WO2020168511A1 - 染色体异常检测模型、其检测系统及染色体异常检测方法 - Google Patents

染色体异常检测模型、其检测系统及染色体异常检测方法 Download PDF

Info

Publication number
WO2020168511A1
WO2020168511A1 PCT/CN2019/075693 CN2019075693W WO2020168511A1 WO 2020168511 A1 WO2020168511 A1 WO 2020168511A1 CN 2019075693 W CN2019075693 W CN 2019075693W WO 2020168511 A1 WO2020168511 A1 WO 2020168511A1
Authority
WO
WIPO (PCT)
Prior art keywords
chromosome
abnormality detection
target
image
chromosomes
Prior art date
Application number
PCT/CN2019/075693
Other languages
English (en)
French (fr)
Inventor
蔡辅仁
黄宗祺
廖英凯
游家鑫
谢柏欣
Original Assignee
中国医药大学附设医院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国医药大学附设医院 filed Critical 中国医药大学附设医院
Priority to PCT/CN2019/075693 priority Critical patent/WO2020168511A1/zh
Priority to JP2021505288A priority patent/JP2021531812A/ja
Publication of WO2020168511A1 publication Critical patent/WO2020168511A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding

Definitions

  • the present invention relates to a medical information analysis model, system and method, in particular to a chromosome abnormality detection model, chromosome abnormality detection system and chromosome abnormality detection method.
  • Chromosomal abnormalities are mostly used for genetic disease screening, or cancer cell mutation detection such as blood cancer and lymphoma.
  • genetic disease screening is mainly for pregnant women to receive relevant tests during pregnancy. Because the fetus carries both the father’s sperm cell and the mother’s egg cell chromosomes from meiosis, each time the embryonic life occurs. , Chromosomal mutations in the embryo may occur, and the health of the fetus needs to be confirmed by detecting whether the fetal chromosome is abnormal.
  • Chromosomal abnormalities can generally be divided into abnormal number of chromosomes, abnormal chromosome structure, and abnormal chromosomal patchwork.
  • the abnormal number of chromosomes means that when the germ cell undergoes meiosis, if a certain chromosome does not separate (nondisjunction), it will cause the abnormal number of chromosomes in sperm or egg cells. After conception, the number of chromosomes will become haploid or multiple. Embryos, and give birth to abnormal fetuses.
  • Common chromosome number abnormalities include trisomy 21 (Down's disease), trisomy 18 (Edward's disease), and single chromosome X (Turner's disease).
  • Abnormal chromosome structure is caused by one or more defects or abnormal combination of chromosome structure.
  • the more common chromosomal patchwork abnormalities are 46, XX/47, XX, +21 Down syndrome patchwork, 45, X/46, XX, 45, X/46, XY or 45, X/46, X, i(Xq) is a patchwork of Turner's disease.
  • it is a patchwork of cells with normal chromosomes, and the symptoms are usually milder than a single pure chromosomal abnormality.
  • the well-known chromosome abnormality detection method is to take the image of the metaphase of chromosome cell division and artificially arrange the chromosome karyotype by the examiner to determine whether the chromosome is haploid or polyploid to determine whether the number of chromosomes is abnormal, and Whether the chromosomes are missing, circular, inverted, or dislocated to determine whether there is an abnormal chromosome structure, the evaluation results of chromosome abnormality detection vary greatly among different examiners, and the process is cumbersome and time-consuming. Therefore, how to develop a chromosome abnormality detection system with high accuracy and rapid detection is indeed a technical subject with commercial value.
  • an object of the present invention is to provide a chromosomal abnormality detection model, a chromosomal abnormality detection method, and a chromosomal abnormality detection system, which can objectively and accurately determine whether a subject has a chromosomal abnormality and can use this to classify diseases And risk assessment.
  • One aspect of the present invention is to provide a chromosome abnormality detection model, which includes the following establishment steps: obtaining a reference database, performing an image conversion step, performing a preliminary classification step, performing a feature selection step, and performing a training step.
  • the reference database includes a plurality of reference chromosomal cell division metaphase images.
  • the image conversion step is to use an unsupervised learning method classifier to arrange 23 pairs of chromosomes in the metaphase image of the reference chromosome cell division to obtain multiple reference chromosome karyotype images.
  • the preliminary classification step is to classify according to the number of chromosomes in the reference chromosome karyotype image.
  • the feature selection step is to use the feature selection module to analyze the reference chromosome karyotype image to obtain at least one image feature value.
  • the training step is to train the aforementioned at least one image feature value through a convolutional neural network learning classifier to achieve convergence, so as to obtain the chromosome abnormality detection model, wherein the chromosome abnormality detection model is used to determine the subject Whether the subject has abnormal chromosomal structure or chromosomal patchwork abnormality.
  • the unsupervised learning method classifier can be a Generative Adversarial Network (GAN).
  • GAN Generative Adversarial Network
  • At least one image feature value may include chromosome size, chromosome position or chromosome shape.
  • the convolutional neural network learning classifier can be Inception-ResNet-v2 convolutional neural network or Inception-V3 convolutional neural network.
  • Another aspect of the present invention is to provide a chromosome abnormality detection method, which comprises the following steps.
  • the unsupervised learning method classifier is used to arrange 23 pairs of chromosomes in the metaphase image of the target chromosome cell division to obtain the target chromosome karyotype image.
  • the aforementioned chromosome abnormality detection model is used to analyze the aforementioned target chromosome karyotype image to determine whether the subject has a chromosome abnormality.
  • the chromosomal abnormality can include abnormal chromosome number, abnormal chromosome structure, or abnormal chromosomal patchwork.
  • the abnormal number of chromosomes may include that the subject's target chromosome is haploid or polyploid
  • the chromosome structure abnormality may include that the subject's target chromosome is chromosome deletion, circular chromosome, chromosome translocation, chromosome inversion, or chromosome repeat.
