FISH adaptive z-layer acquisition method, apparatus, and medium

Abstract

The application provides a FISH adaptive Z layer acquisition method and device and medium, to solve the problem that the existing adaptive layer scanning is easy to miss the unobserved nuclear domain weak signal by relying on local signal attenuation. The method establishes a conservative local nuclear domain with a boundary uncertainty margin under a unified physical Z coordinate; the confidence lower bound of the axial width is constrained according to the lowest standard signal to coarsely screen the sample and establish blank exclusion evidence; the candidate axial interval is finely scanned, and the axial signal trajectory is constructed according to the actual physical Z coordinate; the trajectory closure and unobserved nuclear domain exclusion are used as the core evidence, the nuclear domain coverage and imaging quality are used as the gate for joint evaluation and supplementary sampling control, and the method stops and outputs the de-duplication fluorescent signal.

Description

Technical Field

[0001] This invention relates to the fields of microscopic optical imaging and biomedical image processing technology, and in particular to a FISH adaptive Z-layer acquisition method, device and medium based on nuclear domain exclusion evidence and fluorescence signal axial trajectory closure feedback. Background Technology

[0002] Fluorescence in situ hybridization (FISH) detection typically utilizes a microscopic imaging device capable of precise movement along the X, Y, and Z axes to spatially locate target cells in an ex vivo slide. Images are acquired in nuclear labeling channels used to characterize the target cell nucleus and in one or more probe fluorescence channels. Due to the inherent axial point diffusion characteristics of microscope objectives, the same physical fluorescence spot in the object space will exhibit a bidirectional defocused response at multiple adjacent Z-axis physical locations in the image space, ranging from weak to strong, reaching a peak, and then gradually decaying. In addition, the effective acquisition axial range required varies for different cells and even different channels due to factors such as uneven cell nucleus thickness, slide tilt, different probe hybridization depths, chromatic aberration shifts in multi-channel optical focal planes, and imaging noise interference.

[0003] For the aforementioned axial imaging characteristics, the first common layer acquisition scheme is fixed-range layer scanning. This scheme pre-sets fixed axial start and end positions, the number of scanning layers, and a fixed interlayer spacing for selected cells. After acquiring images of all preset layers, signal detection, 3D fusion, and counting analysis are performed uniformly. This open-loop acquisition method cannot dynamically control the Z-axis movement based on real-time image evidence. If the preset scanning range is too wide, it will result in invalid exposure of a large number of signal-free areas, mechanical idle running, wasted image transmission and memory buffering, and accelerated photobleaching of fluorescent probes. If the preset range fails to completely cover the axial peak or attenuation tail of the actual fluorescence signal, it will lead to signal truncation and missed acquisition of edge layers and repeated counting across layers.

[0004] The second approach is adaptive tomography based on local signal feedback. This approach attempts to dynamically expand the acquisition range based on the sharpness or signal brightness of the detected target in the current layer image, but typically only uses "whether the detected signal has completely attenuated" as the feedback stopping criterion. In actual samples, a strong initial signal and a weak signal in an unscanned axial region that meets the preset minimum detection standard may coexist within the cell nucleus. When the initial signal completes axial attenuation closure on both sides of the peak, if the system makes decisions based solely on the integrity of the existing trajectory and its local neighborhood, it may mistakenly equate "no new signal detected" with "no signal in the undetected region," thus triggering false premature termination and resulting in severe under-collection of weak signals in the unobserved nuclear region. In essence, the evidence of the integrity of the detected signal cannot, physically and statistically, replace the evidence of signal exclusion in the unobserved nuclear region. If a full-domain, small-step, high-exposure, dense tomography is applied to the entire cell nucleus to avoid under-collection, the significance of adaptive acquisition in improving timeliness is completely lost. Furthermore, during adaptive acquisition, neglecting exposure time, gain, light source fluctuations, and grayscale distortion caused by photobleaching, ignoring the disconnect between the acquisition time sequence and the actual physical Z coordinate under reciprocating scanning, or treating abnormal layer errors such as mechanical inaccuracy, invalid image quality, and excessive mapping as blank exclusion evidence will severely compromise the accuracy of axial closed-loop control. Summary of the Invention

[0005] This invention provides a FISH adaptive Z-layer acquisition method, device, and medium. Existing fixed-layer scanning or adaptive layer scanning based only on locally fed-back fluorescence signals cannot simultaneously ensure the integrity of axial acquisition closure of detected signals and the effectiveness of signal exclusion for unobserved nuclear regions at the lowest detectable standard without performing a comprehensive and dense layer scan of the entire cell nucleus. This can easily lead to problems such as missed detection of hidden weak signals, redundant and invalid exposures, and erroneous termination.

[0006] The core technology of this invention is to establish a conservative local kernel domain and a closure verification safety domain under a unified physical Z coordinate system, to establish blank exclusion evidence for unobserved kernel domains by coarse screening of the limit physical constraints through axial point diffusion detection, to establish closure evidence for the actual physical Z-axis signal trajectory of the decoupled time sequence by fine scanning of candidate intervals, and to drive supplementary acquisition decision and adaptive stopping through dual gating of kernel domain coverage and channel imaging quality.

[0007] In a first aspect, the present invention provides a FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback, the method comprising the following steps: Nuclear anchoring data of target cells is obtained in the nuclear labeling channel. The device command position, the actual position of the Z-axis and the axial focal plane offset of each imaging channel are integrated into a unified physical Z coordinate system. Under the unified physical Z coordinate system, the main three-dimensional region of the target cell with boundary uncertainty margin is determined as a conservative local nuclear domain based on the nuclear anchoring data. In the probe fluorescence channel, the coarse screening sampling parameters are determined based on the lower bound of the detectable axial width confidence of the lowest standard signal in the channel under effective imaging parameters, the Z-axis comprehensive positioning error, and the safety margin. Coarse screening layer acquisition is performed within a conservative local kernel domain, and the normalized layer response of each layer is obtained. For coarse screening layers with effective quality, candidate signals are detected within the conservative local kernel domain. When no candidate signal is detected and the normalized layer response of the corresponding coverage area is measured to be lower than the preset detection threshold, blank exclusion evidence is generated to characterize that there is no signal in the corresponding nuclear axis sub-region that meets the preset minimum detectable condition. Based on the candidate signals detected by the coarse screening layer, candidate axial intervals are constructed in a unified physical Z coordinate system and fine scanning layer acquisition is performed. Adjacent candidate signals are associated according to the actual physical Z coordinates to construct axial signal trajectories. The trajectory closure status is evaluated based on the normalized axial response of each axial signal trajectory, and the coverage status of the conservative local kernel, the exclusion status of unobserved sub-regions within the conservative local kernel, and the channel imaging quality status are jointly updated. The joint assessment uses trajectory closure and exclusion status as core evidence, and coverage status and channel imaging quality status as validity gating conditions for joint stopping conditions. When the joint stopping conditions are not met, a supplementary acquisition action is generated based on the evidence gap to drive the microscopic imaging equipment to acquire data until the joint stopping conditions are met, and the acquisition stops. Based on the axial signal trajectory with double-sided closure, the cross-layer deduplication fluorescence signal is output.

[0008] Furthermore, under a unified physical Z-coordinate system, a conservative local nuclear region with boundary uncertainty margin is determined based on nuclear anchoring data for the target cell. This includes: extracting a planar nuclear mask based on the nuclear label channel image and obtaining the local axial estimation boundary of the target cell nucleus; superimposing the boundary uncertainty margin, which integrates boundary estimation error, nuclear segmentation error, channel axial offset calibration error and Z-axis positioning error, on the local axial estimation boundary to generate a conservative local nuclear region. The method also includes: extending a non-negative safety margin along the Z-axis outside the conservative local core domain to form a closed verification safety domain, and defining the region outside the conservative local core domain as the closed verification safety extension band; wherein, the closed verification safety extension band is only used for the actual measurement of the prediction area and the tail attenuation verification of the existing axial signal trajectory, and does not participate in the calculation of the coverage state and the exclusion state.

[0009] Furthermore, the coarse screening sampling parameters include at least the maximum permissible interlayer spacing of the coarse screening. The maximum permissible interlayer spacing of the coarse screening is constrained by the upper limit of the detectable axial width confidence lower bound minus twice the Z-axis comprehensive positioning error and safety margin, so that the signal that is in the conservative local kernel region and reaches the preset minimum target intensity can be detected by at least one coarse screening layer with effective quality under the most unfavorable sampling phase. When the detectable axial width confidence lower bound is no greater than twice the sum of the Z-axis comprehensive positioning error and the safety margin, it is prohibited to establish a sparse coarse screening exclusion plan. Instead, reduce the interlayer spacing, increase the exposure or recalibrate after the signal-to-noise ratio, switch to the preset dense complete layer scan, or mark the corresponding nuclear region as unable to form exclusion evidence.

[0010] Furthermore, the process of obtaining the normalized layer response includes: performing local background estimation and subtraction on the original images of each acquisition layer, and eliminating the influence of exposure time, camera gain response coefficient, light source output ratio, and fluorescence residual ratio on image grayscale through channel calibration; converting the response after background subtraction to a uniform scale through parameter normalization within the camera linear response range, and through pre-calibrated lookup table or monotonic piecewise mapping in the nonlinear gain range or near saturation range.

