CN-122023399-A - Automatic chromosome nuclear type image identification method and system based on artificial intelligence

CN122023399ACN 122023399 ACN122023399 ACN 122023399ACN-122023399-A

Abstract

The application provides a chromosome nuclear type image automatic identification method and system based on artificial intelligence, which relate to the technical field of image automatic identification and are characterized in that a plurality of section outlines with lines distributed at equal intervals in the length direction of a chromosome in a chromosome nuclear type image are extracted; the method comprises the steps of constructing a banded cross section profile sequence of a chromosome, inputting the banded cross section profile sequence into a pre-trained cyclic neural network, outputting the banded contrast at each cross section position by the cyclic neural network, generating a banded degradation curve of the chromosome according to the banded contrast at the adjacent cross section position, extracting an attenuation coefficient of the banded contrast of the banded degradation curve along a long axis of the chromosome, and judging that the banded distortion of the chromosome is the excessive enzymolysis artificial distortion when the attenuation coefficient is smaller than a preset threshold value, otherwise judging that the structure is abnormal. The application can realize the accurate identification of the excessive pattern distortion of the chromosome enzymolysis, thereby distinguishing the artificial distortion from the real abnormality of the chromosome.

Inventors

WANG HUI
WANG YIBO
LUO ZHU
LI XUE
LIU CHAN
ZHANG HUI
JIANG ZHIZHONG

Assignees

湖南信息学院

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The automatic identification method of the chromosome karyotype image based on the artificial intelligence is characterized by comprising the following steps: acquiring a chromosome karyotype image to be identified, and extracting a plurality of section profiles with patterns distributed at equal intervals in the length direction of a chromosome in the chromosome karyotype image; arranging all the textured cross-sectional profiles of the chromosome in the order from centromere to telomere, and constructing a textured cross-sectional profile sequence of the chromosome; Inputting the sequence of the sectional profiles with the stripes into a pre-trained cyclic neural network, outputting the stripe contrast at each section position by the cyclic neural network, and generating a stripe degradation curve of a chromosome according to the stripe contrast at the adjacent section position; Extracting attenuation coefficients of the band contrast of the band degradation curve along the long axis of the chromosome, judging that the band type distortion of the chromosome is the enzymolysis excessive artificial distortion when the attenuation coefficients are smaller than a preset threshold value, and judging that the structure is abnormal in reality otherwise.
2. The automatic identification method for a chromosome karyotype image based on artificial intelligence according to claim 1, wherein extracting a plurality of textured cross-sectional profiles distributed at equal intervals in the length direction of a chromosome in the chromosome karyotype image specifically comprises: chromosome example segmentation is carried out on the chromosome karyotype image to obtain a binary mask image of a single chromosome; Extracting a central line of a chromosome according to the binary mask image, and determining the length direction from a centromere to a telomere on the central line; And setting a plurality of sampling points on the central line at equal intervals, sampling pixel gray values at each sampling point along the direction perpendicular to the central line, and generating the textured cross-section profile at the sampling point.
3. The automatic identification method of chromosome karyotype image based on artificial intelligence according to claim 1, wherein the step of arranging all the textured cross-section profiles of the chromosome in order from centromere to telomere to construct the textured cross-section profile sequence of the chromosome specifically comprises: Identifying the centromere position of each chromosome in the chromosome karyotype image, and determining the short arm telomere direction and the long arm telomere direction of the chromosome; Ordering the textured section profiles at each sampling point along the short arm telomere direction and the long arm telomere direction respectively by taking the positions of the centromeres as starting points; All sequenced textured section profiles are spliced in sequence from the short arm telomeres to the long arm telomeres, and a textured section profile sequence of the chromosome is constructed.
4. The automatic identification method of chromosome nuclear type image based on artificial intelligence as claimed in claim 1, wherein inputting the sequence of textured cross-section profiles into a pre-trained recurrent neural network, the recurrent neural network outputting textured contrast at each cross-section position specifically comprises: Performing length normalization on each textured cross section contour in the textured cross section contour sequence, and converting each contour into a one-dimensional gray scale feature vector with fixed length to obtain a feature vector sequence corresponding to the cross section position one by one; Inputting the feature vector sequence into a pre-trained cyclic neural network in sequence according to the sequence of the section positions, wherein the cyclic neural network receives the feature vector of the current section position at each time step and updates the internal memory state of the feature vector; Extracting hidden state vectors output by the cyclic neural network at each time step, wherein all the hidden state vectors are arranged according to the section position sequence to form a hidden state drift field reflecting the evolution rule of the chromosome band patterns along the long axis; performing a contrast response activation operation on each hidden state vector in the hidden state drift field, mapping Gao Weiyin states to a textured contrast initial value in scalar form; And carrying out amplitude normalization correction on the initial value of the band contrast, and eliminating contrast baseline drift caused by the difference of dyeing conditions among different chromosomes to obtain the band contrast at each section position.
5. The automatic identification method of a chromosome karyotype image based on artificial intelligence according to claim 1, wherein the generating of the banding degradation curve of the chromosome according to the banding contrast at the adjacent section positions specifically comprises: traversing all adjacent section positions according to the increasing sequence of the section positions, and calculating the difference value of the contrast ratio of the stripe between each pair of adjacent sections to obtain the attenuation amplitude at each section position; All attenuation amplitudes are arranged according to the sequence of the section positions, and a contrast gradient field with patterns, which reflects the change speed of the contrast with patterns along the long axis of the chromosome, is constructed; Sequentially accumulating all attenuation amplitudes in the contrast gradient field with the stripes along the long axis direction of the chromosome from the initial section position to obtain an accumulated attenuation amplitude sequence of each section position relative to the initial section position; Generating an initial textured degradation curve from the sequence of cumulative decay amplitudes for each cross-sectional position relative to the starting cross-sectional position; And carrying out moving average smoothing treatment on the initial band degradation curve to eliminate burr fluctuation caused by image noise or local irregular band, so as to obtain a final band degradation curve of the chromosome.