  • Another aspect of the present invention is to provide a chromosome abnormality detection system, including an image capture unit and a non-transitory machine-readable medium.
  • the image capturing unit is used to obtain the metaphase image of the target chromosomal cell division of the subject.
  • the non-transitory machine-readable medium is signally connected to the image capturing unit.
  • the non-transitory machine-readable medium is used to store a program. When the aforementioned program is executed by the processing unit, it is used to determine whether the subject has a chromosomal abnormality.
  • the program includes a reference database acquisition module, a reference image conversion module, a reference preliminary classification module, a reference feature selection module, a training module, a target image conversion module, a target preliminary classification module, a target feature selection module, and a comparison module.
  • the reference database obtaining module is used to obtain a reference database, and the aforementioned reference database is established by a plurality of reference chromosomal cell division metaphase images.
  • the reference image conversion module uses an unsupervised learning method classifier to arrange 23 pairs of chromosomes in the metaphase image of the reference chromosome cell division to obtain multiple reference chromosome karyotype images. Refer to the preliminary classification module to classify the reference chromosome karyotype image based on the number of reference chromosomes.
  • the reference feature selection module is used to analyze the reference chromosome karyotype image to obtain at least one reference image feature value.
  • the training module is used to train at least one reference image feature value through the convolutional neural network learning classifier to achieve convergence, so as to obtain a chromosome abnormality detection model.
  • the target image conversion module uses an unsupervised learning method classifier to arrange the 23 pairs of chromosomes in the metaphase image of the target chromosome cell division to obtain the target chromosome karyotype image.
  • the target preliminary classification module is used to classify the target chromosome karyotype image according to the number of target chromosomes. If the number of target chromosomes is 46, it is classified as normal; if the number of target chromosomes is greater than or less than 46, it is classified as The number of chromosomes is abnormal.
  • the target feature selection module is used to analyze the target chromosome karyotype image to obtain at least one target image feature value.
  • the comparison module is used to analyze the target image feature value using the chromosome abnormality detection model to obtain target image feature value weight data, and determine whether the subject has chromosome structure abnormality or chromosomal patchwork abnormality according to the target image feature value weight data .
  • the unsupervised learning method classifier can be a Generative Adversarial Network (GAN).
  • GAN Generative Adversarial Network
  • At least one reference image feature value can include chromosome size, chromosome position or chromosome shape
  • at least one target image feature value can include chromosome size, chromosome position or chromosome shape.
  • the convolutional neural network learning classifier can be an Inception-ResNet-v2 convolutional neural network or an Inception-V3 convolutional neural network.
  • the non-transitory machine-readable medium may further include an evaluation module for calculating the risk value of the subject having a chromosome abnormality based on the target image feature value weight data.
  • the chromosome abnormality detection model, chromosome abnormality detection system and chromosome abnormality detection method of the present invention automatically convert the target chromosome metaphase image into the target chromosome karyotype image, and use the target feature selection module to analyze the target chromosome karyotype image Then, the method of obtaining at least one target image feature value can effectively reduce the error caused by the subjective consciousness of different judges in the detection of chromosome abnormality.
  • the chromosome abnormality detection model with deep neural network learning function can not only effectively improve the accuracy and sensitivity of chromosome abnormality detection, but also greatly shorten the determination time of chromosome abnormality, so that the chromosome abnormality detection model and chromosome of the present invention
  • the abnormality detection system and the chromosome abnormality detection method are more efficient in chromosome abnormality detection.
  • Fig. 1 shows a flow chart of establishing a chromosome abnormality detection model according to an embodiment of the present invention
  • Fig. 2 shows a flowchart of a method for detecting chromosome abnormalities according to another embodiment of the present invention
  • Fig. 3 shows a block diagram of a chromosome abnormality detection system according to still another embodiment of the present invention.
  • Figure 4 shows the result of the conversion of the target chromosome cell metaphase image into the target chromosome karyotype image
  • FIG. 5 is a schematic structural diagram of a convolutional neural network learning classifier of a chromosome abnormality detection model according to an embodiment of the present invention
  • FIG. 6 is a schematic diagram of the architecture of a convolutional neural network learning classifier of a chromosome abnormality detection model according to another embodiment of an embodiment of the present invention.
  • Fig. 7 is a confusion matrix used by the chromosome abnormality detection model of the present invention to judge the chromosome abnormality of the subject.
  • FIG. 1 shows a flowchart of steps 100 for establishing a chromosome abnormality detection model according to an embodiment of the present invention.
  • the establishment step 100 of the chromosome abnormality detection model of the present invention includes step 110, step 120, step 130, step 140 and step 150.
  • the established chromosome abnormality detection model can be used to determine whether the subject has abnormal chromosome number, abnormal chromosome structure or Patchwork abnormalities.
  • Step 110 is to obtain a reference database, which includes a plurality of reference chromosomal metaphase images.
  • their chromatin is mostly distributed in the nucleus in the range of 30nm to 300nm.
  • the chromosomes will gradually become tightly packed.
  • the nuclear membrane of the cell disappears completely and the spindle filaments begin to become clear.
  • the centromere on each chromosome is attached to the spindle filament (or star ray). The centromere starts to move up and down under the pulling force of its two poles. Finally, the pulling force of the two poles reaches equilibrium.
  • the centromeres are arranged on the equatorial plate in the center of the cell.
  • the sharpness of the chromosome reaches its highest point. Therefore, before obtaining the metaphase image of the reference chromosome, the cells of the reference subject are injected into the metaphase by administering hormones, and then specific cells of the reference subject are extracted, and the reference chromosome cells are obtained through staining and microscopic observation Mid-split image.