[0011] Furthermore, the construction and status assessment of the axial signal trajectory include: The effective acquisition layers within the candidate axial interval are sorted according to their actual physical Z coordinates. Based on a weighted cost function that includes the distance between the centroids of the plane, the degree of regional overlap, the difference in feature continuity, and the difference in actual physical Z position, a correlation is established and a trajectory is constructed between candidate signals whose physical Z position difference does not exceed the maximum correlation interval. Establish a trajectory layer observation state including candidate matching, measured predicted region, and observation missing; for existing axial signal trajectories where no candidate signal is matched in the effective acquisition layer, determine the predicted region based on adjacent matched positions and channel registration error; when the predicted region can be reliably mapped, directly measure the normalized support response within the predicted region and record the observation state as the measured predicted region, and only use the measured response below a preset threshold as evidence of attenuation; when the current layer image quality is invalid, the predicted region exceeds the reliable mapping range, or there is a correlation conflict, record the observation state as observation missing, and prohibit using observation missing as a basis for attenuation or blanking; When adjacent signals are stuck together in a single layer, the local maxima of the distance transform are combined with the predicted centroids of the adjacent layers to generate segmentation seeds. After the region is split, the sub-regions are assigned to the axial signal trajectories by the allocation algorithm.

[0012] Furthermore, assessing the trajectory closure status includes: Based on the effective observation of the axial signal trajectory under each actual physical Z coordinate, the peak position and peak response are determined, and sampling positions that meet the attenuation conditions are found below and above the peak position respectively. When the normalized axial response at the sampling position is not greater than the peak response multiplied by the relative attenuation threshold and not greater than the background termination threshold, the lower attenuation evidence and the upper attenuation evidence are formed respectively. When the conservative local kernel boundary corresponding to the trajectory centroid has been effectively sampled and covered, the trajectory has obtained a measured response below the background termination threshold before reaching the boundary, and the boundary confidence state meets the preset requirements, kernel domain boundary termination evidence is formed; when the boundary confidence state is insufficient, the extended layer is obtained along the unclosed side within the closed verification safe domain, and the attenuated response is measured within the prediction region. When the axial signal trajectory simultaneously obtains evidence of attenuation on both sides, or obtains evidence of attenuation on one side and termination evidence of the nuclear domain boundary on the other side, it is determined to be a bilateral closure.

[0013] Furthermore, the joint stopping conditions require that the following conditions be met simultaneously: all non-noise axial signal trajectories have obtained evidence of bilateral closure; there are no unexcluded areas within the conservative local kernel that are not effectively covered by measured low response; the adjacent effective sampling interval of the conservative local kernel on the actual physical Z coordinate is not greater than the maximum allowable interlayer spacing and the upper and lower boundaries fall within the detectable coverage range; and all necessary layers participating in the determination meet the channel imaging quality requirements. When the current probe fluorescence channel has not yet formed an effective axial signal trajectory, the coverage state, exclusion state and channel imaging quality state must still be met in order to form a state of sufficient acquisition of no candidate signals; When the joint stopping condition is not met, candidate acquisition actions are generated, including filling the coarse screening gap, expanding the unclosed trajectory within the closed verification security domain, encrypting the candidate interval, quality re-sampling, and filling the conservative boundary coverage. The action to be executed is selected based on the expected reduction in evidence gaps and equipment costs. When the next acquisition location exceeds the device's safe travel range, maximum number of layers, cumulative exposure, or time budget, it enters a restricted state. Dense layer scanning or local resampling is performed on the area with insufficient evidence. If the evidence still cannot be supplemented, the corresponding uncovered area, unexcluded area, unclosed trajectory, or quality gap is retained, and an uncertain state is output.

[0014] Secondly, the present invention provides a FISH adaptive Z-layer acquisition device based on nuclear domain exclusion evidence and fluorescence signal axial trajectory closure feedback, comprising: The conservative nuclear domain and closure verification safety domain establishment unit is used to determine the conservative local nuclear domain of the target cell with boundary uncertainty margin based on the nuclear anchoring data obtained by the nuclear labeling channel in a unified physical Z coordinate system, and to set the closure verification safety domain along the Z axis outside the conservative local nuclear domain. The coarse screening planning unit is used to determine the coarse screening sampling parameters based on the detectable axial width confidence lower bound of the lowest standard signal in the probe fluorescence channel, the comprehensive positioning error of the Z-axis, and the safety margin, and to generate a coarse screening sampling plan that covers the complete conservative local nuclear domain. The quality assessment unit is used to evaluate the imaging quality of each sampling layer image in terms of Z-axis accuracy, saturation ratio, background fluctuation, signal-to-noise level, and light source stability. The candidate interval generation unit is used to generate candidate axial intervals to be collected by fine scanning in a unified physical Z coordinate system based on the candidate signals detected by the coarse screening layer. The trajectory observation unit is used to establish the trajectory layer observation status, including candidate matching, measured prediction area and observation missing, and to measure the normalized support response when the prediction area can be reliably mapped. The trajectory construction unit is used to associate adjacent candidate signals ordered according to their actual physical Z coordinates to construct axial signal trajectories, and to perform multi-target splitting and allocation when the single layer is adhered. The closure judgment unit is used to determine the trajectory closure status based on the relative attenuation degree of the axial signal trajectory on both sides of the peak, the background termination state, or the kernel domain boundary termination state. Coverage exclusion assessment unit is used to dynamically maintain the coverage exclusion map and jointly assess the geometric coverage of conservative local nuclear domains and the measured blank exclusion status of unobserved nuclear axis subregions; The candidate action evaluation unit is used to select the next acquisition action based on the expected reduction in evidence gap and equipment cost of each candidate acquisition action when the joint stopping condition is not met. The feedback control unit is used to convert the next acquisition action into a drive control command for the microscopic imaging equipment, and to stop acquisition when both the closure evidence of the axial signal trajectory and the exclusion evidence of the unobserved nuclear region are satisfied, and the coverage state and the channel imaging quality state pass the gating conditions. Based on the double-closed axial signal trajectory, the cross-layer deduplication fluorescence signal is output.

[0015] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to execute the above-described FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback.

[0016] Fourthly, the present invention provides a readable storage medium storing a computer program, which, when executed by a processor, implements the FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback as described above.

[0017] The main contributions and innovations of this invention are as follows: 1. This invention breaks away from the inherent technical bias of traditional adaptive tomography which "stops only based on the integrity of the existing target." It establishes "evidence of exclusion from unobserved nuclear regions" and "evidence of closure of the axial trajectory of the detected signal" as two independent and parallel core technical evidences. Furthermore, it applies dual gating of nuclear region geometric coverage and imaging quality, allowing acquisition to stop only when all non-noise trajectories within the entire nuclear region are bilaterally closed and all unobserved areas have obtained measured low-response exclusion evidence. This fundamentally eliminates the risk of false negatives due to premature closure of the initial local signal, thus achieving a high degree of reliability in balancing the integrity of target signal acquisition with the timeliness of adaptive scanning.

[0018] 2. This invention constructs a rigid inequality based on the lower bound of the detectable axial width of the lowest standard signal under effective imaging parameters, the comprehensive positioning error of the Z-axis, and the safety margin. This inequality determines the maximum permissible interlayer spacing for the coarse screening, ensuring that the lowest standard signal within the core domain can be detected by at least one high-quality coarse screening layer even under the most unfavorable sampling phase. The system first covers the entire core domain with such sparse layers that have physical exclusion effects, and then performs dense fine scanning only on the axial sub-intervals where candidate signals are detected. This eliminates the need for uniform high-density exposure across the entire core domain, significantly reducing the number of invalid exposures in signal-free areas, the Z-axis motor reciprocating motion stroke, data throughput, and peak memory usage, effectively delaying the photobleaching decay of the fluorescent probe.

[0019] 3. This invention extends the estimated boundary of the microscopic nuclear labeling channel into a conservative local nuclear domain by incorporating multiple physical calibration errors. A closed-loop verification safety domain is reserved outside the nucleus, strictly limiting the outside-nuclear closed-loop verification safety extension band to only be used for measured attenuation verification of existing signal trajectories at the nuclear edge, prohibiting the provision of coverage or blank exclusion proofs to the nucleus. This mechanism avoids the mechanical truncation of axial trajectories at the boundary due to blurred nuclear edges or diffuse light spots, and eliminates the technical risk of low-response outside the nucleus contaminating unobserved areas inside the nucleus, ensuring the reliability of boundary signal deduplication and the legitimacy of spatial decision-making.

[0020] 4. After acquiring images of each layer, this invention eliminates the influence of exposure time, gain nonlinear response, light source fluctuation, and fluorescence bleaching, achieving normalized layer response measurement and avoiding misjudgment of changes in instrument hardware parameters as axial point diffusion attenuation of fluorescent bright spots. At the same time, candidate signals are strictly sorted according to the actual physical Z coordinate, completely decoupling the acquisition timestamp from the spatial axial topology. Supplemented by a single-layer adhesion resolution mechanism based on the distance transformation maxima and bipartite graph matching, even if the system performs bidirectional alternation, skip sampling, or local backtracking, it can accurately reconstruct the true axial trajectory of the physical fluorescence signal and evidence of bilateral attenuation closure.

[0021] 5. In the three-state observation, this invention strictly separates "predicted low response" from "observation gaps" and strictly prohibits treating abnormal layers, out-of-range layers, or incomplete layers as blanks or attenuation evidence; it introduces position feedback combined with calibration delay for dual stability confirmation in Z-axis motion control; and when encountering limited working conditions such as mechanical limit travel, exposure limit, or task timeout and unable to supplement evidence, it actively retains the corresponding evidence gap and outputs an uncertain state with an unresolved mark, preventing the automatic treatment of equipment failure or sampling dead zones as negative conclusions, thus ensuring the high rigor of the technical output results and the verifiability of clinical auxiliary applications.

