Organ-like three-dimensional image enhanced segmentation method and system

Abstract

The invention relates to the technical field of biomedical image processing and analysis, in particular to an organ-like three-dimensional image enhancement segmentation method and system. The system comprises a data acquisition module, a light field analysis module, a physical enhancement module, a topological feature extraction module and a boundary evolution module. The system constructs a light transmission attenuation map by using deep learning, carries out voxel-level reverse illumination compensation, and carries out dynamic evolution segmentation in combination with a centripetal vector and a topological rejection potential energy field; the core is that digital transparency is realized based on a Beer-Lambert law inverse process, optical transmission physical degradation is eliminated, and brightness distribution of deep and shallow cells is consistent; according to the method, depth invariance of feature extraction is realized, and the problem that a subsequent segmentation algorithm is sensitive to depth change is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of biomedical image processing and analysis technology, specifically to a method and system for enhancing and segmenting three-dimensional images of organoids. Background Technology

[0002] With the increasing application of organoid technology in the biomedical field, the demand for high-throughput three-dimensional imaging and refined analysis of it is growing; such analysis heavily relies on high-quality three-dimensional microscopic image data.

[0003] Currently, organoid samples are typically scanned tomographically using confocal microscopy or two-photon microscopy, and segmented using conventional image processing algorithms. However, due to the optical properties of biological tissues, light signals undergo severe scattering and attenuation when penetrating deep tissues, resulting in a significant reduction in the signal-to-noise ratio of deep regions and loss of texture details. Furthermore, in high-density culture scenarios, adjacent organoids are often in close contact or even adhered, making the optical boundaries of the contact area extremely blurry and exhibiting zero-gradient features. Existing segmentation methods struggle to effectively recover structural information from degraded physical signals and cannot accurately define blurred instance boundaries, easily leading to difficulties in identifying deep targets and failure in separating adhered regions.

[0004] Therefore, overcoming signal attenuation caused by optical transmission and achieving precise separation of high-density organoids under weak boundary conditions has become an urgent problem to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for enhancing and segmenting three-dimensional images of organoids. Specifically, the technical solution of this invention is as follows: A method for enhancing and segmenting three-dimensional images of organoids, comprising: Acquire initial 3D image data containing organoid targets; The initial three-dimensional image data is analyzed using a deep learning model to construct an optical transmission attenuation map. The optical transmission attenuation map includes background light field data characterizing the background illumination intensity and a local attenuation coefficient matrix characterizing the light transmission properties of the medium. Based on the optical transmission attenuation spectrum, voxel-level inverse illumination compensation is performed on the initial three-dimensional image data to generate physically enhanced image data. The physically enhanced image data is input into a preset topology prediction model to extract a multidimensional topology feature field. The multidimensional topology feature field includes a target probability map representing the probability of the target's existence, a centripetal vector field representing the direction pointing towards the target's center, and a topological repulsion potential field representing the separation barrier between targets. A dynamic evolution equation is constructed based on the centripetal vector field and the repulsive force field originating from the topological repulsive potential field. The instance boundary evolution is performed on the organoid target to generate a three-dimensional segmentation result. The topological repulsion potential field is used to define the segmentation boundary between different organoid targets in regions where the gradient magnitude is lower than a preset threshold.

[0006] Preferably, the method for constructing the optical transmission attenuation map includes: The initial 3D image data is input into a convolutional neural network; Output background light field distribution data with the same size as the initial three-dimensional image data; Outputs a local attenuation coefficient matrix that characterizes the rate attenuation of light along the depth direction within biological tissue; The background light field distribution data and the local attenuation coefficient matrix are combined to form the optical transmission attenuation map.

[0007] Preferably, the method for generating physically enhanced image data includes: Subtract the background light field distribution data from the initial three-dimensional image data to obtain the net signal data; The local attenuation coefficient matrix is ​​integrated along the depth direction to generate the cumulative attenuation factor; Based on the cumulative attenuation factor, exponential gain compensation is performed on the net signal data to restore the voxel intensity in the deep region, thereby obtaining the physically enhanced image data.