  • Step 120 is an image conversion step, which uses an unsupervised learning method classifier to arrange 23 pairs of chromosomes in the metaphase image of the reference chromosome cell division to obtain multiple reference chromosome karyotype images.
  • the reference chromosome karyotype image is to use the aforementioned reference chromosome metaphase image to analyze, compare, sort and number the chromosomes according to the length of the chromosome, the position of the centromere, the ratio of long and short arms, and the presence or absence of satellites.
  • the resulting image may be a Generative Adversarial Network (GAN).
  • GAN Generative Adversarial Network
  • Step 130 is a preliminary classification step, which is based on the number of chromosomes in the reference chromosome karyotype image. If the number of chromosomes is 46, the classification is normal; if the number of chromosomes is greater than or less than 46, then the classification is performed The number of chromosomes is abnormal.
  • Step 140 is a feature selection step, which is to obtain at least one image feature value after analyzing the reference chromosome karyotype image with the feature selection module.
  • At least one image feature value may include chromosome size, chromosome position or chromosome shape.
  • Step 150 is a training step, which is to train the aforementioned at least one image feature value through a convolutional neural network learning classifier to achieve convergence, so as to obtain the chromosome abnormality detection model.
  • the convolutional neural network learning classifier can be Inception-ResNet-v2 convolutional neural network or Inception V3 convolutional neural network.
  • FIG. 2 shows a flowchart of a chromosome abnormality detection method 200 according to another embodiment of the present invention.
  • the chromosome abnormality detection method 200 of the present invention includes step 210, step 220, step 230, and step 240.
  • Step 210 is to provide a chromosome abnormality detection model, and the chromosome abnormality detection model is established through the aforementioned steps 110 to 140.
  • Step 220 is to provide the subject's target chromosome cell metaphase image.
  • the subject's cells are injected into the metaphase by administering hormones, and then the subject's specific cells are extracted. , And obtain the metaphase image of target chromosome cell division through staining and microscope observation.
  • Step 230 uses an unsupervised learning method classifier to arrange 23 pairs of chromosomes in the metaphase image of the target chromosome cell division to obtain the target chromosome karyotype image.
  • the target chromosome karyotype image is to image the aforementioned target chromosome cell division metaphase image, according to the length of the chromosome, the centromere position, the ratio of long and short arms, the presence or absence of satellites, etc., to analyze, compare, sort and number the chromosomes The image obtained afterwards.
  • the unsupervised learning method classifier may be a Generative Adversarial Network (GAN).
  • GAN Generative Adversarial Network
  • Step 240 is to analyze the target chromosome karyotype image using a chromosome abnormality detection model to determine whether the subject has a chromosome abnormality.
  • chromosomal abnormalities can include abnormal number of chromosomes, abnormal chromosome structure, or abnormal chromosomal patchwork.
  • the abnormal number of chromosomes may include that the subject's target chromosome is haploid or polyploid
  • the chromosome structure abnormality may include that the subject's target chromosome is chromosome deletion, circular chromosome, chromosome translocation, chromosome inversion, or chromosome repeat.
  • the chromosome abnormality detection model and chromosome abnormality detection method of the present invention automatically convert the target chromosome metaphase image into the target chromosome karyotype image, and use the feature selection module to analyze the target chromosome karyotype image to obtain at least one image
  • the feature value method can effectively reduce the error caused by the subjective consciousness of different judges in the detection of chromosome abnormalities.
  • the chromosome abnormality detection model with deep neural network learning function can not only effectively improve the accuracy and sensitivity of chromosome abnormality detection, but also greatly shorten the determination time of chromosome abnormality, so that the chromosome abnormality detection model and chromosome abnormality detection model of the present invention
  • the abnormality detection method is more efficient in chromosome abnormality detection.
  • FIG. 3 shows a block diagram of a chromosome abnormality detection system 300 according to another embodiment of the present invention.
  • FIG. 4 shows a target chromosome metaphase image 610 converted into a target chromosome karyotype image 620 result graph.
  • the chromosome abnormality detection system 300 of the present invention includes an image capturing unit 400 and a non-transitory machine-readable medium 500.
  • the chromosome abnormality detection system 300 can be used to determine whether the subject has abnormal chromosome number, abnormal chromosome structure, or abnormal chromosomal patchwork.
  • the image capturing unit 400 is used to obtain the metaphase image 610 of the target chromosomal cell division of the subject.
  • the image capturing unit can be an image capturing device with a microscope for capturing chromosome images observed by the microscope.
  • the non-transitory machine-readable medium 500 is signally connected to the image capturing unit 400.
  • the non-transitory machine-readable medium is used to store a program.
  • the aforementioned program is executed by the processing unit, it is used to evaluate whether the subject has a chromosomal abnormality.
  • the aforementioned program includes a reference database acquisition module 510, a reference image conversion module 520, a reference preliminary classification module 530, a reference feature selection module 540, a training module 550, a target image conversion module 560, a target preliminary classification module 570, and a target feature selection module 580 And the comparison module 590.
  • the reference database obtaining module 510 is used to obtain a reference database, and the aforementioned reference database is established by a plurality of reference chromosomal metaphase images.
  • the reference image conversion module 520 uses an unsupervised learning method classifier to arrange the 23 pairs of chromosomes in the metaphase image of the reference chromosome cell division to obtain multiple reference chromosome karyotype images.
  • the unsupervised learning method classifier may be a generative adversarial neural network.
  • the reference preliminary classification module 530 is used to classify the reference chromosome karyotype image according to the number of reference chromosomes. If the number of reference chromosomes is 46, it is classified as normal; if the number of reference chromosomes is greater than or less than 46, it is classified as abnormal. Preferably, the abnormal number of chromosomes may include that the subject's target chromosome is haploid or polyploid.
  • the reference feature selection module 540 is used to analyze the reference chromosome karyotype image to obtain at least one reference image feature value.