[0022] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the conservative local kernel domain and the closed verification security domain according to an embodiment of the present invention. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0025] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0026] Example 1 This embodiment proposes a FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback. Figure 1 This is the core processing flowchart of this embodiment, as follows: Figure 1 As shown, the main flow of the entire method is as follows: Establish a dual-domain system: unified physical Z-coordinate, conservative local kernel domain, and closed verification security domain; Coarse screening and acquisition → generating normalized layer response, and splitting into candidate axial intervals / blank exclusion evidence according to "whether candidate signal is detected"; Fine scan and trajectory → merge candidate intervals, perform fine scan, construct trajectory according to actual physical Z coordinates, distinguish three-state observations, and update closure status; Joint state update → three states: kernel coverage, exclusion of unobserved kernels, and channel imaging quality; Stop judgment → If the joint stop condition is met, output the result; otherwise, generate a supplementary sampling action to continue sampling, and set a "limited and unable to be supplemented → output uncertain state" branch.

[0027] This method is jointly executed by a microscopic imaging device, a Z-axis drive mechanism, a channel switching mechanism, a camera, a device status acquisition unit, and a processor.

[0028] In this embodiment, the nuclear labeling channel is an imaging channel used to characterize the target cell nuclear region, such as the 4',6-diamidinyl-2-phenylindole (DAPI) nuclear staining channel, or other fluorescent labels or imaging channels capable of distinguishing nuclear regions. Nuclear anchoring data is the target cell nuclear image data obtained through the nuclear labeling channel, used to determine the planar nuclear mask, axial reference position, and nuclear axial support range of the target cell. Nuclear anchoring data can be a short layer sequence obtained around a clear nuclear location, or a combination of a single clear image and pre-calibrated nuclear thickness parameters.

[0029] The unified physical Z-coordinate is the axial coordinate after converting the device command position, the actual Z-axis position, and the axial focal plane offset of each imaging channel to the same physical reference system of the slide. The actual device position is expressed as:

[0030] in, For device command location, This refers to the positioning deviation obtained based on position feedback or repeated positioning calibration. (Probe channel) The physical Z-position is represented as:

[0031] in, For channel Axial calibration offset relative to the reference channel. When position-dependent axial color difference exists, It can be extended to a planar position function. When the device can directly read back the actual position, the read-back value can be used instead. The calculation form.

[0032] The channel sampling status is the probe channel. The incremental acquisition record on the current target cell includes at least the planned and actual physical Z-position set, images of each position, exposure parameters, gain, light source status, quality indicators, candidate signals, candidate axial intervals, trajectory status, conservative local kernel domains, closure verification safety domains, equipment safety travel, cumulative exposure, maximum allowed number of layers, and mission time budget.

[0033] In this embodiment, the nuclear axial support range of the target cell, after the uncertainty margin of the superposition boundary, is the following conservative local nuclear region. The axial support range of the core in the claims and the conservative local core domain refer to the same spatial region when they have a boundary uncertainty margin.

[0034] In this embodiment, the conservative local kernel domain and the closed verification security domain are established as follows: This step performs device and channel calibration, establishing a conservative local kernel domain and a closed verification security domain under a unified physical Z-coordinate.

[0035] First, read the calibration data of the current objective and probe channels, including the axial point diffusion response and the lower confidence limit of the detectable axial width of the lowest standard signal. axial offset of the channel or position-related offset The calibration data includes Z-axis repeatability error, camera gain response, light source output correction relationship, and background noise. The calibration data can be updated after equipment installation, objective lens switching, maintenance, or when light source drift is detected to exceed a threshold.

[0036] Subsequently, the microscopic imaging device is moved to the XY position of the target cell to obtain a clear nuclear image or short slice sequence in the nuclear labeling channel. Background correction, thresholding, connected component filtering, and morphological correction are then performed on the image to obtain a planar nuclear mask. In one implementation, the nuclear labeling channel is the DAPI nuclear staining channel.

[0037] When obtaining multi-layer kernel anchoring data, thresholding and continuity constraints are applied to the kernel label responses at the same planar location along the Z-direction to obtain the estimated boundary. and Specifically, background subtraction and 0-1 normalization are first performed on the kernel label response of each layer. The threshold is taken as the background high quantile threshold and coefficient. Multiply by the larger of the axial peak values ​​at that location. The value can be 0.1–0.3, or it can be obtained by calibration using positive nuclei and background. Continuity constraints require that the boundary must fall within a continuous suprathreshold segment containing the axial main peak, generally requiring at least two consecutive suprathreshold layers, and allowing bridging of no more than one isolated dark gap. When the layer spacing is uneven, it should be judged according to the actual physical length. When only a single-layer nuclear image is available, the clear nuclear position is used as the axial reference center. Based on the objective lens, sample preparation method, and cell type, the nuclear thickness is statistically analyzed from representative multilayer nuclear sequences. The 95th percentile value or its one-sided confidence upper bound is taken, and then the localization, segmentation, and model errors are superimposed to form the estimated boundary. When there is no corresponding calibration, a globally conservative upper bound or supplementary short-layer nuclear sequences are used.

[0038] By superimposing a boundary uncertainty margin onto the estimated boundary, a conservative boundary is obtained:

[0039]

[0040] in, and These represent the lower boundary uncertainty margin and the upper boundary uncertainty margin, respectively, forming a conservative local kernel region:

[0041] The uncertainty margin should comprehensively consider one or more of the following: nuclear label boundary estimation error, nuclear segmentation error, channel axial offset calibration error, Z-axis positioning error, point diffusion tail, and monolayer nuclear thickness model error. When a position-by-position boundary cannot be reliably obtained, the global axial boundary of the target cell and its uncertainty margin are used as a degenerate form, and the boundary source and boundary confidence state are recorded; subsequent coarse screening coverage and evidence for excluding unobserved nuclear domains are limited to... Inside.

[0042] For probe channels with insufficient confidence in the conservative boundary and potential need for external tail measurements, a closed verification safety region is established along the Z-axis outside the conservative local kernel domain, denoted as:

[0043] in, This indicates that it only expands along the Z-axis. and These are the safety margins for the lower and upper closure verification, respectively, and both are non-negative. The safety margins are determined based on the channel axial point diffusion tail, fine scan layer spacing, Z-axis positioning error, equipment safety travel, cumulative exposure, and applicable constraints in the task budget. They can also be calibrated separately for different channels or different planar positions. The safe travel range of the equipment must not be exceeded. The portion of the closure verification safety domain located outside the conservative local kernel domain is called the closure verification safety extension zone. This extension zone is only used for field measurements of the predicted area of ​​existing trajectories, tail attenuation verification, and imaging quality assessment, and does not participate in the calculation of kernel domain coverage status or the exclusion status of unobserved kernel domains. Figure 2 As shown, the conservative local kernel region is located in the middle, with the upper and lower sides being the closure verification safety extension zones. These extension zones are only used for measured attenuation verification at the tail of existing trajectories and are not used for kernel region coverage or internal exclusion verification. When no verification outside the corresponding side boundary is required, the safety extension margin on that side can be zero. It should be noted that a safety extension margin of zero only indicates that there is no need for external boundary verification at initialization, not that it cannot be adjusted subsequently: if the trajectory later approaches or crosses the conservative kernel boundary but still does not close, the extension margin on this side is recalculated based on the point spread tail, fine scan interval, positioning error, remaining travel, exposure, and time budget of that channel; if there is usable margin, the layer continues to be extracted into the safety extension zone; if there is no margin or no attenuation evidence is found even after extending to the upper limit, the unclosed state is retained and the limited result is output.

[0044] First determine based on location feedback. Then, by superimposing the current channel axial offset, we obtain... and will , The actual positioning of each channel is mapped to a unified physical Z-coordinate. When there are resolution, clipping regions, or planar optical path offsets, scaling, translation, or distortion calibration parameters are used to map the planar kernel mask and boundary confidence state to the probe channels.

[0045] In this embodiment, the coarse screening sampling parameters are determined as follows: Establish channel sampling status and generate a coarse-screening sampling plan covering the complete conservative local kernel domain. The coarse-screening sampling parameter set is a set of parameters used to cover the complete conservative local kernel domain and establish signal exclusion evidence, including at least the maximum permissible interlayer spacing for coarse screening. The parameters to consider include: coarse screening exposure time, camera gain, light source output range, detection threshold, and allowable noise level.

[0046] Based on probe channel The lower confidence limit of the detectable axial width of the lowest standard signal under effective imaging parameters Z-axis comprehensive positioning error and safety margin It is determined that, in one implementation, the following is satisfied:

[0047] When using a one-sided confidence lower bound, a 95% confidence level is typically used, but it can also be set between 90% and 99% depending on the acceptable risk of missed detections. For each "objective-channel-exposure / gain setting" combination, it is recommended to acquire at least 30 independent standard signals, or continuously increase the sample size until the change in the lower confidence bound is less than the set tolerance. The detectable width distribution is determined using standard slides, calibrated microspheres, historical qualified samples, or a combination of point spread function and camera noise model. The minimum target intensity is the lowest background subtraction response that should still be considered a valid FISH signal in positive standard slides or historical qualified samples, and should be higher than the high quantile threshold of negative background. The most unfavorable sampling phase is calculated one by one based on the different offsets of the signal center relative to the sampling grid within one layer spacing, including at least half a layer spacing offset, and the smallest detectable width among all phases is taken. This ensures that signals within the conservative kernel region and reaching the preset minimum target intensity are detected by at least one high-quality coarse screening layer under the set confidence level and the most unfavorable sampling phase. The coarse screening exposure can be lower than the fine scanning exposure, but must not be lower than the lower limit of the parameter that can detect the lowest standard signal after calibration.

