CN-121998929-A - Sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability

CN121998929ACN 121998929 ACN121998929 ACN 121998929ACN-121998929-A

Abstract

The invention discloses a sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability, and belongs to the technical field of medical image processing and computer vision intersection. Aiming at the problems of unstable detection results, frequent cross-frame identity switching, single index dimension and the like in a complex scene of a microscopic video, the system comprises a track reliability evaluation module, a parameter set fusion evaluation module and the like, space and time calibration is carried out on the microscopic video, an improved RT-DETR model comprising ESC convolution, MSLA attention, SNI semantic injection and GSConvE fusion is constructed, high-precision detection of a small target is realized, track reliability is generated by combining Kalman prediction, statistical gating and Hungary matching, the comprehensive quality score is obtained by extracting kinematics and geometric parameters according to physical coordinates and carrying out normalized fusion, and screening judgment is completed under time window weighting and reliability gating.

Inventors

LIU YUANYUAN
LIU KAI
Sun Jiangtai
ZHANG JUN
WANG JINGZHI
WANG LONG
ZHU LU

Assignees

华东交通大学

Dates

Publication Date: 20260508
Application Date: 20260120

Claims (9)

1. A sperm quality evaluation and screening judgment method based on parameter set fusion and track reliability is characterized by comprising the following steps: S1, acquiring microscopic imaging video data, completing physical calibration of a space scale and a time scale, and establishing a time index taking a starting frame as a reference; S2, constructing an improved detection model based on RT-DETR, setting a convolution simulation self-attention unit in a backbone network stage, adopting a multi-scale linear attention mechanism in a feature interaction stage, carrying out semantic injection in a hybrid coding stage, and adopting a hybrid convolution unit to carry out feature fusion; S3, loading the trained detection model, detecting each frame of image, and outputting a sperm target set, position parameters and confidence coefficient thereof; S4, performing cross-frame correlation on continuous multi-frame detection results according to the time index, generating a motion track, distributing a unique track identifier, and calculating and maintaining track reliability according to the matching stability and the detection confidence of the track; S5, converting pixel coordinates and frame time of the track points into physical units according to the track information and the physical calibration parameters, calculating kinematic parameters according to the physical units, intercepting a target area by taking the position of a sperm target in a corresponding frame as the center, and extracting geometric shape parameters of the target area; S6, summarizing the kinematic parameters and the geometric shape parameters according to the unique track identification and the time index to form a parameter set, carrying out normalization processing on the parameter set, and generating a comprehensive quality score through a feature fusion algorithm; and S7, calculating a time weighted average value of the comprehensive quality scores in a preset time window when the track reliability is not lower than a preset threshold value, and outputting a screening judgment result of the corresponding sperms when the weighted average value is not lower than the preset score threshold value.
2. The sperm quality evaluation and screening determination method based on parameter set fusion and trajectory reliability as described in claim 1, wherein in step S1, said physical scaling parameters include spatial scale and temporal scale, said spatial scale is used for establishing quantitative mapping relationship between image pixel domain and physical spatial domain, respectively And (3) with Representing the actual physical length corresponding to the unit pixel length in the horizontal direction and the vertical direction, wherein the time scale is the inter-frame time interval Starting with video frame time Establish time index for reference, for the first Frame assignment unique time value 。
3. The sperm quality evaluation and screening judgment method based on parameter set fusion and track reliability is characterized in that an improved model in the step S2 comprises the following structural points that a convolution simulation self-attention unit is arranged in a backbone network stage, large-core depth separable convolution and dynamic convolution are arranged in parallel in a residual frame, relative position bias is introduced into local window multi-head attention, features consistent with an original level are output after point-by-point convolution aggregation, a multi-scale linear attention mechanism is arranged in a feature interaction stage, cross-scale information interaction and scale alignment are achieved in a coding layer through multi-scale key mapping and value mapping and combined with nuclear linear attention, so that feature codes consistent with semantics are obtained, and semantic injection is carried out in the mixed coding stage so as to supplement high-level semantic information to low-level detail features, and a mixed convolution unit is adopted in the down sampling process to complete feature fusion, mixing and alignment, and P3, P4 and P5 multi-scale features are output to a decoder.
4. The sperm quality evaluation and screening determination method based on parameter set fusion and track reliability as set forth in claim 1, wherein step S3 comprises two steps of pre-training and frame-by-frame detection, wherein the microscopic imaging video data obtained in step S1 is subjected to frame slicing to construct a training data set, supervised training is carried out to obtain detection model parameters, the optimal weight is loaded after training is completed to detect the video sequence frame by frame, and the first step in the video sequence is carried out Inputting the frame image into the detection model to obtain the frame detection set The position parameter Representing the center coordinates of the bounding box and its width and height, said confidence level Scoring the confidence that the target belongs to the sperm class, and aggregating the set Time index to the frame And establishing a one-to-one correspondence and storing.