  • the at least one reference image feature value may include chromosome size, chromosome position or chromosome shape.
  • the training module 550 is used to train at least one reference image feature value through the convolutional neural network learning classifier to achieve convergence, so as to obtain a chromosome abnormality detection model.
  • the convolutional neural network learning classifier may be an Inception-ResNet-v2 convolutional neural network or an Inception V3 convolutional neural network.
  • the target image conversion module 560 uses an unsupervised learning method classifier to arrange 23 pairs of chromosomes in the target chromosome metaphase image 610 to obtain the target chromosome karyotype image 620.
  • the unsupervised learning method classifier may be a generative adversarial neural network.
  • the target preliminary classification module 570 is used to classify the target chromosome karyotype image according to the number of target chromosomes. If the number of target chromosomes is 46, it is classified as normal; if the number of target chromosomes is greater than or less than 46, it is classified as abnormal. Preferably, the abnormal number of chromosomes may include that the subject's target chromosome is haploid or polyploid.
  • the target feature selection module 580 is used to analyze the target chromosome karyotype image to obtain at least one target image feature value.
  • the at least one target image feature value may include chromosome size, chromosome position or chromosome shape.
  • the comparison module 590 is used to analyze the target image feature value using the chromosome abnormality detection model to obtain the target image feature value weight data, and determine whether the subject has a chromosome structure abnormality or a chromosomal patchwork type according to the target image feature value weight data abnormal.
  • the abnormal chromosome structure may include that the subject's target chromosome is a chromosome deletion, a circular chromosome, a chromosome translocation, a chromosome inversion or a chromosome duplication.
  • non-transitory machine-readable medium 500 may further include an evaluation module (not shown in the figure) for further calculating the risk value of the subject having a chromosomal abnormality based on the target image feature value weight data.
  • the reference database used in the present invention is the retrospectively delinked clinical content of the subjects collected by the China Medical University Hospital (CMUH), which is a clinical content approved by the Research Ethics Committee of China Medical University and affiliated Hospital
  • the test plan its number is: CMUH107-REC3-151.
  • the aforementioned reference database contains 30,000 reference chromosome metaphase images of the subject, and there is no particular limitation on the gender of the subject to which the aforementioned reference chromosome metaphase images belong, and there is no special interval for age.
  • each reference chromosome metaphase image will use the reference image conversion module to classify each reference chromosome metaphase image with an unsupervised learning method classifier
  • 23 pairs of chromosomes are arranged to obtain multiple reference chromosome karyotype images.
  • the current deep neural network model requires a large amount of training data (Training Data, that is, the metaphase image of each reference chromosome cell division of the chromosome abnormality detection model of the present invention) to achieve stable convergence and high classification accuracy. If the number of training data is not sufficient, it will cause the deep neural network to produce overfitting and cause the error value of the judgment result to be too high, resulting in low credibility of the deep neural network model.
  • the chromosome abnormality detection model of the present invention further includes an image pre-processing step, which performs black and white contrast correction on each reference chromosome karyotype image, and normalizes the image value so that the image value is between 0 and 1.
  • a preliminary classification step is performed to determine whether the subject has an abnormal number of chromosomes, which is classified based on the number of chromosomes in each reference chromosome karyotype image. If the number of chromosomes is 46, it is classified as normal; if the number of chromosomes is greater than or less than 46, it is classified as abnormal.
  • each reference chromosome karyotype image will be analyzed by the feature selection module to obtain at least one image feature value.
  • the feature selection module can further distinguish the image feature values of the chromosome size, chromosome position, or chromosome shape in each reference chromosome karyotype image.
  • the aforementioned image feature values will be trained by the convolutional neural network learning classifier to achieve convergence, so as to obtain the chromosome abnormality detection model of the present invention.
  • the chromosome abnormality detection model will be used to determine whether the subject has abnormal chromosome number, abnormal chromosome structure, or abnormal chromosomal patchwork.
  • the convolutional neural network learning classifier can be Inception-ResNet-v2 convolutional neural network or Inception V3 convolutional neural network.
  • FIG. 5 is a schematic diagram illustrating the architecture of the convolutional neural network learning classifier 700 of the chromosome abnormality detection model of the present invention.
  • the convolutional neural network learning classifier 700 is an Inception-ResNet-v2 convolutional neural network, which includes multiple convolutional layers (Convolution), multiple maximum pooling layers (MaxPool), Multiple average pooling layers (AvgPool) and multiple cascading layers (Concat) to train and analyze image feature values.
  • Convolution convolution
  • MaxPool multiple maximum pooling layers
  • AvgPool Multiple average pooling layers
  • Concat cascading layers
  • the Inception-ResNet-v2 convolutional neural network is a large-scale visual recognition convolutional neural network based on the ImageNet visualization data database, and the image data in the ImageNet visualization data database are all two-dimensional color images, so it is well known
  • the GoogLeNet convolutional neural network model has RGB three-channel filters in its first convolutional layer.
  • the original image files of each reference chromosome karyotype image are all three-dimensional grayscale images.
  • the chromosome abnormality detection model of the present invention further passes the GoogLeNet convolutional neural network model including RGB three-channel filters through the arithmetic average method.
  • the number of training sessions can be 100 (Epochs) and the gradient descent method using 96Mini-Batch Size, and adjust the initial learning rate (Learning Rates), where the learning rate is the control weight and bias when the neural network is trained
  • Learning Rates the initial learning rate
  • the chromosome abnormality detection model of the present invention can further ensure that the loss function (Loss Function) can reach stable convergence by adjusting the value of the learning rate.
  • the image feature value of each reference chromosome karyotype image is subjected to a two-layer convolution layer and a maximum pooling layer (MaxPool) processing to extract Perform the maximum output of the image feature value of the image, and repeat the aforementioned two-layer convolutional layer and one-layer maximum pooling layer output again, and use multiple convolutional layers for parallel towers training to complete the primary image feature value Training (Inception).