[0048] when If the current minimum executable layer spacing still does not meet the above constraints, the current parameters cannot form a reliable sparse coarse screening spacing. Instead of establishing a sparse coarse screening exclusion plan, the system will reduce the layer spacing, improve the exposure or signal-to-noise quality, recalibrate, switch to the preset dense complete layer scan, or mark the corresponding range as unable to form exclusion evidence and enter a restricted rollback.

[0049] A set of physical Z-positions for coarse screening is generated between the lower and upper axial boundaries of the conservative local kernel domain. The set of positions can be arranged at equal intervals or non-equal intervals based on the local kernel thickness, device movement direction, and known channel offset. However, the interval between any adjacent positions must not exceed the maximum allowable interlayer spacing at that location, and both the conservative upper and lower boundaries must fall within the effective detectable coverage area. Coarse screening positions can be executed alternately from the axial reference center to both the upper and lower sides, or in a unidirectional movement sequence. Regardless of the actual acquisition order, all images and candidate signals are recorded and sorted according to the actual physical Z-coordinates. The set of coarse screening positions represents the coverage constraint that must ultimately be satisfied; it does not require that all coarse screening layers be completed before candidate verification or fine scanning can be performed.

[0050] In this embodiment, the normalization layer response is obtained as follows: The normalized layer response is the response of the original layer image after background subtraction and elimination of the influence of acquisition parameter variations. For the original intensity... First estimate the background Then through the channel calibration function Converted to:

[0051] in, For the exposure time, For camera gain, This refers to the output ratio of the light source. The remaining fluorescence percentage is estimated based on the cumulative exposure or acquisition time, and satisfies... .

[0052] Within the linear response range, It can be represented as:

[0053] in, is the gain response coefficient. Indicates channel Gain setting The response ratio at a given gain level is a coefficient within the camera's linear operating range for a fixed gain level. When calibrating across multiple gain levels, it is represented by a lookup table or function composed of discrete calibration points. During calibration, a stable, uniform fluorescent standard film or standard light source is used, with fixed exposure time and light source output. Dark and bright field data are repeatedly acquired for each gain level. After subtracting the dark field, the median ratio or regression slope of the response at that gain level relative to the reference gain response is used as the reference value. Different objectives, channels, and camera readout modes are saved separately, and no linear extrapolation is performed when the range exceeds the calibration range or saturation occurs.

[0054] When the gain is nonlinear or close to saturation, calibration lookup tables or piecewise mapping are used. During table lookup, stable standard sources are acquired for multiple exposure times, gain levels, and light source output levels under fixed channels and objectives. The digital response after background subtraction and the reference irradiance are recorded. The median value of multiple acquisitions is taken for each grid point, and fluctuation and saturation indicators are saved. During runtime, interpolation is only performed between adjacent valid grid points. If the value is below the noise lower limit, enters the saturation region, or exceeds the calibration range, that layer is not included in the evidence judgment; the exposure or gain is adjusted and the data is re-acquired. Segmentation nodes are preferentially selected from the camera's native gain level, exposure / light source control level, and locations where the slope of the response curve changes significantly. Monotonic piecewise linear fitting is performed on the calibration points, minimizing the number of nodes while ensuring the relative error of each segment does not exceed 1%–5%. Linear interpolation is performed between adjacent nodes, and cubic interpolation to maintain monotonicity is used when necessary. Recalibration is performed after equipment maintenance, changes in objective or camera mode, or when light source drift exceeds limits.

[0055] When no validated photobleaching model is established Independent photobleaching correction is not performed. Background estimation is mainly based on grayscale morphological opening operations, but can also be achieved using the rolling sphere method. The structuring element or rolling sphere radius is taken as 2 to 5 times the maximum equivalent radius of the candidate fluorescent spot to avoid absorbing point signals into the background surface. Saturated pixels and known bad pixels are masked before calculation. If the illumination field only changes slowly at low frequencies, second-order or third-order polynomial surface fitting can be used after removing candidate regions. The specific method and radius are determined by the background residual of the negative control and the retention of the positive spot response. Regardless of the method used, the results of adjacent layers are compared at the same response scale.

[0056] In this embodiment, normalization does not apply independent photobleaching correction when a validated photobleaching model has not been established, in order to maintain the objectivity of response comparison and avoid introducing additional errors from an unvalidated model; once a photobleaching model has been established and validated, normalization is then applied based on... Perform corrections.

[0057] In this embodiment, the preliminary screening and evidence generation are as follows: The physical Z-position of the coarse sieve is sequentially converted into a Z-axis control command. After the equipment is in position and the vibration stabilizes, the current probe channel image is acquired, and the command position, actual position, exposure, gain, light source status, and acquisition time are recorded simultaneously. In one implementation, the position determination criterion adopts a combination of "position feedback + continuous stabilization time + calibration delay": after Z-axis movement, the motion state is waited for to end; when position feedback is enabled, the absolute value of the position error must not exceed a preset position error upper limit. The system then verifies that the actual position has reached the target position; after that, it waits for a configurable continuous stabilization delay (in one exemplary configuration, this continuous stabilization delay is configured to 100ms); devices without position feedback can use a pre-calibrated fixed delay. If the wait times out or the position error exceeds the limit, the layer is treated as having invalid quality and is retried.

[0058] The quality of the current layer is assessed based on Z-axis positioning error, saturation ratio, background fluctuation, signal-to-noise level, light source stability, and image integrity. If the quality is invalid, priority is given to repositioning, adjusting exposure or gain, and then re-imaging. If re-imaging is still invalid, the location is marked as a quality gap, and the kernel coverage or unobserved kernel exclusion status should not be updated based on this.

[0059] Estimating non-uniform backgrounds from high-quality original images and generating normalized layer responses, only when... Candidate signals are detected within the registration tolerance range. Candidate signals are fluorescent bright spots that meet intensity, area, shape, and signal-to-noise (SNR) conditions, located within the conservative local kernel domain and registration tolerance range in the normalized layer response of effective quality. Candidate signal detection is performed in the following order: First, the background is estimated and normalized within the kernel domain and registration tolerance range. Then, the standard deviation is calculated using the background median plus 3–6 times the median absolute difference (MAD) to form the intensity threshold. Alternatively, the high quantile value of the negative control can be used to control the false positive rate. Subsequently, connected component labeling is performed, isolated hot pixels are removed, and peak value, integral response, area, SNR, roundness, and aspect ratio are calculated. The SNR is generally not lower than 3, and 4 is used for greater stability. The area is set to 0.25–4 times the effective area of ​​the calibrated diffuse spot. Objects with a roundness lower than 0.3 or an aspect ratio greater than 3 are not directly deleted but are transferred to adhesion assignment or awaiting verification. Each threshold is calibrated separately according to the objective lens, channel, and pixel size. Extranuclear bright spots are preserved as interference evidence but do not participate in the construction of intranuclear trajectories.

[0060] When candidate signals are detected, an initial candidate axial interval is formed centered on the physical Z-position. When no candidate signals are detected, blank exclusion evidence is formed only when the current layer quality is effective, the kernel domain mapping is reliable, and the normalized response of the corresponding coverage area is measured to be lower than the detection threshold. This is to characterize that there is no signal in the corresponding kernel axis sub-region that meets the preset minimum detectable condition. Weak responses, single-pixel bright spots, saturated large areas, objects with uncertain shapes, or abnormal detection processes are not directly used as blank evidence, but rather form candidates to be verified or observational gaps.

[0061] After each coarse screening layer is completed, the current coverage area, insufficient evidence area, and candidate axis area are updated. The system can continue to execute the next coarse screening layer, or it can first perform verification on weak candidates, perform fine scanning on clear candidates, or perform resampling on locations with invalid quality. Regardless of the staggered sequence used, the requirements for coverage of the complete conservative core domain and exclusion of evidence must still be met before normal termination.

[0062] Candidate axial intervals are the intervals to be refined in a unified physical Z-coordinate system, formed by expanding the range around the corresponding coarse screening position after one or more candidate signals are detected in the coarse screening stage. The expansion width is determined based on the channel axial point diffusion width, the spacing between coarse screening layers, the Z-axis positioning error, and the conservative kernel boundary. Candidate axial intervals that overlap or are less than a preset merging distance are merged. During merging, the currently obtained coarse screening and verification results are sorted according to the actual physical Z-coordinate, and candidate axial intervals triggered by the same or adjacent layers and that overlap are merged. The merged intervals are usually trimmed to [a smaller size]. When an unclosed trajectory requires measured evidence of attenuation outside the boundary, extension to the corresponding side is only permitted. And it must not exceed this range. The candidate axial interval is located in The portion outside of this is marked as the closed verification extension segment. The closed verification extension segment is only used for the actual measurement of the prediction area of ​​the existing trajectory and the tail attenuation verification. It is not included in the range calculation of the nuclear domain coverage, excluded intervals or intervals with insufficient evidence, nor is it used to establish new intranuclear candidate signals.

[0063] Complete The axial intervals are divided into candidate axial intervals, axial intervals with calibrated blank exclusion evidence, and intervals with insufficient evidence; candidate intervals cannot be converted into blank exclusion intervals before the fine scan and trajectory determination are completed. If there are more than [a certain number of] positions between the coarse screening positions... If there are gaps, uncovered boundaries, quality gaps, or candidates to be verified, the corresponding range is marked as insufficient evidence and will be prioritized for supplementary sampling in subsequent decisions.

