequency components but also high-frequency boundary details. Furthermore, by analyzing the perturbation energy... Norm weighting, which explicitly selects the frequency components most sensitive to the structure for fusion, can achieve the optimal expression of image content. In order to ensure that the boundary responses of different cells can be effectively distinguished in dense or overlapping cell areas, and effectively avoid the blurring or breakage of cell motion trajectories.

[0024] 2. Based on the spatial-frequency joint image response, a nonlinear image reconstruction function combining power function enhancement and gradient penalty is constructed to compress the brightness range to the effective dynamic range, thereby avoiding local overexposure or underexposure. At the same time, the gradient magnitude of the spatial-frequency joint image response is used to constrain the response intensity of the edge region, forming a dual "enhancement-suppression" mechanism. This effectively weakens the ringing effect and unstructured enhancement generated during the fusion process, adaptively enhances the cell body in different regions, and suppresses background interference, making it particularly suitable for scenes with complex backgrounds in cell images. Attached Figure Description

[0025] Figure 1 This is a flowchart of an image processing method for monitoring dynamic cell proliferation according to the present invention. Detailed Implementation

[0026] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] The following description, in conjunction with the accompanying drawings, details the specific scheme of the image processing method for dynamic monitoring of cell proliferation provided by the present invention.

[0029] See attached document Figure 1 The diagram illustrates an image processing method for monitoring cell proliferation dynamics according to an embodiment of the present invention. The method includes the following steps:

[0030] S1. Acquire the original microscopic image and preprocess it to obtain the preprocessed microscopic image; based on the preprocessed microscopic image, construct the three-channel structural tensor and introduce a perturbation mechanism driven by structural response to construct the perturbation response function and obtain the perturbation tensor of the preprocessed microscopic image; project the perturbation tensor of the preprocessed microscopic image onto the joint space-frequency domain to obtain the multi-scale frequency response.

[0031] Cell image sequences are continuously acquired using a microscopic imaging system. Any frame from the cell image sequence is used as the original microscopic image and preprocessed to obtain a preprocessed microscopic image. The preprocessing steps, such as smoothing, denoising, edge detection, standardization, and normalization, employ techniques well-known to those skilled in the art and will not be elaborated upon here. To avoid problems such as low contrast, uneven illumination, unclear structure, and background noise in the preprocessed microscopic image, and to enhance the latent structural information in the preprocessed microscopic image to adapt to subsequent perturbation and frequency responses, the brightness value and its first-order gradient information of the preprocessed microscopic image are explicitly embedded into a three-dimensional tensor space to characterize the edge response direction and intensity of the preprocessed microscopic image, thereby constructing a three-channel structural tensor. The specific representation is as follows:

[0032] ;

[0033] in, It is a three-channel structural tensor; These are the two-dimensional spatial coordinates of the preprocessed microscopic image. and These represent the pixel indices, i.e., pixel coordinates, in the horizontal and vertical directions, respectively. This represents the channel index, corresponding to the brightness channel, horizontal gradient channel, and vertical gradient channel, respectively. It is the brightness function of the preprocessed photomicrograph, representing the brightness of the preprocessed photomicrograph at the pixel position. The pixel grayscale value at that location; and These represent the brightness gradients in the horizontal and vertical directions of the preprocessed microscopic image, respectively, reflecting... The rate of change of direction was approximated using the Sobel operator, a technique well-known to those skilled in the art, and will not be elaborated upon here. The three-channel tensor of the structure not only represents the preprocessed microscopic image, but also carries the directional and intensity information of cell edge changes, providing a structural basis for the perturbation guidance mechanism.