  • a two-layer convolution layer and a maximum pooling layer MaxPool
  • the image feature values of each reference chromosome karyotype image will be performed 10 times (10 ⁇ ), 20 times (20 ⁇ ) and 10 times (10 ⁇ ) of different depths, levels and aspects.
  • the Residual module is trained to train the image feature values of each reference chromosome karyotype image and achieve convergence.
  • each residual module performs different calculations and judgments on the image feature values of each standardized medial foot X-ray image data , Causing errors to accumulate.
  • the training of the Inception-ResNet convolutional neural network will pull the node operation value of a specific level back to the input of that level to perform the operation again, so as to prevent the convolutional neural network learning classifier 700 from comparing the foregoing
  • the image feature value of the image feature value undergoes multi-layer weight calculation training, and the degradation phenomenon of gradient disappearance occurs, and the error accumulation can avoid the loss of information, and can effectively improve the training efficiency of the convolutional neural network learning classifier 700.
  • a convolutional layer After completing the deep and repetitive residual module training, a convolutional layer, an average pooling layer, a global average pooling layer (Global Average Pooling 2D, GloAvePool2D), and a linear rectification unit training layer (Rectified Linear Unit, ReLU) perform final training and processing on the convergent image feature values to determine the subject's chromosome abnormality.
  • the average pooling layer can first calculate the image feature values after the residual module training is completed to find the average value of each image feature value.
  • the convolutional neural network can learn the classifier 700
  • the overall network architecture performs regularization processing to prevent the convolutional neural network learning classifier 700 from overfitting in the training mode pursuing low error, which leads to excessively high error values in the judgment results.
  • linear rectification The unit training layer further activates the image feature values after the training is completed, and outputs target image feature value weight data 701 for subsequent comparison and analysis.
  • the aforementioned linear rectification unit training layer can prevent the target image feature value weight data 701 output by the foot deformity detection model from approaching zero or approaching infinity, so as to facilitate the subsequent comparison steps, thereby improving the chromosome abnormality detection of the present invention The judgment accuracy of the model.
  • the judgment result of the aforementioned subject's chromosome abnormality condition will be further integrated into the reference database to optimize the chromosome abnormality detection model of the present invention, thereby further improving the training effect and judgment accuracy of the chromosome abnormality detection model of the present invention.
  • the convolutional neural network learning classifier 800 is an Inception V3 convolutional neural network, which includes multiple convolutional layers (Convolution), multiple average pooling layers (AvgPool), and multiple maximum Pooling layer (MaxPool) and multiple cascading layers (Concat), and use the dropout layer (Dropout), fully connected layer (Fully connected) and normalization layer (Softmax) to solve the problem of overfitting in machine learning to Train and analyze image feature values.
  • Convolution convolution
  • AvgPool multiple average pooling layers
  • MaxPool multiple maximum Pooling layer
  • Concat cascading layers
  • Dropout dropout
  • Fully connected layer Fully connected layer
  • Softmax normalization layer
  • Inception V3 Convolutional Neural Network is a factorization based on large filter size decomposition of convolutional network.
  • Parallel parameter reduction can not only solve the problem of overfitting, but also increase the parameters by increasing the depth of the network. The number is closer to the mathematical model originally intended to be approximated.
  • the image feature values of each reference chromosome karyotype image are respectively subjected to an average pooling layer and a convolution layer; five convolution layers; three Convolutional layer: After a convolutional layer is operated, the characteristic matrix values of each group of operations are stacked in cascade. After that, repeat 2 times to perform an average pooling layer and a convolutional layer; five convolutional layers; three convolutional layers; one convolutional layer operation, and the characteristic matrix values of each group of operations are Cascade layer stacking. Then perform a maximum pooling layer; three convolutional layers; and a convolutional layer.
  • the characteristic matrix values of each group of operations are stacked in cascade layers. After repeating 4 times, perform an average pooling layer and a convolutional layer; five convolutional layers; three convolutional layers and a convolutional layer. Cascade layer stacking. Then perform an average pooling layer, two convolutional layers, a fully connected layer, and a normalization layer. The calculated feature matrix values are repeated twice for an average pooling layer and a convolution layer. Layers; three layers of convolutional layers and one layer of cascade; two layers of convolutional layer and one layer of cascade; after one layer of convolutional layer operations, the feature matrix values of each set of operations are stacked in cascade layers. Finally, after performing an average pooling layer, a discarding layer, a fully connected layer, and a normalization layer, the target image feature value weight data 801 is output to obtain a trained chromosome anomaly detection model.
  • the judgment result of the aforementioned subject's chromosome abnormality condition will be further integrated into the reference database to optimize the chromosome abnormality detection model of the present invention, thereby further improving the training effect and judgment accuracy of the chromosome abnormality detection model of the present invention.
  • FIG. 7 is a confusion matrix used by the chromosome abnormality detection model of the present invention to judge the chromosome abnormality of the subject.
  • the convolutional neural network learning classifier for establishing the chromosome abnormality detection model is the convolutional neural network learning classifier 800 shown in FIG. 6 to determine whether the subject's chromosome is abnormal, and the result Divided into normal and abnormal.
  • the horizontal axis is the predicted label, and the vertical axis is the actual label.
  • the confusion matrix can be divided into True Positive (TP), True Negative (TN), False Positive (FP) and False Negative (False).
  • Negative, FN four parts, and calculate the accuracy, sensitivity, specificity, positive predictive value and negative predictive value of the chromosome abnormality detection model of the present invention based on the data of TP, TN, FP and FN.