[0064] In this embodiment, the fine scan and axial signal trajectory are constructed as follows: Generating the physical Z-position of the fine scan and the fine scan interlayer spacing within each candidate axial interval. The interlayer spacing should be less than or equal to the maximum allowable interlayer spacing for coarse screening, and determined based on point diffusion response, adjacent signal separation requirements, and equipment positioning capabilities; a smaller interlayer spacing should be selected when candidate responses are strong, there may be adhesion, or the peak position is unclear. Layer-by-layer execution includes quality assessment, background subtraction, and response normalization. Invalid quality layers are processed according to the aforementioned quality re-acquisition path and cannot be used as evidence of signal attenuation or trajectory closure. A detection threshold is formed based on local background fluctuations, camera noise, and the lowest signal calibration. Candidate bright spots are extracted from the inner core, and the planar centroid, coverage area, normalized peak response, normalized integral response, area, and shape descriptor are calculated. For the closed verification extension segment, new candidates within the kernel are not extracted independently; instead, the normalized support response is directly measured according to the predicted region of the existing trajectory. Objects whose area significantly exceeds the signal point calibration range, elongated reflective areas, large saturated areas, or lack consistency with adjacent layers are marked as noise pending determination. Their positions and responses are retained for subsequent verification, and their regions are not immediately converted to blank exclusion states.

[0065] Axial signal trajectories are ordered sets formed by associating candidate signals from adjacent Z-positions that are determined to originate from the same physical fluorescent target, after sorting them according to their actual physical Z-coordinates. Trajectories The normalized axial response sequence is denoted as The adjacency relationship is determined by the difference in physical Z-position, not by the order of acquisition time. All valid layers within the current candidate axial interval are reordered according to their actual physical Z-coordinates, and candidate associations are established only between layers whose physical Z-position difference does not exceed the upper limit of association; candidates with adjacent acquisition times but not adjacent physical Z-positions cannot be directly associated.

[0066] For candidate signals and Calculate the association cost:

[0067] in, The distance between the centroids of the plane. The correlation radius is adjusted according to the physical Z-interval and registration error. To normalize the overlap of candidate regions, Differences in area, shape, and normalized response continuity, The physical Z-position difference, The maximum correlation interval, to Non-negative weights are used. Weights and association upper limits are determined through device calibration samples or provided by configuration. The association combination with the lowest total cost and no conflicts is selected; alternatively, nearest neighbor, bipartite graph matching, state estimation, maximum likelihood association using point spread models, or graph optimization can be used as equivalent implementations. Associated candidates are added to the same axial signal trajectory according to their actual physical Z-position; candidates that cannot be associated are established with new trajectories, and the observation state at the corresponding position is recorded as a candidate match.

[0068] The trajectory layer observation state is the trajectory At the physical Z position The description of the types of evidence obtained at each location includes three categories: candidate matches, measured data in the predicted region, and missing observations. For locations where existing trajectories have no matching candidates in the effective quality layer, the predicted region is determined based on adjacent matched locations, correlation radii, and channel registration errors. In one implementation, the prediction region is circular by default: first, the target plane position is predicted based on the two most recent valid matching centroids, and then... Calculate the radius, where It is the equivalent radius of the diffusion patch or the nearest candidate in the channel. It is a robust upper bound for the relative Z-difference in the centroid displacements of adjacent layers. This is the upper bound of the channel registration error. This represents the nearest valid matching location. If the registration error has a clear directionality or the candidate itself is elliptical, then it is changed to an elliptical region with the major and minor semi-axes superimposed with the prediction errors in the x and y directions, respectively.

[0069] If the predicted region can be reliably mapped, then the normalized support response is measured directly in that region:

[0070] The observed state is recorded as the actual measurement of the predicted area. For the regional measurement function, the regional peak value, background subtraction integral response, or robust quantile value can be used. These three measurement methods are selected during the channel calibration stage and are not switched layer by layer within the same trajectory: the regional peak value is used for isolated, unsaturated bright spots; the background subtraction integral response is used when the profile is stable but the peak position may drift slightly, i.e., it is calculated within the prediction area. When hot pixels or local outliers exist, the 90th to 99th percentile values ​​are used, minus the median of the neighboring background. The final selection is determined by the sensitivity, coefficient of variation, and false positive rate on positive and negative calibration samples, and recorded in the channel parameters. Missing observations indicate invalid image quality, unreliable predicted region mapping, correlation conflicts, or registration errors, thus preventing the acquisition of a response suitable for attenuation assessment.

[0071] Only when the measured response in the predicted region is lower than the background or relative attenuation threshold can it be used as evidence of attenuation; only when it is located within a conservative local kernel region and meets the minimum detectable condition can it be used as evidence for blank exclusion verification. The absence of a matching candidate alone should not be automatically recorded as a low response; missing observations should not be used as evidence of attenuation, blank status, or trajectory termination.

[0072] When two neighboring signals become stuck in a single layer, the stuck region is multi-objectively assigned using multiple centroid positions, response peak numbers, and region shapes from the preceding and following layers to avoid permanently merging the two physical signals. In one implementation, the stuck region is first seeded based on the local maxima of the distance transform and the predicted centroids from the preceding and following layers, and then split into multiple sub-regions using watershed or constrained region growing. Subsequently, a cost matrix is ​​established between the existing trajectory predictions and the sub-regions, with costs including centroid distance, response continuity, and shape and area differences. The Hungarian algorithm is used to assign each sub-region one by one. Pairings exceeding the association radius or response / shape change thresholds are directly prohibited. If the number of seeds is inconsistent with the number of trajectories, or if the optimal assignment still exceeds the threshold, the layer is marked as having an association conflict or missing observations, and the adjacent Z layer is resampled.

[0073] In this embodiment, the trajectory closure state is updated as follows: Trajectory Form a normalized axial response sequence arranged according to the actual physical Z-position. The candidate matching location uses the candidate region response, and the predicted region's measured location uses... For locations with missing observations, only gap markers are retained without filling in low responses. The current peak location is determined based on the sequence with valid observations. and peak response In one implementation, a three-point weighted median based on the actual physical Z-neighborhood is used as a peak robustness parameter: for Without crossing any missing observation points, take the nearest valid observations above and below, and then sort them according to distance. Distance weighting is used; when the layer spacing is uneven, only observations falling within a window of 1 to 1.5 times the half-width of the diffusion point in that channel are used. The peak value is taken as the actual Z position corresponding to the maximum value of the smoothed sequence, and the weighted center of the plateau is taken when a plateau appears. Other robust processing methods such as median smoothing of adjacent positions, confidence intervals, or point diffusion response models can also be used. Smoothing is only used for locating stable peak regions, and closure is still determined by the original measured response satisfying the attenuation threshold at continuous positions.

[0074] Let the relative attenuation threshold be The background termination threshold is When there are sampling locations with effective quality and clear observation status on both the lower and upper sides of the peak. and and satisfy as well as At that time, evidence of attenuation on the lower and upper sides is formed respectively. Determined by the axial point diffusion calibration of the channel. The attenuation condition is determined by normalized background fluctuations and the lowest detectable response; locations with missing observations should not be substituted into the above inequalities. To avoid false closures caused by accidental dark layers, the attenuation condition can be required to hold at a predetermined number of consecutive physical Z-positions, or the overall trend from the peak to the boundary can be required to decrease and the boundary response can fall within the background confidence range; the number of consecutive positions is determined based on the fine scan layer spacing and the axial point diffusion width.

[0075] When the conservative local kernel boundary corresponding to the trajectory centroid has been covered by high-quality sampling, the trajectory has obtained a measured response below the background termination threshold before reaching the conservative boundary, and the boundary confidence state meets preset requirements, then this side forms kernel domain boundary termination evidence. If the boundary confidence is insufficient, the conservative boundary cannot constitute closure evidence on its own; the system should... At least one or a predetermined number of effective extended layers of mass are obtained on the unclosed inner edge, and measured attenuation responses are obtained within the existing trajectory prediction region. The low response of the extended layers is only used for trajectory tail attenuation verification and not for... Proof of internal kernel domain coverage or exclusion; achieving If the boundary still cannot form effective evidence of attenuation, the unclosed state of that side shall be retained.

[0076] Trajectories that simultaneously obtain attenuation evidence from both the upper and lower sides or termination evidence from the core domain boundary are updated to bilateral closure; those lacking evidence from only one side are updated to the corresponding side not closed; those lacking evidence from both sides are updated to bilateral not closed; short trajectories that consistently fail to meet the morphological, signal-to-noise, or cross-layer continuity conditions remain under noise pending until verification is complete. Trajectory closure can also be achieved using Gaussian response, point spread response, or other single-peak models, with the model's tails falling into the background confidence range as closure evidence; model fitting is an alternative to trajectory closure and is not a necessary step in determining the central focal plane.

[0077] In this embodiment, the joint update of the kernel domain coverage status, the unobserved kernel domain exclusion status, and the channel imaging quality status is as follows: The coverage interval of each effective quality layer is calculated based on its actual physical Z-position and detectable axial coverage range, and only the total coverage interval is considered. The union of intersecting portions. If there are sampling gaps exceeding the maximum permissible interlayer spacing, uncovered conservative kernel boundaries, or quality invalid locations, the kernel domain coverage remains incomplete; the closed verification safety extension band does not increase the kernel domain coverage.