5. The sperm quality evaluation and screening determination method based on parameter set fusion and trajectory reliability as claimed in claim 1, wherein step S4 of cross-frame correlation and trajectory generation of continuous multi-frame detection results comprises initializing initial frame detection results into temporary trajectories and assigning unique identifiers, and starting from sub-frame, performing frame-by-frame time interval on trajectories in an active state Performing Kalman prediction and mapping to an observation domain to obtain: ; Wherein, the In order to predict the observation, the system, In order to observe the matrix, As a priori the state vector, In order to create the covariance of the innovation, For the a priori covariance matrix, Measuring the noise covariance; To be used for Calculating the mahalanobis distance: ; Wherein, the In order to be a mahalanobis distance, Detecting and observing for the current frame; implementing statistical gating when The method comprises the steps of taking the matching track as a candidate pair, constructing a comprehensive cost function based on geometric overlapping degree and center offset on the candidate pair, carrying out one-to-one optimal allocation by adopting a Hungary algorithm, updating matched tracks, adding track points, initializing unmatched detection as a new track, accumulating loss of the unmatched tracks and exceeding a preset termination time length Terminating when the temporary track is continuously matched for a preset confirmation time period Then converting into a confirmation state and solidifying into a unique identification ID; The track reliability calculating and maintaining step includes calculating track reliability after each update The reliability is determined by weighting the matching stability, the detection stability and the prediction consistency when Suspending track expansion when the threshold value is lower than the preset time period Terminating, wherein the track with the reliability not lower than the threshold value continuously participates in subsequent physical conversion and quality score calculation; Wherein, the Is a square-clamping threshold value, and the square-clamping threshold value is set as a square-clamping threshold value, Is the reliability minimum threshold.
6. The sperm quality evaluation and screening determination method based on parameter set fusion and trajectory reliability according to claim 1, wherein the kinematic parameters in step S5 are used for quantifying one of a velocity, a swing and a rhythm, at least comprising one of a linear velocity and an average path velocity in a velocity class parameter, a lateral head displacement amplitude in a swing class parameter, one of a cross frequency and a direction retention in a rhythm class parameter, adaptively selecting a linear velocity or an average path velocity based on a trajectory linearity index and a direction retention in each time window, and adaptively selecting a cross frequency or a direction retention in combination with a sampling rate and a window duration condition; the target area is based on the first The detected boundary frame in the frame is subjected to expansion according to a preset proportion to generate a target area For covering sperm head and partial middle section area, and establishing corresponding relation with the frame time index and track mark Performing rotation normalization and scale normalization for subsequent geometric feature calculation, and performing gray scale normalization, illumination correction, background suppression and denoising treatment to eliminate the influence of microscopic imaging condition difference on the result; the geometric shape parameter is based on pairs of Head mask calculation for image segmentation including at least area Aspect ratio Principal direction angle Degree of solidity Wherein Is the convex hull area of the mask. The said The method is obtained by performing second moment analysis on the mask, and performing internal scale normalization on the size class parameters according to the head length.
7. The sperm quality assessment and screening determination method based on parameter set fusion and trajectory reliability as described in claim 1, wherein step S6 said parameter set uses trajectory identification ID and time window Forming feature vectors within the window for index keys The feature vector comprises a kinematic parameter linear velocity or average path velocity, lateral head displacement amplitude, crossover frequency or direction retention, and a geometric profile parameter area Aspect ratio Principal direction angle Degree of solidity And is associated with a time index and a track identification ID for subsequent fusion and scoring calculation; normalizing the size parameters according to the head length, normalizing the rest components to uniform scale, scoring the indexes unfavorable to undersize or oversize by adopting a band-pass function to weaken the influence of extreme values on the comprehensive score, respectively fusing the normalized kinematic quantity and geometric components into kinematic sub-scores And geometric sub-scoring And linearly weighting according to preset non-negative weights to obtain the basic score of the time window: ; the integrated quality score introduces trajectory reliability Modulating the basic score to obtain a comprehensive quality score of the time window: ; Wherein, the As a basis for the score of the score, As the weight of the material to be weighed, In order to achieve a degree of reliability of the track, In order to modulate the index of the light, For use as input for a subsequent time weighting and threshold decision.
8. The sperm quality evaluation and screening determination method based on parameter set fusion and trajectory reliability as described in claim 1, wherein the time weighted average in step S7 is calculated by the method of Is set for a predetermined time window of (2) In, divide the window into Adjacent time slices, record First, a window reliability weighted average is calculated: ; When (when) When calculating the window comprehensive quality weighted average: ; When (when) Outputting the result of screening and judging passing or not passing, and outputting the track mark and time window of the sperm target Is a result of the evaluation of (2); Wherein, the For the time-slice index to be used, For the duration of the time slice, For the track reliability of the time-slices, As a threshold value for the degree of reliability, Scoring the composite quality of the time slices; As the weight of the time in question, For a window composite quality weighted average, Is the composite quality threshold.
9. The system based on the sperm quality evaluation and screening determination method of the parameter set fusion and track reliability comprises a data access and calibration module, a model management and reasoning module, a cross-frame association management module, a target area processing module, a parameter calculation and summarization module, a scoring module, a time weighting evaluation and screening module and a screening result output module, wherein the data access and calibration module is used for accessing acquired microscopic imaging videos and frame time information, completing physical calibration of space and time scales and establishing a time index, the model management and reasoning module is used for constructing, training, loading and running a detection model, outputting target position information and confidence level and associating with the time index, the cross-frame association management module is used for generating a motion track and assigning a unique identifier in a cross-frame association manner, calculating and maintaining track reliability, the target area processing module is used for generating a target area based on a detection boundary frame and executing rotation normalization and imaging condition correction for subsequent feature extraction, the parameter calculation and summarization module is used for carrying out physical unit conversion, extracting motion and geometric parameters and summarization to form parameter sets, carrying out normalization processing, the scoring module is used for carrying out feature fusion on the parameter sets, generating comprehensive quality scores, the time weighting evaluation and the screening module is used for carrying out time weighting evaluation in a preset time window, outputting screening determination result according to the reliability and threshold value.