  • the calculation method of correct rate is (TP+TN)/(TP+FP+TN+FN)
  • the calculation method of sensitivity is TP/(TP+FN)
  • the calculation method of specificity is TN/(TN+FP)
  • the positive predictive value is calculated as TP/(TP+FP)
  • the negative predictive value is calculated as TN/(FN+TN).
  • the number of subjects in the TP block is 206
  • the number of subjects in the TN block is 201
  • the number of subjects in the FP block is 3
  • the number of subjects in the FN block The number is 0 people.
  • the prediction result of the chromosome abnormality detection model of the present invention for judging the subject's chromosome abnormality is shown in Table 1.
  • the chromosomal abnormality detection model of the present invention can be used to accurately determine whether a subject has a chromosomal abnormality, and the chromosomal abnormality can include abnormal chromosome number, abnormal chromosome structure, and abnormal chromosomal patchwork.
  • the chromosome abnormality detection system of the present invention can effectively improve the accuracy and sensitivity of chromosome abnormality detection, and can shorten the evaluation time of whether a subject has chromosome abnormality. From the original image input to the interpretation result, it only needs 0.1- on average. It can be completed in 1 second, making it more widely used.

Landscapes

  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Public Health (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioethics (AREA)
  • Biotechnology (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Theoretical Computer Science (AREA)
  • Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
  • Apparatus Associated With Microorganisms And Enzymes (AREA)
  • Image Analysis (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

本发明提供一种染色体异常检测模型，其建立步骤包含：取得参照数据库，利用非监督式学习法分类器进行影像转换，得到多个参照染色体核型影像，依据多个参照染色体核型影像中的染色体条数进行初步分类，分析多个参照染色体核型影像得到至少一个影像特征值，并通过卷积神经网路学习分类器进行训练，得到所述染色体异常检测模型。本发明还提供一种染色体异常检测方法和检测系统。

Description

染色体异常检测模型、其检测系统及染色体异常检测方法 技术领域
本发明是有关于一种医疗信息分析模型、系统以及方法，特别是一种染色体异常检测模型、染色体异常检测系统以及染色体异常检测方法。
背景技术
染色体异常检查大多用于遗传疾病筛检，或血癌和淋巴癌等癌细胞变异侦测。其中遗传疾病筛检主要为孕妇于怀孕过程中皆会接受相关的检测，因胎儿同时携带父亲的精细胞与母亲的卵细胞经过细胞减数分裂而来的染色体，因此每次胚胎生命的发生过程中，有可能产生胚胎的染色体突变，需通过检测胎儿染色体异常与否确认胎儿的健康状态。
染色体异常一般可分为染色体数目异常、染色体结构异常以及染色体拼凑型异常。其中染色体数目异常为生殖细胞行减数分裂时，若发生某个染色体不分离(nondisjunction)现象时，便会导致精子或卵细胞染色体数目的异常，受孕之后就成为染色体数目为单倍体或多倍体的胚胎，而生出畸型的胎儿。常见的染色体数目异常包含三染色体21症(唐氏症)、三染色体18症(艾德华氏症)及单染色体X症(特娜氏症)等。染色体结构异常为染色体构造有一处或多处以上的缺损、异常组合等情况所造成。而较常见的染色体拼凑型异常有46,XX/47,XX,+21的唐氏症拼凑体、45,X/46,XX、45,X/46,XY或45,X/46,X,i(Xq)为透纳氏症的拼凑体。一般来说为含有部分正常染色体细胞的拼凑体，其症状通常要比单一纯粹的染色体异常为轻。
公知的染色体异常检测方式为拍摄染色体细胞分裂中期影像后，由检验人员进行人工排列为染色体核型图，再以此判断染色体是否出现单倍体或多倍体以判断是否出现染色体数目异常，以及染色体是否具有脱失、环状、倒位或错位的状况以判断是否出现染色体结构异常，是以染色体异常检测的评 估结果在不同检验人员间存在极大的差异，且过程也较为繁琐耗时。因此，如何发展出一种具有高度准确率及快速检测的染色体异常检测系统，实为一具有商业价值的技术课题。