[0034] Furthermore, to capture the curvature changes and texture non-uniformity of the image structure, a structure response-driven perturbation mechanism is introduced, enabling the perturbation applied to the preprocessed microscopic image to adaptively adjust according to cell edges or texture structure. This structure response-driven perturbation mechanism is implemented through a perturbation response function, defined as follows:

[0035] ;

[0036] in, It is a perturbation response function, representing the position of a pixel. The structural response intensity at the point is used to guide the formation of the subsequent perturbation tensor field; and These represent the second derivatives of the preprocessed microscopic image in the horizontal and vertical directions, respectively, and are used to measure the degree of concavity and convexity of the pixel values ​​in the preprocessed microscopic image in the horizontal and vertical directions. They are implemented using the discrete difference approximation method, which is a well-known technique in the art and will not be described in detail here. It is a nonlinear enhancement index used to control the degree of response amplification in high curvature regions. It is a key factor affecting the shape of the perturbation response function. Based on expert experience, the reference value range is [value range missing]. ; This represents the nonlinear energy aggregation of the second-order curvature structure response along the horizontal and vertical directions at each pixel point in the preprocessed microscopic image, used to simulate the enhanced perception of high-concave-convex regions by the human visual system.

[0037] Based on the aforementioned perturbation response function, a perturbation tensor for the preprocessed microscopic image is constructed. The technical objective is to introduce a periodic perturbation controlled by structural changes into the three-channel tensor space of the structure, thereby generating larger frequency components in the subsequent frequency response while maintaining smoothness. The specific expression of the perturbation tensor for the preprocessed microscopic image is as follows:

[0038] ;

[0039] in, It is the perturbation tensor of the preprocessed microscopic image, representing the tensor after perturbation; This is a sinusoidal perturbation amplitude control factor used to adjust the perturbation intensity. It is obtained based on image texture entropy estimation, and the reference value range is [value range missing]. The method for calculating the image texture entropy is a well-known technique in the art and will not be described in detail here. It is a structural perturbation function, used to implement the perturbation response function. Periodic activation mapping; It is the disturbance energy modulation term, representing the magnitude of the disturbance at each location; It is the normalized modulation factor of the disturbance amplitude, which serves as a relative amplification factor for multiplicative scaling.

[0040] Furthermore, to identify the multi-scale spatial response features in the preprocessed microscopic image, the perturbation tensor of the preprocessed microscopic image is projected into the joint spatial-frequency domain to capture multi-scale information and obtain the multi-scale frequency response. Specifically, firstly, a two-dimensional frequency response kernel function is constructed using a direction-controllable Gaussian-sine composite frequency kernel. Then, the two-dimensional frequency response kernel function is applied to the brightness channel of the perturbation tensor of the preprocessed microscopic image (i.e., the... The frequency response is obtained, and the specific formula is as follows:

[0041] ;

[0042] in, Indicated in scale The preprocessed microscopic image at pixel position Frequency response at that location; It refers to the kernel size, which is set based on expert experience, such as... ; , It represents the relative displacement, indicating the displacement index, and describes the range of the domain of the two-dimensional frequency response kernel function. It is the brightness channel of the perturbation tensor of the preprocessed microscopic image at the pixel position. The value at; In scale The two-dimensional frequency response kernel function; It is a Gaussian attenuation term used to control the suppression effect of the two-dimensional frequency response kernel function on pixels far from the center, and has spatial locality and frequency selectivity; In scale The frequency attenuation factor is used to control the width of the convolution kernel in the frequency domain, corresponding to the sharpness of the frequency response at different scales. It is obtained through existing image frequency domain analysis methods such as Fourier theory. The reference value range is: low frequency scale: High-frequency scale: ; It is a sinusoidal modulation term used to control the directionality of the frequency response, enabling the two-dimensional frequency response kernel function to be sensitive to structural features at different angles. This is the horizontal modulation factor, used to control the sensitivity of the two-dimensional frequency response kernel function to the horizontal structure of the preprocessed microscopic image. It is obtained through analysis based on the principal axis direction of the image's Hessian matrix. Reference values ​​are as follows: The analysis method based on the principal axis direction of the image Hessian matrix is ​​a well-known technique in the art and will not be described in detail here. In scale The vertical modulation factor is used to control the degree of response of the two-dimensional frequency response kernel function to the vertical structure. .