[0078] From complete The process involves deducting unclosed candidate axial regions, noise-pending regions, observation-missing regions, and quality-invalid regions. For the remaining axial sub-regions, it is checked whether they are covered by measured blank evidence of valid quality that meets the preset minimum detectable condition. The unobserved core region exclusion state is only complete when there are no sub-regions with insufficient evidence; blank or low-response areas in the closure verification safety extension band are not included in this state calculation. After all objects in a candidate axial region are verified as noise, the region can only be converted to an excluded region if it simultaneously has a valid blank verification layer that meets the minimum detectable condition; it should not be automatically considered blank due to the deletion of candidate trajectories.

[0079] Check whether all necessary layers involved in coverage, exclusion, and closure determination meet the channel imaging quality requirements. The channel imaging quality status is determined at least based on Z-axis positioning error, image saturation ratio, background fluctuation, signal-to-noise ratio, exposure and gain effectiveness, light source stability, channel registration error, and image integrity. If there are incomplete re-imaging issues such as Z-axis misalignment, saturation, abnormal background, insufficient signal-to-noise ratio, light source drift, out-of-tolerance channel registration, or image acquisition failure, the channel imaging quality status remains incomplete. Layers with invalid quality cannot be used to prove blanks, attenuation, kernel coverage, or trajectory closure.

[0080] In this embodiment, the combined stopping conditions, supplementary mining decisions, and restricted rollback are as follows: make This indicates that evidence of bilateral closure has been obtained for all non-noise trajectories. express Evidence has been obtained to exclude all areas that have not yet been observed. This indicates that the kernel domain coverage status is complete. This indicates that the channel imaging quality status is complete. and Corresponding to two types of core evidence, and This serves as a threshold condition for the validity of both types of evidence. Only if:

[0081] Only when established is the current probe channel allowed to stop normally. Before a valid trajectory has been formed, Treat it as an empty set, but it must still satisfy the following conditions: , and Only when sufficient acquisition of candidate signals is achieved can a state of no candidate signal acquisition be formed; failure to detect candidates in a single layer or completion of only local kernel domain acquisition cannot trigger a stop.

[0082] When the joint stop condition is not met, measures are taken to fill the gaps in the coarse screen. Candidate acquisition actions are generated for internal expansion of unclosed trajectories, densification of candidate intervals, quality resampling, and completion of conservative boundary coverage. Each action includes at least the action type, target physical Z-position or interval, required imaging parameters, expected reduction in evidence gaps, and equipment cost. Internal coarse screening failure layers, large gaps within candidate intervals, and in-situ resampling are all treated as independent candidate actions and are not forcibly assigned to the upper or lower boundary.

[0083] When multiple candidate actions exist simultaneously, they can be prioritized according to their evidentiary validity. One implementation employs a two-level selection strategy: first, actions that are safe and executable and can address the current highest-priority evidence gap are selected; then, actions of the same priority that all meet the evidentiary requirements are compared to determine the expected evidence increment; finally, Z-axis travel, exposure, and time cost are compared. Alternatively, an action-level indeterminacy can be constructed.

[0084] in, , , and These represent the expected reductions in the trajectory closure gap, the unobserved kernel exclusion gap, the kernel coverage gap, and the quality gap, respectively. Indicates the cost of device movement, exposure, time, or buffering. , , , and These are non-negative weights determined through preset configuration or historical calibration; the aforementioned quantities are determined by the number of unclosed trajectories covered by the action, the volume or axial length of insufficient evidence, the size of the coverage gap, the degree of quality invalidity, and the expected verification results. Indeterminacy evaluation can be used only for ranking candidate actions of the same priority. Equipment motion and exposure costs are secondary selection factors only when all candidate actions meet the evidence requirements. Boundary indeterminacy calculated for extensions above or below the trajectory is a specific implementation of this action-level evaluation.

[0085] The selected next physical Z-position, channel, exposure, gain, and action type are converted into Z-axis drive, channel switching, and camera acquisition commands. After the position is reached and stabilized, the new layer is acquired. New layers used to fill gaps in the coarse screening or cover conservative boundaries return to the coarse screening acquisition step. New layers used for weak candidate verification, candidate interval densification, or expansion of unclosed trajectories return to the fine scanning step. After quality resampling is completed, the process returns to the original step that generated the resampling action.

[0086] When a single layer is of invalid quality, the camera fails to capture an image, or the Z-axis is not in position, the abnormal layer is not considered as evidence of blankness, attenuation, or coverage. The system will re-enter the position within the allowed number of retries, adjust the exposure or gain, and then re-sample. If the original position is unreachable, an alternative physical Z-position can be selected that is no more than a safe margin away from the original position, and the maximum allowable interlayer spacing will be rechecked.

[0087] A restricted state is formed when the next position exceeds the allowable range of the corresponding action, the safe travel range of the equipment, the maximum number of layers, the cumulative exposure, or the task time budget. The evidence calculation range for coarse screening coverage or core area exclusion actions is limited to... The acquisition range for trajectory closure verification actions is limited to If equipment conditions permit, then... If the evidence is insufficient within a given region, perform a pre-defined dense full-layer scan or a local resampling with reduced interlayer spacing. If the evidence still cannot be supplemented, retain the uncovered region, the unexcluded region, the unclosed trajectory, or the quality gap, and output an uncertain state. Restricted rollback must not automatically consider equipment boundaries, failed layers, or blank layers that have not been quality verified as signal attenuation or core region exclusion evidence.

[0088] In this embodiment, the data acquisition results output and resource release are as follows: For each bilaterally closed trajectory, a cross-layer deduplication signal is output. The planar position is determined by the weighted position of the response of the centroid of each layer of the trajectory. The Z-position represents the normalized response peak position or the weighted center of the peak neighborhood, and the response represents the normalized peak response or the peak neighborhood integral response. Signals with similar planar positions but separable trajectories, peak Z-positions, or response sequences are output separately; candidates for the same trajectory appearing across multiple layers are output only once. The output content is the intermediate technical result of acquisition and image processing, and does not directly generate disease diagnosis or health status conclusions.

[0089] When there is no valid trajectory and the joint stopping condition is met, the output is a state of sufficient acquisition of no candidate signals; when there is an unclosed trajectory, an uncovered kernel region, an unexcluded kernel region, a quality gap, or a restricted backoff that has not been completed, the output is an uncertain state and the corresponding gap, and the state is not converted into a definite negative conclusion.

[0090] Generate representative images for display or verification. These representative images can be locally synthesized peak layers of each closed trajectory, multi-layer fusion images constrained by the trajectory, or kernel images with deduplication signal position markers.

[0091] When the original layer no longer participates in unclosed trajectory judgment, coverage exclusion verification, representative image generation, or anomaly verification, its image cache is released; after the current channel ends, the channel acquisition results are retained and the caches of the remaining layers are released. The channel acquisition results can save the complete effective layers, save only the trajectory peak and evidence layers, save the fused image constrained by the trajectory, or transfer the original layers to external storage and then release the memory; different saving methods do not change the joint stopping mechanism formed by dual evidence and coverage and quality gating.

[0092] In this embodiment, the non-limiting example parameters are as follows: The following are non-limiting example parameters for illustrating the calculation relationship only. These example parameters are only used to clarify the value transfer relationship of the formula and do not constitute a numerical limitation on the scope of protection of this invention, nor do they represent the measured performance of any specific device.

[0093] Assuming the estimated axial boundary of the target cell nucleus is The region is defined as follows: 3.0 μm to +3.2 μm, with an uncertainty margin of 0.2 μm at both the upper and lower boundaries. The conservative local kernel region is: 3.2μm to +3.4μm. If the current channel's 1.4μm 0.1μm and If it is 0.2μm, then according to The maximum permissible interlayer spacing for coarse screening was determined to be 1.0 μm. The system covers conservative local kernel regions with adjacent spacing not exceeding 1.0 μm. After weak candidates are found near +2.8 μm, candidate verification and fine scanning can be performed with an interlayer spacing of 0.25 μm. If the upper boundary confidence is insufficient, an upper closure verification safety margin is set. If the depth is 0.5 μm, then the upper bound of the safe zone for closure verification is +3.9 μm. Low-quality responses obtained between +3.4 μm and +3.9 μm are only used for attenuation verification on the upper tail of the trajectory and do not increase the coverage or exclusion range of the conservative core region. The joint stopping condition is met when the candidate forms a peak with effective measured attenuation on both sides, the corresponding trajectory closes on both sides, and the remaining conservative core regions are all excluded by evidence formed by high-quality coarse screening layers. If any coarse screening layer is of invalid quality or the trajectory prediction region only shows observational gaps, then the corresponding range remains insufficiently supported by evidence and triggers resampling, without normal stopping.

[0094] In this embodiment, the multi-channel expansion implementation is as follows: Nuclear anchoring data can be reused across multiple probe channels for the same target cell, but each probe channel maintains its own coarse screening parameters, normalization function, candidate interval, trajectory status, coverage exclusion status, and channel acquisition results. Multiple channels share a unified physical Z-coordinate, but the axial offset, point spread response, minimum detectable width, exposure parameters, and quality thresholds can differ between channels, thus allowing different channels to obtain different actual Z-sequences and layers.