[0008] Preferably, the topological repulsive potential field in the multidimensional topological feature field is generated in the following manner: Identify local extrema regions in the physically enhanced image data; A repulsive force field is constructed based on the distance between the local extreme regions; In the repulsive force field, an equipotential surface with zero potential energy is defined as the dividing boundary, and the voxel position corresponding to the equipotential surface is the contact interface between adjacent organoid targets.

[0009] Preferably, the method for performing instance boundary evolution on the organoid target includes: Calculate the signal-to-noise ratio of the organoid target at the current depth position; Based on the signal-to-noise ratio index, adjust the weighting coefficients of the geometric rigidity constraints; The configuration is as follows: when the signal-to-noise ratio index is lower than a preset threshold, the weighting coefficient is increased to enhance the shape constraint of the organoid target towards a regular sphere; The contour position of the organoid target is iteratively updated by combining the cohesive force provided by the centripetal vector field, the separation force provided by the topological repulsive potential energy field, and the shape-preserving force provided by the geometric rigidity constraint.

[0010] Preferably, the weighting coefficients of the geometric stiffness constraints are adjusted according to the following rules: Establish a nonlinear mapping relationship between depth coordinates and the weighting coefficients; The weighting coefficient increases monotonically as the depth coordinates increase; The geometric rigidity constraint is achieved by minimizing the gradient magnitude of the level set function.

[0011] Preferably, the training process of the topology prediction model includes: Construct a composite loss function that includes physical consistency loss and geometric topology loss; The physical consistency loss is used to constrain the generation of the optical transmission attenuation map, ensuring that the physically enhanced image data conforms to the physical laws of optical transmission. The generation of the centripetal vector field and the topological repulsion potential field are constrained by the geometric topological loss, ensuring the continuity of the three-dimensional topological structure under sparse labeling conditions.

[0012] An organoid three-dimensional image enhancement and segmentation system, comprising: The data acquisition module is used to acquire initial three-dimensional image data containing organoid targets; The physical modeling module is used to perform light field analysis on the initial three-dimensional image data using a deep learning model and construct an optical transmission attenuation map. The inverse enhancement module is used to perform voxel-level inverse illumination compensation on the initial three-dimensional image data based on the optical transmission attenuation spectrum, and generate physically enhanced image data. The feature prediction module is used to input the physical enhancement image data into a preset topology prediction model and extract a multidimensional topology feature field. The multidimensional topology feature field includes a target probability map representing the probability of the existence of the target, a centripetal vector field representing the direction pointing to the center of the target, and a topological repulsion potential field representing the separation barrier between targets. The topology segmentation module is used to construct a dynamic evolution equation based on the centripetal vector field and the repulsive force field originating from the topological repulsive potential energy field, perform instance boundary evolution on the organoid target, and generate a three-dimensional segmentation result.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes a deep learning model to decouple image degradation into a background light field and a local attenuation coefficient, and performs exponential gain compensation on voxels based on the inverse process of the Beer-Lambert law. This physical-level digital transparency processing enables deep cells to have a brightness distribution similar to that of shallow cells in the enhanced image, eliminating the impact of physical degradation of light transmission on image quality, thereby reducing the sensitivity of subsequent segmentation algorithms to depth changes and achieving depth invariance in feature extraction. 2. To address the problem of blurred optical boundaries or even zero gradient features at the contact surfaces of adjacent organoids, which cause traditional algorithms to fail, this invention does not rely solely on image gradients. Instead, it constructs a repulsive force field by identifying local extreme regions and defines equipotential surfaces with zero potential energy as segmentation boundaries in regions where the gradient magnitude is below a threshold. This method is equivalent to establishing an invisible wall in the adhesion region, forcing the initial seed point to avoid crossing the boundary during the evolution process, thereby achieving accurate segmentation of closely contacting or even adhered targets. 3. This invention establishes a nonlinear mapping relationship between depth coordinates, signal-to-noise ratio (SNR) index, and geometric weight coefficients. In shallow regions with clear images, the system allows the segmentation results to present irregular, realistic biological morphologies. In deep regions with low SNR, the system automatically increases the weight coefficients, utilizing strong prior modes to enhance the shape constraint of organoids towards regular spheres. This dynamic balance mechanism effectively combats deep speckle noise and prevents contour edge jaggedness and fragmentation caused by signal degradation. 4. In the model training phase, the physical consistency loss of this invention forces the parameters decomposed by the network to satisfy the physical laws of light transmission, preventing the model from generating false illusory textures; at the same time, the geometric topology loss uses vector field smoothness and potential field regression constraints to strengthen the spatial continuity between voxels; this dual constraint mechanism enables the system to maintain the integrity of the three-dimensional topology even under sparse labeling conditions, avoiding unreasonable discrete points or topological breaks in the segmentation results. Attached Figure Description