发明内容
有鉴于此，本发明的一目的为提供染色体异常检测模型、染色体异常检测方法以及染色体异常检测系统，其可客观且准确的判断受试者是否存在染色体异常的状况，并可借此进行疾病分类和风险评估。
本发明的一方面在于提供一种染色体异常检测模型，包含以下建立步骤：取得参照数据库、进行影像转换步骤、进行初步分类步骤、进行特征选取步骤以及进行训练步骤。所述参照数据库包含多个参照染色体细胞分裂中期影像。所述影像转换步骤是利用非监督式学习法分类器将参照染色体细胞分裂中期影像中23对染色体进行排列，以得到多个参照染色体核型影像。所述初步分类步骤是依据参照染色体核型影像中的染色体条数进行分类，若染色体条数为46条，分类为染色体数目正常；若该色体条数为大于或小于46条，则分类为染色体数目异常。所述特征选取步骤是利用特征选取模块分析参照染色体核型影像后以得到至少一个影像特征值。所述训练步骤是将前述的至少一个影像特征值透过卷积神经网路学习分类器进行训练而达到收敛，以得到所述染色体异常检测模型，其中所述染色体异常检测模型是用以判断受试者是否具有染色体结构异常或染色体拼凑型异常。
依据前述的染色体异常检测模型，其中非监督式学习法分类器可为生成对抗神经网络(Generative Adversarial Network，GAN)。
依据前述的染色体异常检测模型，其中至少一个影像特征值可包含染色体大小、染色体位置或染色体形状。
依据前述的染色体异常检测模型，其中卷积神经网路学习分类器可为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
本发明的另一方面在于提供一种染色体异常检测方法，其包含下述步骤。提供如前段所述的染色体异常检测模型。提供受试者的目标染色体细胞分裂中期影像。利用所述非监督式学习法分类器将所述目标染色体细胞分裂中期影像中23对染色体进行排列，以得到目标染色体核型影像。利用前述的染色体异常检测模型分析前述的目标染色体核型影像，以判断受试者是否具有染色体异常。
依据前述的染色体异常检测方法，其中染色体异常可包含染色体数目异常、染色体结构异常或染色体拼凑型异常。优选地，染色体数目异常可包含受试者的目标染色体为单倍体或多倍体，染色体结构异常可包含受试者的目标染色体为染色体缺失、环状染色体、染色体转位、染色体倒转或染色体重复。
本发明的又一方面在于提供一种染色体异常检测系统，包含影像撷取单元以及非暂时性机器可读媒体。影像撷取单元用以取得受试者的目标染色体细胞分裂中期影像。非暂时性机器可读媒体信号连接影像撷取单元，其中非暂时性机器可读媒体用以储存程序，当前述的程序由处理单元执行时是用以判断受试者是否具有染色体异常，且前述的程序包含参照数据库取得模块、参照影像转换模块、参照初步分类模块、参照特征选取模块、训练模块、目标影像转换模块、目标初步分类模块、目标特征选取模块及比对模块。参照数据库取得模块用以取得参照数据库，且前述的参照数据库是由多个参照染色体细胞分裂中期影像所建立。参照影像转换模块，其是利用非监督式学习法分类器将参照染色体细胞分裂中期影像中23对染色体进行排列，以取得多个参照染色体核型影像。参照初步分类模块，用以将参照染色体核型影像依据参照染色体条数进行分类，若参照染色体条数为46条，分类为染色体数目正常，若参照染色体条数为大于或小于46条，则分类为染色体数目异常。参照特征选取模块用以分析参照染色体核型影像后以得到至少一个参照影像特征值。训练模块用以将至少一个参照影像特征值通过卷积神经网路学习分类 器训练达到收敛，以得到染色体异常检测模型。目标影像转换模块其是利用非监督式学习法分类器将目标染色体细胞分裂中期影像中23对染色体进行排列，以得到目标染色体核型影像。目标初步分类模块用以将目标染色体核型影像依据目标染色体条数进行分类，若目标染色体条数为46条，分类为染色体数目正常；若目标染色体条数为大于或小于46条，则分类为染色体数目异常。目标特征选取模块用以分析目标染色体核型影像后以得至少一个目标影像特征值。比对模块用以将目标影像特征值以所述染色体异常检测模型进行分析以得到目标影像特征值权重数据，并依据目标影像特征值权重数据判断受试者是否具有染色体结构异常或染色体拼凑型异常。
依据前述的染色体异常检测系统，其中非监督式学习法分类器可为生成对抗神经网络(Generative Adversarial Network,GAN)。
依据前述的染色体异常检测系统，其中至少一个参照影像特征值可包含染色体大小、染色体位置或染色体形状，至少一个目标影像特征值可包含染色体大小、染色体位置或染色体形状。
依据前述的染色体异常检测系统，其中卷积神经网路学习分类器可为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
依据前述的染色体异常检测系统，其中非暂时性机器可读媒体可还包含评估模块，用以依据目标影像特征值权重数据计算受试者具有染色体异常的风险值。
借此，本发明的染色体异常检测模型、染色体异常检测系统以及染色体异常检测方法通过将目标染色体细胞分裂中期影像自动化地转换为目标染色体核型影像，并利用目标特征选取模块分析目标染色体核型影像后以得至少一个目标影像特征值的方式可有效降低染色体异常检测时因不同判断者的主观意识所产生的误差。再者，通过具有深度神经网路学习功能的染色体异常检测模型不仅能有效提升染色体异常检测的准确度与敏感度，并可大幅缩短染色体异常的判定时间，使本发明的染色体异常检测模型、染色体异常检测 系统以及染色体异常检测方法在染色体异常检测方面更有效率。
上述发明内容旨在提供本揭示内容的简化摘要，以使阅读者对本揭示内容具备基本的理解。此发明内容并非本揭示内容的完整概述，且其用意并非在指出本发明实施例的重要/关键元件或界定本发明的范围。
附图说明
为让本发明的上述和其他目的、特征、优点与实施例能更明显易懂，结合附图说明如下:
图1绘示依照本发明的一实施方式的一种染色体异常检测模型的建立步骤流程图；
图2绘示依照本发明另一实施方式的一种染色体异常检测方法的步骤流程图；
图3绘示依照本发明再一实施方式的一种染色体异常检测系统的方块图；
图4绘示目标染色体细胞分裂中期影像转换为目标染色体核型影像的结果图；
图5绘示本发明的一实施方式的一实施例的染色体异常检测模型的卷积神经网路学习分类器的架构示意图；
图6绘示本发明的一实施方式的另一实施例的染色体异常检测模型的卷积神经网路学习分类器的架构示意图；以及