Description

Sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability Technical Field The invention relates to the technical field of intersection of medical image processing and computer vision, in particular to a sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability. Background Objective assessment of sperm quality is critical to the success of male infertility diagnosis and assisted reproductive technology. Traditional manual interpretation is greatly influenced by subjective experience, and has poor repeatability and limited consistency. Although the existing computer-aided semen analysis (CASA) can automatically analyze, the conditions of detection, unstable tracking, identity switching and the like caused by high density and shielding still exist under the complex microscopic condition, so that the reliability of statistics is further affected. The existing evaluation is based on static average of few speed indexes, and is difficult to effectively fuse with morphology, texture, gesture and other information, so that the evaluation dimension is single. The weighting and dual threshold decision on the trace reliability and time window level is not enough, and the result is susceptible to transient noise, and the stability is to be improved. Chinese patent CN120374622a discloses a sperm quality assessment method and system based on multi-modal fusion, which performs level assessment by multi-source indexes such as space-time alignment, trajectory linearity, thermodynamic diagram, abnormal time window and deformity rate, and is suitable for comprehensive interpretation of conventional microscopic videos. Despite the expansion in the index dimension, the cross-frame association link is not sufficiently practical and robust. Under the conditions of high density, shielding or short-time mismatch, identity switching and time sequence fluctuation are easy to occur. The lack of quantitative coupling between the quality of the trajectory and the score results in a sensitivity to short tracks, mis-matches and transient noise. The weighting of the time window scale and the double threshold decision are insufficient, and the output stability is limited. Chinese patent CN112036384B proposes a sperm head morphology recognition method, a device and equipment, wherein a one-stage detector represented by YOLOv is adopted to complete head region positioning, and morphology discrimination is carried out by combining traditional image processing, so that the method is suitable for static morphology evaluation, but the method focuses on single frame recognition, end-to-end detection and cross-frame correlation of microscopic videos are not formed, and the robustness processing of a small-target multi-scale scene is insufficient. The evaluation dimension of the scheme is mainly focused on the geometric shape of the head, and the lack of the joint modeling of the kinematic quantity related to the vitality is difficult to reflect the functional difference. The two-stage flow from frame to mask is sensitive to illumination, contrast and background of microscopic imaging, and threshold and morphological operations are easy to introduce error segmentation and index deviation. In summary, the prior art has insufficient correlation between robust detection and frame crossing reliability of small targets under complex microscopic conditions, insufficient quantitative coupling of multidimensional information and track reliability, and imperfect time window weighting and dual-threshold judgment, so that the evaluation is sensitive to transient noise and identity switching, and has insufficient stability and consistency. Therefore, it is necessary to propose an evaluation method and system that combines detection, correlation and multi-source fusion. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides a sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability, aiming at the problems of unstable detection and tracking, non-uniform units and scales, single evaluation and the like under microscopic imaging conditions. The invention aims at realizing the following technical scheme that the sperm quality evaluation and screening judgment method and system based on parameter set fusion and track reliability comprises the following steps: S1, acquiring microscopic imaging video data, completing physical calibration of a space scale and a time scale, and establishing a time index taking a starting frame as a reference; S2, constructing an improved detection model based on RT-DETR, setting a convolution simulation self-attention unit in a backbone network stage, adopting a multi-scale linear attention mechanism in a feature interaction stage, carrying out semantic injection in a hybrid coding stage, and adopting a hybrid convolut