[0014] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0016] Example 1

[0017] Please see Figure 1 A method for enhancing and segmenting three-dimensional images of organoids, comprising: Acquire initial 3D image data containing organoid targets; A deep learning model is used to analyze the light field of the initial three-dimensional image data and construct a light transmission attenuation map. The light transmission attenuation map includes background light field data representing the background illumination intensity and a local attenuation coefficient matrix representing the light transmission characteristics of the medium. Based on the optical transmission attenuation spectrum, voxel-level inverse illumination compensation is performed on the initial three-dimensional image data to generate physically enhanced image data. The physical augmented image data is input into a preset topology prediction model to extract a multidimensional topology feature field. The multidimensional topology feature field includes a target probability map representing the probability of the target's existence, a centripetal vector field representing the direction pointing towards the target's center, and a topological repulsion potential field representing the separation barrier between targets. Based on the centripetal vector field and the topological repulsion potential field, a dynamic evolution equation is constructed to perform instance boundary evolution on the organoid target and generate a three-dimensional segmentation result. Among them, the topological repulsion potential field is used to define the segmentation boundary between different organoid targets in regions where the gradient magnitude is lower than a preset threshold.

[0018] This embodiment details the physical-topological dual-loop execution logic of the method, aiming to solve the boundary blurring problem caused by physical signal attenuation in the deep regions of high-density organoids. The system executes the data acquisition step, using a confocal microscope or two-photon microscope to perform tomographic scanning on the fluorescently labeled organoid sample to obtain initial three-dimensional image data. The data originates from high-throughput sampling of optical sensors, and its physical meaning is the original spatial distribution of light intensity, including attenuated signals affected by biological tissue scattering and environmental background noise. The system initiates a light field analysis program, using a deep learning model to decouple image brightness degradation into two physical processes: additive noise and multiplicative attenuation, constructing a light transmission attenuation map. Based on this, the system performs a physical-level digital transparency operation, restoring the signal for each voxel according to the inverse process of the Beer-Lambert law, generating physically enhanced image data. Furthermore, this data is input into the topology prediction model, which outputs a target probability map, a centripetal vector field, and a core topological repulsion potential field. The system constructs a dynamic evolution equation based on the centripetal vector field and the topological repulsion potential field, responding to edge ambiguity regions where the gradient magnitude is below a preset threshold. The preferred setting is the quantile value of the gradient magnitude histogram of the current input image, or the adaptive low threshold obtained by calculating the gradient map using the Otsu method. The topological repulsive potential field is used to forcibly define the segmentation boundary, so that the initial seed points are aggregated under the action of the centripetal field, while avoiding crossing the boundary under the action of the repulsive potential field. The final generated equipotential surface is the segmented cell surface. This embodiment uses reverse physical enhancement by constructing an optical transmission attenuation map, which effectively restores the texture details of the deep regions of organoids. In the high-density culture scenario of organoids, by introducing a topological repulsion potential energy field, an artificial boundary defined by geometric potential energy is established in the zero-gradient region where the optical boundary is blurred, thereby achieving accurate separation of deep weak signal targets and high-density contact cells, and solving the problem of failure of the traditional watershed algorithm in the adhesion region.