6. The method for automatically identifying a nuclear type image of a chromosome based on artificial intelligence as set forth in claim 1, wherein extracting attenuation coefficients of the contrast of the band degradation curve along the long axis of the chromosome specifically includes: Executing attenuation curve segment interception operation on the textured degradation curve, intercepting an effective curve segment from a starting section position to an ending section position, and eliminating abnormal fluctuation areas generated by textured boundary effects at two ends of the curve to obtain an attenuation curve segment to be analyzed; and carrying out attenuation registration on the attenuation curve segments to obtain attenuation coefficients of the textured contrast along the long axis of the chromosome.
7. The automatic recognition method of chromosome nuclear type image based on artificial intelligence according to claim 4, wherein the feature vector sequence is sequentially input into a pre-trained recurrent neural network according to the sequence of section positions, and the recurrent neural network receives the feature vector of the current section position at each time step and updates the internal memory state thereof specifically comprises: Establishing a time step alignment mechanism between each feature vector and a section position in the feature vector sequence, and mapping each section position from a centromere to a telomere direction into a time step input node of a pre-trained cyclic neural network; At each current time step, the recurrent neural network reads the history memory state saved at the last time step, and takes the history memory state as the context background information of the current time step to participate in the processing of the current feature vector; the feature vector input in the current time step and the historical memory state of the previous time step are fused according to a preset gating weight through a gating fusion unit in the cyclic neural network, and a candidate memory state of the current time step is generated; And writing the candidate memory state into a memory unit of the cyclic neural network, finishing progressive refreshing from the memory state of the previous time step to the memory state of the current time step, and transmitting the refreshed memory state to the next time step.
8. The method of claim 4, wherein performing a contrast response activation operation on each hidden state vector in the hidden state drift field, mapping Gao Weiyin states into a scalar version of a striped contrast initial value comprises: according to the increasing sequence of the section positions, performing point-by-point traversing operation on each hidden state vector in the hidden state drift field, and sequentially taking out the hidden state vector corresponding to each section position as an input object of activating operation; performing full-connection dimension reduction mapping on each extracted hidden state vector, and performing linear aggregation on all dimension components of the hidden state vector according to trained weight coefficients to obtain an aggregation response value with a single dimension; Performing nonlinear activation compression on the aggregation response value, and limiting the dynamic range of the aggregation response value to a preset contrast ratio response interval through a nonlinear activation function to obtain a compressed contrast ratio activation value; And taking the banding contrast activating value as the initial banding contrast value at the section position, and completing mapping from Gao Weiyin state space to scalar contrast space.
9. The automatic identification method of chromosome nuclear type image based on artificial intelligence as claimed in claim 6, wherein performing attenuation registration on the attenuation curve segment to obtain attenuation coefficient with contrast along long axis of chromosome specifically comprises: Dividing the attenuation curve section into a plurality of continuous subintervals according to the section positions, wherein each subinterval covers a fixed number of continuous section positions; Estimating local attenuation rate of the attenuation curve segment in each subinterval, and determining the segment attenuation rate corresponding to the subinterval according to the longitudinal axis variation of the starting end and the terminating end of the segment; the sectional attenuation rates corresponding to all subintervals are arranged according to the sequence of the section positions, and a sectional attenuation rate sequence reflecting variation fluctuation of the attenuation rates along the long axis of the chromosome is constructed; And carrying out weighted fusion processing on the segmented attenuation rate sequence, distributing higher fusion weights for subintervals close to the centromere region, and taking the weighted fusion result as an attenuation coefficient of the band contrast along the long axis of the chromosome.
10. An artificial intelligence based automatic identification system for a chromosome nuclear type image for performing an artificial intelligence based automatic identification method for a chromosome nuclear type image according to any one of claims 1 to 9, wherein the automatic identification system for a chromosome nuclear type image comprises: The chromosome karyotype feature extraction module is used for acquiring a chromosome karyotype image to be identified, and extracting a plurality of textured cross-sectional profiles distributed at equal intervals in the length direction of a chromosome in the chromosome karyotype image; The chromosome band vein section profile identification module is used for arranging all band vein section profiles of the chromosome in the sequence from centromeres to telomeres to construct a band vein section profile sequence of the chromosome; the chromosome band degradation identification module is used for inputting the band cross section profile sequence into a pre-trained cyclic neural network, outputting band contrast at each cross section position by the cyclic neural network, and generating a band degradation curve of the chromosome according to the band contrast at the adjacent cross section position; The chromosome band type distortion judging module is used for extracting attenuation coefficients of band contrast of the band degradation curve along a long axis of a chromosome, judging that the band type distortion of the chromosome is enzymolysis excessive artificial distortion when the attenuation coefficients are smaller than a preset threshold value, and judging that the chromosome is true structural abnormality otherwise.