[0043] S2. A perturbation energy weighting mechanism is introduced to superimpose the multi-scale frequency responses to obtain a spatial-frequency joint image response; brightness compression and edge smoothing are performed on the spatial-frequency joint image response to obtain a reconstructed microscopic image; based on the reconstructed microscopic image, cell tracking and behavior recognition are performed to achieve dynamic monitoring of cell proliferation.

[0044] To fuse frequency responses at multiple scales, a perturbation energy weighting mechanism is introduced to weight and superimpose the multi-scale frequency responses, resulting in a joint space-frequency image response:

[0045] ;

[0046] in, At pixel position Spatial-frequency joint image response at [location]; This is the total number of multi-scale values, which is determined based on specific application requirements and is not limited here. Reference values ​​include 2, 3, 4, and 5. In scale The fusion weighting factor is used to reflect the perturbation energy density at each scale to ensure that the scale with the strongest structural response is retained preferentially during fusion. It is the disturbance response function of The power integral represents the integral over a scale. The energy accumulation of the overall structural change intensity of the preprocessed microscopic image, i.e., the perturbation energy; It is the disturbance response function of Power-law integral. Indicates scale index; It is a disturbance of energy. Norm form (generalized energy average), based on generalized norm The norm is obtained through a technical means well-known to those skilled in the art, and will not be elaborated upon here;

[0047] Furthermore, to suppress the ringing effect and unstructured enhancement that may occur after frequency response fusion, the following nonlinear image reconstruction function is introduced to address the spatial-frequency joint image response. Brightness compression and edge smoothing are performed to obtain the reconstructed microscopic image:

[0048] ;

[0049] in, This indicates the pixel position of the reconstructed microscopic image. Pixel value at; It is a very small constant used to prevent undefined operations caused by a value of zero. A reference value is shown below. ; It is the nonlinear enhancement exponent of the power function, determined based on the logarithmic power mapping of Retinex theory, with a reference range of values. ; This is the gradient penalty coefficient, used to suppress high-frequency edge jitter. It is obtained from a variant algorithm based on Retinex theory, and the reference value range is [value range missing]. ; This is the edge gradient amplification factor, used to control the edge suppression intensity. It is obtained based on the local contrast adjustment model of Retinex theory, and the reference value range is [value range missing]. ; It is the gradient magnitude of the spatial-frequency joint image response, representing the spatial-frequency joint image response at the pixel location. The intensity of edge change at the location, , and represents the first-order partial derivatives of the spatial-frequency joint image response in the horizontal and vertical directions, respectively, and represents the local edge intensity in the horizontal and vertical directions, obtained through the Sobel operator; the reconstructed microscopic image output by the above formula can suppress the background response while preserving the cell structure edges; the above-mentioned logarithmic power mapping, variant algorithm and local contrast adjustment model based on Retinex theory are all technical means well known to those skilled in the art, and will not be elaborated here.

[0050] Furthermore, based on the reconstructed microscopic images, a high-quality cell image sequence is obtained, and cell tracking and behavior recognition are performed based on this high-quality cell image sequence. The cell tracking utilizes a Kalman filter to predict the cell's position in the next frame based on the cell's current motion state. The cell's motion state, such as position and velocity, is obtained by extracting features from the high-quality cell image sequence using existing feature extraction techniques. Then, the cost (e.g., Euclidean distance) between the predicted position and the actual detected cell position is calculated using a Hungarian algorithm, and a globally optimal allocation is sought to achieve accurate correspondence between individual cells in consecutive frames of the high-quality cell image sequence, thereby establishing a continuous trajectory describing the cell proliferation process and obtaining the cell's motion trajectory. The Kalman filter and the Hungarian algorithm are techniques well-known to those skilled in the art and will not be elaborated upon here.