[0019] Example 2 Methods for constructing optical transmission attenuation maps include: Input the initial 3D image data into the convolutional neural network; Output background light field distribution data with the same dimensions as the initial 3D image data; Outputs a local attenuation coefficient matrix that characterizes the rate attenuation of light along the depth direction within biological tissue; The background light field distribution data and the local attenuation coefficient matrix are combined to form an optical transmission attenuation map.

[0020] This embodiment further specifies the above-described light field analysis steps. The process employs a lightweight 3D fully convolutional neural network, such as 3DU-Net and its variants, as the core of the analysis. Specifically, the convolutional neural network adopts an encoder-decoder architecture, with a multi-task parallel output head designed at the decoder end: the first output head outputs a single-channel background light field B through a 1×1×1 convolutional layer, and the second output head outputs the local attenuation coefficient of a single channel in parallel. Both output heads use the Softplus activation function to satisfy non-negative physical constraints. The initial 3D image data is input into the network; the first branch of the network outputs the background light field B distribution data, which is derived from the network's feature extraction of low-frequency components of the image, and physically represents the distribution of additive background noise caused by non-specific fluorescence and ambient light; simultaneously, the second branch of the network outputs the local attenuation coefficient. Its origin lies in the network's deep learning of tissue heterogeneity. Physically, it represents the light intensity attenuation rate per unit length when light passes through a specific voxel location. This parameter varies with spatial location, reflecting structural differences within organoids, such as cysts or dense cell clusters. To satisfy the non-negativity constraint of light transmission physical quantities—that is, light intensity and attenuation coefficient cannot be negative—a Softplus activation function is introduced at the end of the output layer of the convolutional neural network. Mathematically, it is expressed as: ; This operation ensures the output background light field B and local attenuation coefficient. Constant satisfaction and This avoids complex number or numerical divergence problems in subsequent integration steps; the system outputs the combination of these two sets of physical parameters.

[0021] Methods for generating physically enhanced image data include: Subtract the background light field distribution data from the initial 3D image data to obtain the net signal data; The cumulative attenuation factor is generated by integrating the local attenuation coefficient matrix along the depth direction. Based on the cumulative attenuation factor, exponential gain compensation is performed on the net signal data to restore the voxel intensity in the deep region, thus obtaining physically enhanced image data.

[0022] This embodiment further specifies the above-described reverse illumination compensation steps, aiming to mathematically and rigorously reverse the physical degradation process of optical transmission; the system calculates net signal data. This removes interference from ambient light and non-specific staining; numerical integration is performed along the depth z-axis on the local attenuation coefficient matrix to generate the cumulative attenuation factor. ; The source is the integral of the local attenuation coefficient along the optical path, which physically represents the total attenuation experienced by light rays from the sample surface to depth z, expressed as a dimensionless factor. Based on this, the system performs exponential gain compensation on the net signal data, calculated using the following formula: This process yields physically enhanced image data; instead of blindly amplifying brightness, it specifically compensates for the exponential loss of photon energy due to increased depth. This embodiment employs a physical model-based inverse compensation mechanism to ensure that deep cells in organoid deep imaging exhibit a brightness distribution similar to that of shallow cells in the enhanced image. This approach significantly reduces the sensitivity of subsequent segmentation networks to depth changes, achieving depth invariance in feature extraction and ensuring the consistency and robustness of the segmentation algorithm at different imaging depths.

[0023] Example 3 The topological repulsive potential field in the multidimensional topological feature field is generated in the following way: Identify local extrema regions in physically enhanced image data; A repulsive force field is constructed based on the distance between local extreme regions; In a repulsive force field, an equipotential surface with zero potential energy is defined as the dividing boundary, and the voxel position corresponding to the equipotential surface is the contact interface between adjacent organoid targets.