[0095] Multiple probe channels can perform coarse and fine scanning sequentially, or multiple channels that have not yet stopped can be acquired at the same physical Z-position. In one implementation, each channel that has not yet stopped first generates the next required position in the sample's physical coordinates. Then convert the axial offset of the channel into mechanical position. Use when color difference varies with the plane position. Each requirement corresponds to an allowable error range. When the ranges of multiple channels intersect, the position closest to the current Z-position or with the smallest total error is selected from the intersection, and all active channels that meet the error requirements are sequentially acquired at that position. If there is no intersection, the channels are moved separately. The next position is still selected first according to the evidence gap priority, then the number of channels whose evidence gaps can be reduced simultaneously is considered, and finally the Z-stroke, channel switching, and exposure cost are compared. After acquiring each channel, the channel status is updated with the actual position. Channels that have met the stopping conditions are no longer exposed.

[0096] In this embodiment, the alternative implementation method is as follows: The coarse screening position can be covered by equal-spacing full-core domain coverage, alternating coverage from the center to both sides, coverage along a single direction, or non-equal-spacing coverage based on the local core thickness; regardless of the order, the coverage gap is calculated based on the actual physical Z coordinate and meets the calibrated maximum allowable interlayer spacing.

[0097] Coarse screening and fine scanning can be performed in a strict two-stage sequential order, or coarse screening, weak candidate verification, fine scanning, and quality resampling can be performed alternately after the evidence is updated layer by layer. This staggered scheduling does not change the conditions that conservative core area coverage, exclusion of unobserved core areas, trajectory closure, and channel quality verification must be completed before normal stopping. The system can complete all planned coarse screening layers first, and then uniformly execute candidate interval fine scanning; alternatively, coarse screening, weak candidate verification, and fine scanning can be performed alternately based on incremental evidence. After each layer is completed, the coverage exclusion map and candidate action set are updated, clarifying whether candidates can immediately enter verification or fine scanning, and then returning to the incomplete coarse screening coverage. Both execution methods have the same evidence establishment conditions and joint stopping conditions; the only difference lies in the scheduling order of the acquisition actions.

[0098] In equipment where reliable nuclear morphology priors and signal detection models have been established, the exclusion state of unobserved nuclear regions can also be formed by actual sampling evidence and validated model inferences; model inferences only cover the applicable range that has been calibrated and validated, and the uncertain areas of the model still need to be sampled in practice.

[0099] Coverage exclusion diagrams can be presented in a one-dimensional form organized along axial intervals, or in a three-dimensional kernel domain evidence diagram. Each voxel or voxel block records the candidate, excluded, insufficient evidence, or quality invalid status; the three-dimensional form can improve the spatial accuracy of local candidate association and excluded evidence, and for devices that support region of interest exposure, local illumination, or local readout, it can also be used to perform supplementary sampling only for local areas with insufficient evidence.

[0100] Candidate extraction can employ threshold connected components, scale-space speckle detection, template matching, point diffusion model fitting, or lightweight segmentation models; cross-layer association can employ nearest neighbor, bipartite graph matching, state estimation, maximum likelihood association, or graph optimization, as long as the same physical signal can be organized into an ordered trajectory based on the actual physical Z position.

[0101] Limited rollback can be achieved by using preset dense full layer scanning, reducing interlayer spacing for local resampling, increasing or decreasing exposure for post-exhibition verification, switching to backup channel calibration, entering the manual verification queue, or only outputting an uncertain state; no rollback method will automatically treat invalid quality layers or unreachable equipment boundaries as exclusion evidence.

[0102] Example 2 Based on the same concept, this invention also proposes a FISH adaptive Z-layer acquisition device based on nuclear domain exclusion evidence and fluorescence signal axial trajectory closure feedback. The system of this embodiment includes a microscopic imaging device, a Z-axis drive mechanism, a channel switching mechanism, a camera, a device status acquisition unit, and a processor. The processor executes the method of the above embodiment one.

[0103] The microscopic imaging device can move along the X, Y, and Z axes to locate and image target cells in an in vitro slide; the Z-axis drive mechanism is used to perform motion control in the Z direction; the channel switching mechanism is used to switch between the nuclear labeling channel and one or more probe fluorescence channels; the camera is used to acquire images of each channel; and the device status acquisition unit is used to acquire device status information such as device position status and position feedback.

[0104] The processor includes the following functional units: Conservative nuclear domain and closure verification safety domain establishment unit: used to determine the conservative local nuclear domain of the target cell with boundary uncertainty margin based on the nuclear anchoring data obtained by the nuclear labeling channel in a unified physical Z coordinate system, and to set the closure verification safety domain along the Z axis outside the conservative local nuclear domain.

[0105] Coarse screening planning unit: Used to determine coarse screening sampling parameters based on the detectable axial width confidence lower bound of the lowest standard signal in the probe fluorescence channel, the comprehensive positioning error of the Z-axis, and the safety margin, and generate a coarse screening sampling plan covering the complete conservative local nuclear domain.

[0106] Quality assessment unit: Used to evaluate the imaging quality of each sampled layer image in terms of Z-axis accuracy, saturation ratio, background fluctuation, signal-to-noise level, and light source stability.

[0107] Candidate interval generation unit: used to generate candidate axial intervals to be collected by fine scanning in a unified physical Z coordinate system based on the candidate signals detected by the coarse screening layer.

[0108] Trajectory observation unit: used to establish trajectory layer observation status including candidate matching, measured and missing observations in the predicted region, and to measure the normalized support response when the predicted region can be reliably mapped.

[0109] Trajectory construction unit: used to associate adjacent candidate signals sorted by actual physical Z coordinate to construct axial signal trajectories, and to perform multi-target splitting and allocation when single-layer adhesion is used.

[0110] Closure Judgment Unit: Used to determine the trajectory closure status based on the relative attenuation degree of the axial signal trajectory on both sides of the peak, the background termination state, or the kernel domain boundary termination state.

[0111] Coverage exclusion assessment unit: used to dynamically maintain the coverage exclusion map, jointly assess the geometric coverage of conservative local nuclear domains and the measured blank exclusion status of unobserved nuclear axis subregions.

[0112] Candidate action evaluation unit: used to select the next acquisition action based on the expected reduction in evidence gap and equipment cost of each candidate acquisition action when the joint stop condition is not met.

[0113] Feedback control unit: It is used to convert the next acquisition action into a drive control command for the microscopic imaging equipment, and to control the stop acquisition when both the closure evidence of the axial signal trajectory and the exclusion evidence of the unobserved nuclear region are satisfied, and the coverage status and channel imaging quality status pass the gating conditions. It outputs cross-layer deduplication fluorescence signal based on the double-closed axial signal trajectory.

[0114] Each unit can be implemented by a software module in the same processor, or it can be distributed between the device control computer and the image processing computer, and passed through a single-layer image queue and control result message transmission. In one implementation, after each layer of image acquisition is completed, the acquisition thread submits the original image, the actual physical Z-position, and the imaging parameters. The processing thread returns one of the following control results: continue coarse screening, weak candidate verification, candidate interval fine scanning, trajectory expansion, quality re-acquisition, normal stop, or restricted stop. The device control thread only executes the next Z-axis action when it receives the continue acquisition result, thereby forming a closed loop between kernel domain exclusion evidence, axial trajectory closure evidence, coverage and quality gating, candidate actions, and device control.

[0115] The above units collaborate according to the following data and information flows: The conservative kernel domain and closure verification safety domain establishment unit transmits the established conservative local kernel domain and closure verification safety domain to the coarse screening planning unit; the coarse screening sampling plan generated by the coarse screening planning unit is processed by the quality assessment unit and the candidate interval generation unit, and then the candidate axial interval is submitted to the trajectory observation unit; the trajectory layer observation status output by the trajectory observation unit is processed by the trajectory construction unit and the closure judgment unit, and then submitted to the candidate action evaluation unit along with the kernel domain coverage and exclusion status of the coverage exclusion assessment unit; the candidate action evaluation unit passes the selected action to the feedback control unit to convert it into a driving command for the microscopic imaging equipment, and feeds back the acquisition results to the aforementioned units to form a closed loop.

[0116] Example 3 This embodiment proposes an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the steps of the FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback as described in Embodiment 1. The processor can be a central processing unit, a graphics processing unit, or other processing chip with program execution capabilities, and the memory can be random access memory, flash memory, or other non-volatile storage media capable of storing computer programs.

[0117] Example 4 This embodiment proposes a readable storage medium. This readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the FISH adaptive Z-layer acquisition method based on kernel domain exclusion evidence and fluorescence signal axial trajectory closure feedback as described in Embodiment 1. The readable storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, external hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0118] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0119] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0120] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 1 Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0121] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] The above embodiments are merely illustrative of several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A FISH adaptive Z-layer acquisition method, characterized in that, Includes the following steps: Nuclear anchoring data of target cells is obtained in the nuclear labeling channel. The device command position, actual Z-axis position and axial focal plane offset of each imaging channel are integrated into a unified physical Z coordinate system. Under the unified physical Z coordinate system, the main three-dimensional region of the target cell with boundary uncertainty margin is determined as a conservative local nuclear domain based on the nuclear anchoring data. In the probe fluorescence channel, the coarse screening sampling parameters are determined based on the lower bound of the detectable axial width confidence of the lowest standard signal in the channel under effective imaging parameters, the Z-axis comprehensive positioning error, and the safety margin. Coarse screening layer acquisition is performed within the conservative local core domain, and the normalized layer response of each layer is obtained. For coarse screening layers with effective quality, candidate signals are detected within the conservative local core domain. When no candidate signal is detected and the normalized layer response of the corresponding coverage area is measured to be lower than the preset detection threshold, blank exclusion evidence is generated to characterize that there is no signal in the corresponding core axis sub-region that meets the preset minimum detectable condition. Based on the candidate signals detected by the coarse screening layer, candidate axial intervals are constructed in the unified physical Z coordinate system, and fine scanning layer acquisition is performed. Adjacent candidate signals sorted according to the actual physical Z coordinates are associated to construct axial signal trajectories. The trajectory closure status is evaluated based on the normalized axial response of each axial signal trajectory, and the coverage status of the conservative local kernel, the exclusion status of the unobserved sub-regions within the conservative local kernel, and the channel imaging quality status are jointly updated. The joint evaluation uses the trajectory closure state and the exclusion state as core evidence, and the coverage state and the channel imaging quality state as validity gating conditions for a joint stop condition; when the joint stop condition is not met, a supplementary acquisition action is generated based on the evidence gap to drive the microscopic imaging device to acquire data until the joint stop condition is met, and the acquisition stops, and a cross-layer deduplication fluorescence signal is output based on the axial signal trajectory with double-sided closure.