[0051] The behavior recognition method utilizes existing feature extraction techniques to extract multidimensional dynamic features from high-quality cell image sequences corresponding to the current movement trajectory of cells, and constructs time-series features. These multidimensional dynamic features include area change rate, boundary curvature fluctuation amplitude, and contour discontinuity index. A sliding time window is introduced, and based on the time-series features, a long short-term memory network is used to learn the temporal evolution law, while an isolated forest model is used to capture abnormal behavior patterns. The aforementioned long short-term memory network and isolated forest model are well-known techniques to those skilled in the art and will not be elaborated upon here.

[0052] In summary, an image processing method for dynamic monitoring of cell proliferation has been developed.

[0053] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0055] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An image processing method for dynamic monitoring of cell proliferation, characterized in that, Includes the following steps: S1. Acquire the original microscopic image and perform preprocessing to obtain the preprocessed microscopic image; The brightness values ​​of the preprocessed microscopic image, along with its brightness gradients in the horizontal and vertical directions, are explicitly embedded into a three-dimensional tensor space to construct a three-channel structural tensor. A structure-response-driven perturbation mechanism is introduced. By calculating the second derivatives of the preprocessed microscopic image in the horizontal and vertical directions and combining them with a nonlinear enhancement exponent, a perturbation response function is constructed. ； in, It is a perturbation response function, representing the structural response intensity at pixel position (x,y); It is the brightness function of the preprocessed microscopic image; and α and β represent the second derivatives of the preprocessed microscopic image in the horizontal and vertical directions, respectively; α is the nonlinear enhancement exponent. Based on the perturbation response function, a perturbation energy modulation term is constructed. Combined with the three-channel structural tensor, the perturbation tensor of the preprocessed microscopic image is obtained, specifically expressed as: ； in, is the perturbation tensor of the preprocessed microscopic image; β is the sinusoidal perturbation amplitude control factor; It is a disturbance energy modulation term; The structure represents a three-channel tensor; (x, y) are the two-dimensional spatial coordinates of the preprocessed microscopic image; c represents the channel index; The perturbation tensor of the preprocessed microscopic image is projected onto the joint space-frequency domain to obtain the multi-scale frequency response. S2. A perturbation energy weighting mechanism is introduced. Based on the perturbation response function, a double integral is applied to quantify the energy accumulation of the overall structural change intensity of the preprocessed microscopic image, thus obtaining the perturbation energy. Based on the perturbation energy, a fusion weighting factor is calculated. Based on the fusion weighting factor, the multi-scale frequency responses are weighted and superimposed to obtain the spatial-frequency joint image response. The spatial-frequency joint image response is subjected to brightness compression and edge smoothing to obtain the reconstructed microscopic image. Based on the reconstructed microscopic image, cell tracking and behavior recognition are performed to achieve dynamic monitoring of cell proliferation.

2. The image processing method for monitoring cell proliferation dynamics according to claim 1, characterized in that, S1 specifically includes: A two-dimensional frequency response kernel function is constructed using a Gaussian-sine composite frequency kernel. This kernel function is then applied to the brightness channel of the perturbation tensor of the preprocessed microscopic image to obtain a multi-scale frequency response. The two-dimensional frequency response kernel function is constructed based on a Gaussian attenuation term and a sinusoidal modulation term, and its specific representation is as follows: ,in, It is a Gaussian decay term; is the frequency attenuation factor at scale s, and u and v are relative displacements used to describe the range of the domain of the two-dimensional frequency response kernel function; It is a sinusoidal modulation term; It is the horizontal modulation factor; It is the vertical modulation factor at scale s.

3. The image processing method for monitoring cell proliferation dynamics according to claim 1, characterized in that, S2 specifically includes: By calculating the gradient magnitude of the spatial-frequency joint image response and introducing a gradient penalty coefficient and a nonlinear enhancement exponent of a power function, brightness compression and edge smoothing are performed on the spatial-frequency joint image response to obtain the reconstructed microscopic image.

Images