[0024] This embodiment details the generation mechanism of the topological repulsive potential field, aiming to establish an invisible wall between adjacent targets. The system identifies local extreme regions representing the core of each cell in the physically enhanced image or target probability map. For any point in space, the distance to the two nearest extreme points is calculated, and a repulsive force field is constructed based on this distance difference. Specifically, let the spatial point... The Euclidean distance to the nearest extreme point is The Euclidean distance to the second nearest extremum is Repulsive potential energy The signed distance function is defined as the distance difference: ; Under this definition, when a point is located exactly halfway between the centers of two cells, that is... The potential energy is zero, forming a zero potential energy surface; a repulsive force field. It is then defined as the negative gradient of the potential energy field: ; The force field has opposite directions on both sides of the contact surface, thus forming a separation barrier; the potential energy ridge or zero potential energy surface is defined as the dividing boundary in the repulsive force field, so that even if the gray levels of two cells are continuous on the physical image, they are separated by a high-energy barrier in the topological potential energy space.

[0025] The training process of the topology prediction model includes: Construct a composite loss function that includes physical consistency loss and geometric topology loss; By using physical consistency loss to constrain the generation of optical transmission attenuation maps, we can ensure that physically enhanced image data conforms to the physical laws of optical transmission. By utilizing geometric topological loss to constrain the generation of centripetal vector fields and topological repulsion potential fields, the continuity of the 3D topological structure is maintained under sparse labeling conditions.

[0026] This embodiment further defines the model's training strategy; it constructs a composite loss function. ;in, and These are the hyperparameter weighting factors for the physical loss term and the geometric loss term, respectively; utilizing physical consistency loss... Based on the physical constraints of the Beer-Lambert law, the specific formula is as follows: ; in, For the input image, The attenuation coefficient of the network output is the integral along the optical path. The loss function, which takes the background field output by the network, forces the parameters decomposed by the network to satisfy the physical reconstruction equation, preventing the model from producing illusory textures. It's worth noting that this physical consistency loss constructs a self-supervised learning mechanism, allowing the network to operate without needing to acquire the background light field. and attenuation coefficient Training can be completed using the true labels; at the same time, geometric topological loss is introduced. The constraints are placed on the centripetal vector field and the topological repulsion potential field, specifically using a combination of cosine similarity loss, smoothness constraint, and potential field regression loss: ; in, Represents the sampling space The total number of voxels within, Indicates the current voxel index. express Norm; For the vector field predicted by the model, For a true value vector field, For the predicted topological repulsion potential field, Here is the true value of the potential field calculated based on the real distance transformation; where the true value vector field is... and the true value of the potential field It is an organoid binary mask based on manual annotation, which is pre-generated by calculating the Euclidean distance transformation and its gradient field respectively; For vector field smoothing regularization coefficients, The potential energy field constraint coefficient; In the specific experimental setup of this embodiment, in order to ensure the effectiveness of physical constraints while also taking into account the learning of geometric features, the weighting factor... The preferred setting is 10.0. Set to 1.0, regularization coefficient Set as , Set to 1.0; In response to sparse labeling conditions, this loss function enhances the spatial continuity between voxels, penalizes the main discrete points in the vector field, and forces the potential field to maintain the monotonicity of geometric distance in unlabeled regions, ensuring the continuity of the topology in three-dimensional space. This embodiment introduces a topological repulsion field based on geometric positional relationships rather than simply relying on image gradients. Even in areas where organoid images are extremely blurry or even have invisible edges, the boundaries can still be accurately delineated based on the geometric logic of cell center repulsion. Combined with the joint constraints of physical consistency and geometric topological loss, it effectively solves the problem of zero-gradient contact surface separation and ensures the integrity of the three-dimensional topological structure when labeled data is scarce.

[0027] Example 4 Methods for instance boundary evolution of organoid targets include: Calculate the signal-to-noise ratio of the organoid target at the current depth location; Adjust the weighting coefficients of the geometric rigidity constraint based on the signal-to-noise ratio index; The configuration is as follows: when the signal-to-noise ratio is lower than the preset threshold, the weighting coefficient is increased to enhance the shape constraint of the organoid target towards a regular sphere; By combining the cohesive force provided by the centripetal vector field, the separation force provided by the topological repulsive potential field, and the shape-preserving force provided by the geometric rigidity constraint, the contour position of the organoid target is iteratively updated.