Description

Automatic chromosome nuclear type image identification method and system based on artificial intelligence Technical Field The application relates to the technical field of automatic identification of images, in particular to an automatic identification method and an automatic identification system of chromosome nuclear type images based on artificial intelligence. Background The automatic image recognition technology is used as a key technology of chromosome karyotype intelligent analysis, has important application value in clinical cytogenetic detection, can replace manual completion of chromosome streak feature extraction, morphological discrimination and anomaly recognition, effectively reduces subjective errors of manual interpretation, improves the efficiency and standardization degree of karyotype analysis, is widely applied to scenes such as genetic disease screening, prenatal diagnosis and tumor cytology detection, provides automatic technical support for accurate determination of chromosome structural anomalies, and promotes clinical chromosome detection to develop towards an intelligent and efficient direction. In the practical application of chromosome karyotype image recognition, the problem of artificial distortion such as fuzzy banding, boundary ablation, banding fusion and the like is easily caused by improper enzymolysis parameter control in a flaking link, the distortion form caused by excessive enzymolysis is highly similar to the visual characteristics of real structural anomalies such as natural bending of a chromosome, segment deletion and the like, the conventional image automatic recognition model is difficult to effectively distinguish the two, the artificial distortion is easily misjudged as pathological anomalies, the misjudgment can directly lead to the distortion of a karyotype analysis result, the risk of clinical diagnosis false positive is improved, and the reliability of diagnosis and treatment results is influenced, so that the accurate recognition of the excessive banding distortion of chromosome enzymolysis is realized, and the problem of distinguishing the artificial distortion from the real anomaly of the chromosome becomes an industry faced problem. Disclosure of Invention The application provides an automatic identification method and an automatic identification system for a chromosome nuclear type image based on artificial intelligence, which can realize accurate identification of excessive chromosome enzymolysis with streak distortion, so as to distinguish artificial distortion from real abnormality of a chromosome. In a first aspect, the present application provides an artificial intelligence based automatic identification method for a chromosome karyotype image, the automatic identification method for a chromosome karyotype image comprising the steps of: acquiring a chromosome karyotype image to be identified, and extracting a plurality of section profiles with patterns distributed at equal intervals in the length direction of a chromosome in the chromosome karyotype image; arranging all the textured cross-sectional profiles of the chromosome in the order from centromere to telomere, and constructing a textured cross-sectional profile sequence of the chromosome; Inputting the sequence of the sectional profiles with the stripes into a pre-trained cyclic neural network, outputting the stripe contrast at each section position by the cyclic neural network, and generating a stripe degradation curve of a chromosome according to the stripe contrast at the adjacent section position; Extracting attenuation coefficients of the band contrast of the band degradation curve along the long axis of the chromosome, judging that the band type distortion of the chromosome is the enzymolysis excessive artificial distortion when the attenuation coefficients are smaller than a preset threshold value, and judging that the structure is abnormal in reality otherwise. In this embodiment, extracting a plurality of textured cross-sectional profiles distributed at equal intervals in the length direction for a chromosome in the chromosome karyotype image specifically includes: chromosome example segmentation is carried out on the chromosome karyotype image to obtain a binary mask image of a single chromosome; Extracting a central line of a chromosome according to the binary mask image, and determining the length direction from a centromere to a telomere on the central line; And setting a plurality of sampling points on the central line at equal intervals, sampling pixel gray values at each sampling point along the direction perpendicular to the central line, and generating the textured cross-section profile at the sampling point. In this embodiment, the sequence of textured cross-sectional profiles for constructing the chromosome by arranging all textured cross-sectional profiles of the chromosome in order from centromere to telomere specifically includes: Identifying the centromere position of