2. The FISH adaptive Z-layer acquisition method as described in claim 1, characterized in that, Determining a conservative local nuclear region with boundary uncertainty margin for a target cell based on the nuclear anchoring data in a unified physical Z-coordinate system includes: extracting a planar nuclear mask based on a nuclear labeling channel image and obtaining the local axial estimation boundary of the target cell nucleus; superimposing the boundary uncertainty margin, which integrates boundary estimation error, nuclear segmentation error, channel axial offset calibration error, and Z-axis positioning error, on the local axial estimation boundary to generate the conservative local nuclear region. The method further includes: extending a non-negative safety margin along the Z-axis outside the conservative local core domain to form a closed verification safety domain, and defining the region outside the conservative local core domain as a closed verification safety extension band; wherein, the closed verification safety extension band is only used for the actual measurement of the prediction area and the tail attenuation verification of the existing axial signal trajectory, and does not participate in the calculation of the coverage state and the exclusion state.

3. The FISH adaptive Z-layer acquisition method as described in claim 1, characterized in that, The coarse screening sampling parameters include at least the maximum permissible layer spacing of the coarse screening. The maximum permissible layer spacing of the coarse screening is constrained by the upper limit value after deducting twice the Z-axis comprehensive positioning error and the safety margin from the lower limit of the detectable axial width confidence, so that the signal that is in the conservative local kernel region and reaches the preset minimum target strength is detected by at least one coarse screening layer with effective quality under the most unfavorable sampling phase. When the lower limit of the detectable axial width confidence is not greater than twice the sum of the Z-axis comprehensive positioning error and the safety margin, it is prohibited to establish a sparse coarse screening exclusion plan. Instead, measures such as reducing the interlayer spacing, increasing exposure or recalibrating after the signal-to-noise ratio, switching to a preset dense complete layer scan, or marking the corresponding kernel region as unable to form exclusion evidence should be performed.

4. The FISH adaptive Z-layer acquisition method as described in claim 1, characterized in that, The process of acquiring the normalized layer response includes: performing local background estimation and subtraction on the original images of each acquisition layer, and eliminating the influence of exposure time, camera gain response coefficient, light source output ratio, and fluorescence residual ratio on image grayscale through channel calibration; converting the response after background subtraction to a uniform scale through parameter normalization within the camera linear response range, and through pre-calibrated lookup table or monotonic piecewise mapping in the nonlinear gain range or near saturation range.

5. The FISH adaptive Z-layer acquisition method as described in claim 1, characterized in that, The construction and status assessment of the axial signal trajectory include: The effective acquisition layers within the candidate axial interval are sorted according to their actual physical Z coordinates. Based on a weighted cost function that includes the distance between the centroids of the plane, the degree of regional overlap, the difference in feature continuity, and the difference in actual physical Z position, a correlation is established and a trajectory is constructed between candidate signals whose physical Z position difference does not exceed the maximum correlation interval. An observation state for the trajectory layer is established, including candidate matching, measured prediction region, and observation missing. For existing axial signal trajectories at locations where no candidate signal is matched in the effective acquisition layer, the prediction region is determined based on adjacent matched locations and channel registration error. When the prediction region can be reliably mapped, the normalized support response is directly measured within the prediction region, and the observation state is recorded as the measured prediction region. Only when the measured response is lower than a preset threshold is it used as evidence of attenuation. When the current layer image quality is invalid, the prediction region exceeds the reliable mapping range, or there is a correlation conflict, the observation state is recorded as observation missing, and observation missing is prohibited from being used as a basis for attenuation or blanking. When adjacent signals are stuck together in a single layer, the local maxima of the distance transform are combined with the predicted centroids of the adjacent layers to generate a segmentation seed. After the region is split, the sub-regions are assigned to the axial signal trajectories by the allocation algorithm.

6. The FISH adaptive Z-layer acquisition method as described in claim 2, characterized in that, The evaluation trajectory closure state includes: Based on the effective observation of the axial signal trajectory under each actual physical Z coordinate, the peak position and peak response are determined, and sampling positions that meet the attenuation conditions are found below and above the peak position, respectively. When the normalized axial response at the sampling position is not greater than the peak response multiplied by the relative attenuation threshold and not greater than the background termination threshold, the lower attenuation evidence and the upper attenuation evidence are formed respectively. When the conservative local kernel boundary corresponding to the trajectory centroid has been effectively sampled and covered, the trajectory has obtained a measured response below the background termination threshold before reaching the boundary, and the boundary confidence state meets the preset requirements, kernel domain boundary termination evidence is formed; when the boundary confidence state is insufficient, an extended layer is obtained along the unclosed side within the closed verification safety domain, and the attenuated response is measured within the prediction region. When the axial signal trajectory simultaneously obtains evidence of attenuation on both sides, or obtains evidence of attenuation on one side and termination evidence of the nuclear domain boundary on the other side, it is determined to be a bilateral closure.

7. The FISH adaptive Z-layer acquisition method as described in claim 2, characterized in that, The joint stopping condition requires that the following conditions be met simultaneously: all non-noise axial signal trajectories have obtained evidence of bilateral closure; there are no unexcluded areas within the conservative local kernel that are not effectively covered by measured low response; the adjacent effective sampling interval of the conservative local kernel on the actual physical Z coordinate is not greater than the maximum allowable interlayer spacing and the upper and lower boundaries fall within the detectable coverage range; and all necessary layers participating in the determination meet the channel imaging quality requirements. When the current probe fluorescence channel has not yet formed an effective axial signal trajectory, the coverage state, the exclusion state, and the channel imaging quality state must still be satisfied in order to form a state of sufficient acquisition of no candidate signals. When the joint stopping condition is not met, candidate acquisition actions are generated, including filling the coarse screening gap, expanding the unclosed trajectory within the closed verification security domain, encrypting the candidate interval, quality re-sampling, and filling the conservative boundary coverage, and the action to be executed is selected based on the expected reduction in evidence gaps and equipment costs. When the next acquisition location exceeds the device's safe travel range, maximum number of layers, cumulative exposure, or time budget, it enters a restricted state. Dense layer scanning or local resampling is performed on the area with insufficient evidence. If the evidence still cannot be supplemented, the corresponding uncovered area, unexcluded area, unclosed trajectory, or quality gap is retained, and an uncertain state is output.

8. A FISH adaptive Z-layer acquisition device, characterized in that, include: The conservative nuclear domain and closure verification safety domain establishment unit is used to determine the conservative local nuclear domain of the target cell with boundary uncertainty margin based on the nuclear anchoring data obtained by the nuclear labeling channel in a unified physical Z coordinate system, and to set the closure verification safety domain along the Z axis outside the conservative local nuclear domain. The coarse screening planning unit is used to determine the coarse screening sampling parameters based on the detectable axial width confidence lower bound of the lowest standard signal in the probe fluorescence channel, the Z-axis comprehensive positioning error and safety margin, and generate a coarse screening sampling plan that covers the entire conservative local kernel region. The quality assessment unit is used to evaluate the imaging quality of each sampling layer image in terms of Z-axis accuracy, saturation ratio, background fluctuation, signal-to-noise level, and light source stability. The candidate interval generation unit is used to generate candidate axial intervals to be collected by fine scanning in the unified physical Z coordinate system based on the candidate signals detected by the coarse screening layer. The trajectory observation unit is used to establish the trajectory layer observation status, including candidate matching, measured prediction area and observation missing, and to measure the normalized support response when the prediction area can be reliably mapped. The trajectory construction unit is used to associate adjacent candidate signals ordered according to their actual physical Z coordinates to construct axial signal trajectories, and to perform multi-target splitting and allocation when the single layer is adhered. The closure judgment unit is used to determine the trajectory closure status based on the relative attenuation degree of the axial signal trajectory on both sides of the peak, the background termination state, or the kernel domain boundary termination state. The coverage exclusion assessment unit is used to dynamically maintain the coverage exclusion map and jointly assess the geometric coverage of the conservative local nuclear domain and the measured blank exclusion status of the unobserved nuclear axis sub-region. The candidate action evaluation unit is used to select the next acquisition action based on the expected reduction in evidence gap and equipment cost of each candidate acquisition action when the joint stopping condition is not met. The feedback control unit is used to convert the next acquisition action into a drive control command for the microscopic imaging device, and to control the acquisition to stop when both the closure evidence of the axial signal trajectory and the exclusion evidence of the unobserved nuclear region are satisfied, and the coverage state and the channel imaging quality state pass the gating conditions. Based on the double-closed axial signal trajectory, the cross-layer deduplication fluorescence signal is output.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the FISH adaptive Z-layer acquisition method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the FISH adaptive Z-layer acquisition method as described in any one of claims 1 to 7.

Images