[0028] This embodiment discloses a depth-adaptive geometrically rigid constraint evolution strategy; the system calculates the organoid target at the current depth position. Local signal-to-noise ratio index ; The source is local statistical analysis of images, and its physical meaning is the ratio of signal quality to noise level of the current depth slice; specifically, the signal-to-noise ratio (SNR) metric. The calculation formula is: ; in, Slice at current depth The average gray value of the foreground region voxels after Otsu thresholding. The average gray value of the voxels in the background area. The standard deviation of the voxel gray level in the background region. To prevent small constants with a denominator of zero from being taken as values, ; The system establishes a negative correlation between the weights of the geometric regularization terms in the evolution equation and the signal-to-noise ratio (SNR). In response to an SNR index falling below a preset threshold (identifying a deep high-noise region), the system automatically increases the weight coefficients. The contour position is iteratively updated using level set functions, and the specific dynamic evolution equation is constructed as a level set PDE. ; in, The time step of the evolutionary iteration, The rate at which the characterization profile evolves over time; For level set functions, For the mean curvature, The calculation formula is: ; Here, are the depth-adaptive weighting coefficients, where, Defined as a normalized unit vector field, its calculation formula is: ; in, For the current voxel index, The coordinates of the target center closest to this voxel. Describes the Euclidean norm. A minimal constant to prevent division by zero errors; similarly, a topological repulsive force field. Normalization is also required, that is... To ensure that it is aligned with the level set normal The dimensional consistency of the dot product operation; The aggregation velocity provided for the centripetal vector field The separating force provided by the topological repulsive potential field and These are the convection term weighting coefficients for the centripetal and repulsive force fields, respectively; to ensure the convergence and separation effect of the evolution, in a specific embodiment, they are set to... To drive the main contour convergence, set To provide adequate separation and repulsion force at the contact surface; It should be noted that the weighting coefficient , and It plays a role in balancing physical dimensions in numerical implementation; among them, the curvature flow term It exhibits diffusion properties and is related to the second spatial derivative, while the convection term... Related to the first spatial derivative; to ensure the physical consistency of the dynamic equations, the weighting coefficients... This implies a physical meaning related to the diffusion coefficient, thus balancing the rate of change of time on the left side of the equation. The dimensional difference between the flow term and the geometric flow term on the right ensures the stability of the numerical evolution; the process combines the equilibrium of three forces: the cohesive force provided by the centripetal vector field, the separation force provided by the topological repulsive potential field, and the shape-preserving force provided by the geometric rigidity constraint.

[0029] The adaptive adjustment of the weighting coefficients of geometric stiffness constraints follows these rules: Establish a nonlinear mapping relationship between depth coordinates and weighting coefficients; The weighting coefficient increases monotonically with increasing depth coordinates. The geometric rigidity constraint is achieved by minimizing the gradient magnitude of the level set function.

[0030] This embodiment further quantifies the aforementioned weight adjustment mechanism; it establishes weight coefficients. With signal-to-noise ratio The negative correlation mapping relationship is used to achieve depth adaptive adjustment, and the calculation formula is as follows:

[0031] in, Weights of basic geometric constraints This is the gain coefficient. In the preferred embodiment, the signal-to-noise ratio sensitivity coefficient is used. The value ranges from 0.1 to 1.0 to control the smoothness of the weight variation with the signal-to-noise ratio; the physical meaning of this formula is: when the signal-to-noise ratio... At higher levels, in shallow regions, the exponential term approaches 0, and the evolutionary process is mainly driven by the image data term; when the signal-to-noise ratio... When decreasing, in deeper regions, the weights The geometric rigidity increases exponentially, forcibly strengthening the constraints to maintain the spherical shape; in this process, the geometric rigidity constraints specifically minimize the gradient magnitude of the level set function, i.e., the mean curvature flow term. To achieve this; when the weights increase in depth, the curvature flow term in the PDE equation dominates the transformation process. The main function of the algorithm is to forcibly smooth out the burrs on the contour, so that the segmentation result tends to a smooth sphere or ellipsoid. This embodiment employs a dynamic balancing strategy, allowing the segmentation results to present irregular, realistic biological morphologies in shallow layers where the image is clear, while automatically switching to a strong prior mode in deep layers where the signal-to-noise ratio is extremely low. This counterintuitive design relies on the geometric assumption that the target should be spherical to combat noise, prevent contour fragmentation caused by deep speckle noise, and significantly improve the segmentation robustness across the entire depth range.

[0032] Example 5 Please see Figure 2 A three-dimensional image enhancement and segmentation system for organoids, comprising: The data acquisition module is used to acquire initial three-dimensional image data containing organoid targets; The physical modeling module is used to analyze the light field of the initial 3D image data using a deep learning model and construct a light transmission attenuation map. The inverse enhancement module is used to perform voxel-level inverse illumination compensation on the initial three-dimensional image data based on the optical transmission attenuation map, and generate physically enhanced image data. The feature prediction module is used to input the physical augmented image data into the preset topology prediction model and extract a multi-dimensional topology feature field. The multi-dimensional topology feature field includes a target probability map representing the probability of the target's existence, a centripetal vector field representing the direction pointing towards the target's center, and a topological repulsion potential field representing the separation barrier between targets. The topology segmentation module is used to construct dynamic evolution equations based on the centripetal vector field and the topological repulsion potential field, perform instance boundary evolution on organoid targets, and generate three-dimensional segmentation results.

[0033] This embodiment provides a system architecture based on computer hardware. The system includes the following core components: a data acquisition module connected to an optical microscope interface, configured to cache raw 3D data from high-throughput scanning in real time; a physical modeling module with a built-in pre-trained neural network, such as a variant of 3DU-Net, configured to build and store background light field and attenuation coefficient tensors in GPU memory; an inverse enhancement module calling a high-precision floating-point arithmetic unit to perform voxel-level subtraction and exponentiation operations to generate physically enhanced image data; a feature prediction module running a topology prediction model, employing a multi-channel output architecture to output probability maps, vector fields, and potential energy fields in parallel; and a topology segmentation module as the core computing unit, running a level set evolution algorithm based on partial differential equations, with a depth-weight mapping table in memory for dynamically calling adaptive weights during the evolution process, ultimately outputting 3D segmentation results with instance IDs. This embodiment integrates physical signal recovery with geometric topology segmentation through the pipelined collaboration of hardware modules. The output of the physical modeling module is directly used as the input of the inverse enhancement module, ensuring the realism of the data stream at the physical level. Combined with the dynamic evolution calculation of the topology segmentation module, high-precision and automated analysis is achieved in complex biological tissue imaging scenarios.

[0034] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An organoid three-dimensional image enhanced segmentation method, characterized by, The method comprises the following steps: acquiring initial three-dimensional image data containing organoid targets; performing light field analysis on the initial three-dimensional image data by using a deep learning model to construct a light transmission attenuation atlas, wherein the light transmission attenuation atlas comprises background light field data representing background light intensity and a local attenuation coefficient matrix representing light transmission characteristics of a medium; performing voxel-level inverse light compensation on the initial three-dimensional image data based on the light transmission attenuation atlas to generate physically enhanced image data; inputting the physically enhanced image data into a preset topological prediction model to extract a multi-dimensional topological feature field, wherein the multi-dimensional topological feature field comprises a target probability map representing the probability of the existence of a target, a centripetal vector field representing the direction of the center of the target, and a topological repulsive potential field representing the separation barrier between targets; constructing a dynamic evolution equation based on the centripetal vector field and a repulsive force field derived from the topological repulsive potential field to perform instance boundary evolution on the organoid targets to generate a three-dimensional segmentation result; wherein the topological repulsive potential field is used to define the segmentation boundary between different organoid targets in a region with a gradient module length lower than a preset threshold.

2. The organoid three-dimensional image enhanced segmentation method according to claim 1, wherein, The method for constructing the light transmission attenuation atlas comprises the following steps: inputting the initial three-dimensional image data into a convolutional neural network; outputting background light field distribution data consistent in size with the initial three-dimensional image data; outputting a local attenuation coefficient matrix representing the attenuation rate of light along the depth direction in the biological tissue; combining the background light field distribution data and the local attenuation coefficient matrix into the light transmission attenuation atlas.

3. The organoid three-dimensional image enhanced segmentation method of claim 2, wherein, The method for generating the physically enhanced image data comprises the following steps: subtracting the background light field distribution data from the initial three-dimensional image data to obtain net signal data; integrating the local attenuation coefficient matrix along the depth direction to generate a cumulative attenuation factor; performing exponential gain compensation on the net signal data based on the cumulative attenuation factor to restore the voxel intensity in the deep region, thereby obtaining the physically enhanced image data.

4. The organoid three-dimensional image enhanced segmentation method of claim 1, wherein, The topological repulsive potential field in the multi-dimensional topological feature field is generated by the following method: identifying local extreme value regions in the physically enhanced image data; constructing a repulsive force field based on the distance between the local extreme value regions; in the repulsive force field, defining an equipotential surface with zero potential as a segmentation boundary, and the voxel position corresponding to the equipotential surface is the contact interface of adjacent organoid targets.

5. The organoid three-dimensional image enhanced segmentation method of claim 1, wherein, The method for performing instance boundary evolution on the organoid targets comprises the following steps: calculating a signal-to-noise ratio index of the organoid targets at the current depth position; adjusting the weight coefficient of the geometric rigidity constraint based on the signal-to-noise ratio index; configuring the weight coefficient to increase when the signal-to-noise ratio index is lower than a preset threshold, so as to enhance the shape constraint force of the organoid targets tending to be a regular sphere; iteratively updating the contour position of the organoid targets by combining the aggregation force provided by the centripetal vector field, the separation force provided by the topological repulsive potential field, and the shape maintaining force provided by the geometric rigidity constraint.

6. The organoid three-dimensional image enhanced segmentation method of claim 5, wherein, The adjustment of the weight coefficient of the geometric rigidity constraint follows the following rules: establishing a nonlinear mapping relationship between the depth coordinate and the weight coefficient; monotonically increasing the weight coefficient with an increase of the depth coordinate; wherein the geometric rigidity constraint is achieved by minimizing a gradient norm of a level set function.

7. The organoid three-dimensional image enhanced segmentation method of claim 1, wherein, The training process of the topological prediction model comprises: constructing a composite loss function comprising a physical consistency loss and a geometric topology loss; using the physical consistency loss to constrain generation of the light transmission attenuation atlas, to ensure that the physically enhanced image data conforms to the law of light transmission physics; using the geometric topology loss to constrain generation of the centripetal vector field and the topological repulsive potential field, to ensure continuity of the three-dimensional topological structure under sparse annotation conditions.

8. An organoid three-dimensional image enhanced segmentation system for implementing the organoid three-dimensional image enhanced segmentation method according to any one of claims 1 to 7, characterized in that, comprise: a data acquisition module configured to acquire initial three-dimensional image data containing an organoid target; a physical modeling module configured to perform light field analysis on the initial three-dimensional image data using a deep learning model, and construct a light transmission attenuation atlas; an inverse enhancement module configured to perform voxel-level inverse lighting compensation on the initial three-dimensional image data based on the light transmission attenuation atlas, and generate physically enhanced image data; a feature prediction module configured to input the physically enhanced image data into a preset topological prediction model, and extract a multi-dimensional topological feature field, the multi-dimensional topological feature field comprising a target probability map representing a probability of existence of a target, a centripetal vector field representing a direction pointing to a center of the target, and a topological repulsive potential field representing a separation barrier between targets; a topological segmentation module configured to construct a dynamic evolution equation according to the centripetal vector field and a repulsive force field derived from the topological repulsive potential field, and perform instance boundary evolution on the organoid target, to generate a three-dimensional segmentation result.

Images