图7为本发明的染色体异常检测模型用于判断受试者的染色体异常的混淆矩阵。
具体实施方式
下述将更详细讨论本发明各实施方式。然而，此实施方式可为各种发明概念的应用，可被具体实行在各种不同的特定范围内。特定的实施方式是仅以说明为目的，且不受限于揭露的范围。
请参照图1，绘示依照本发明的一实施方式的一种染色体异常检测模型的建立步骤100流程图。本发明的染色体异常检测模型的建立步骤100包含步骤110、步骤120、步骤130、步骤140和步骤150，建立后的染色体异常检测模型可用以判断受试者是否具有染色体数目异常、染色体结构异常或染色体拼凑型异常。
步骤110是取得参照数据库，所述参照数据库包含多个参照染色体细胞分裂中期影像。在非分裂期的细胞，其染色质多以30nm至300nm的状态分布于细胞核中，当细胞进入有丝分裂期时，染色体才会开始逐步紧密排列。而细胞有丝分裂中期(metaphase)时，细胞的核膜完全消失不见，纺锤丝开始变得清晰。每个染色体上的着丝点分别附着至纺锤丝(或星射线)，着丝点受其两极拉力开始上下移动，最后两极拉力达到均衡，着丝点均排列于细胞中央的赤道板上，为染色体的清晰度达到最高的时点。是以在取得参照染色体细胞分裂中期影像前，先通过施打激素使参照受试者的细胞进入细胞分裂中期后，再抽取参照受试者的特定细胞，并通过染色和显微镜观察取得参照染色体细胞分裂中期影像。
步骤120是进行影像转换步骤，是利用非监督式学习法分类器将参照染色体细胞分裂中期影像中23对染色体进行排列，以得到多个参照染色体核型(karyotype)影像。参照染色体核型影像是将前述的参照染色体细胞分裂中期影像，根据染色体的长度、着丝点位置、长短臂比例、随体的有无等特征，对染色体进行分析、比较、排序和编号后所得到的影像。所述非监督式学习法分类器可为生成对抗神经网络(Generative Adversarial Network，GAN)。
步骤130是进行初步分类步骤，是依据参照染色体核型影像中的染色体条数进行分类，若染色体条数为46条，分类为染色体数目正常；若染色体条数为大于或小于46条，则分类为染色体数目异常。
步骤140是进行特征选取步骤，是利用特征选取模块分析参照染色体核型影像后以取得至少一个影像特征值。其中至少一个影像特征值可包含染色体大小、染色体位置或染色体形状。
步骤150是进行训练步骤，是将前述的至少一个影像特征值通过卷积神经网路学习分类器进行训练而达到收敛，以得到所述染色体异常检测模型。其中所述卷积神经网路学习分类器可为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
请参照图2，绘示依照本发明另一实施方式的一种染色体异常检测方法200的步骤流程图。本发明的染色体异常检测方法200包含步骤210、步骤220、步骤230和步骤240。
步骤210是提供染色体异常检测模型，而染色体异常检测模型是经由前述步骤110至步骤140所建立。
步骤220是提供受试者的目标染色体细胞分裂中期影像，在取得目标染色体细胞分裂中期影像前，先通过施打激素使受试者的细胞进入细胞分裂中期后，再抽取受试者的特定细胞，并通过染色和显微镜观察取得目标染色体细胞分裂中期影像。
步骤230利用非监督式学习法分类器将目标染色体细胞分裂中期影像中23对染色体进行排列，以取得目标染色体核型影像。所述目标染色体核型影像是将前述的目标染色体细胞分裂中期影像，根据染色体的长度、着丝点位置、长短臂比例、随体的有无等特征，对染色体进行分析、比较、排序和编号后所得到的影像。所述非监督式学习法分类器可为生成对抗神经网络(Generative Adversarial Network，GAN)。
步骤240是利用染色体异常检测模型分析所述目标染色体核型影像，以判断受试者是否具有染色体异常。其中染色体异常可包含染色体数目异常、染色体结构异常或染色体拼凑型异常。优选地，染色体数目异常可包含受试者的目标染色体为单倍体或多倍体，染色体结构异常可包含受试者的目标染色体为染色体缺失、环状染色体、染色体转位、染色体倒转或染色体重复。
借此，本发明的染色体异常检测模型与染色体异常检测方法通过将目标染色体细胞分裂中期影像自动化地转换为目标染色体核型影像，并利用特征选取模块分析目标染色体核型影像后以得至少一个影像特征值的方式可有效降低染色体异常检测时因不同判断者的主观意识所产生的误差。再者，通过具有深度神经网路学习功能的染色体异常检测模型不仅能有效提升染色体异常检测的准确度与敏感度，并可大幅缩短染色体异常的判定时间，使本发明的染色体异常检测模型以及染色体异常检测方法在染色体异常检测方面更有效率。
请再参照图3和图4，图3绘示依照本发明再一实施方式的一种染色体异常检测系统300的方块图，图4绘示目标染色体细胞分裂中期影像610转换为目标染色体核型影像620的结果图。本发明的染色体异常检测系统300包含影像撷取单元400和非暂时性机器可读媒体500。染色体异常检测系统300可用以判断受试者是否具有染色体数目异常、染色体结构异常或染色体拼凑型异常。
影像撷取单元400用以取得受试者的目标染色体细胞分裂中期影像610。影像撷取单元可为搭配显微镜的取像装置，用以拍摄显微镜所观察到的染色体影像。
非暂时性机器可读媒体500信号连接影像撷取单元400，其中非暂时性机器可读媒体用以储存程序，当前述的程序由处理单元执行时是用以评估受试者是否具有染色体异常，其中前述的程序包含参照数据库取得模块510、参照影像转换模块520、参照初步分类模块530、参照特征选取模块540、训练模块 550、目标影像转换模块560、目标初步分类模块570、目标特征选取模块580及比对模块590。
参照数据库取得模块510用以取得参照数据库，且前述的参照数据库是由多个参照染色体细胞分裂中期影像所建立。
参照影像转换模块520，其是利用非监督式学习法分类器将参照染色体细胞分裂中期影像中23对染色体进行排列，以取得多个参照染色体核型影像。所述非监督式学习法分类器可为生成对抗神经网络。
参照初步分类模块530用以将参照染色体核型影像依据参照染色体条数进行分类。若参照染色体条数为46条，分类为染色体数目正常；若参照染色体条数为大于或小于46条，则分类为染色体数目异常。优选地，染色体数目异常可包含受试者的目标染色体为单倍体或多倍体。
参照特征选取模块540用以分析参照染色体核型影像后以取得至少一个参照影像特征值。所述至少一个参照影像特征值可包含染色体大小、染色体位置或染色体形状。
训练模块550用以将至少一个参照影像特征值通过卷积神经网路学习分类器训练达到收敛，以得到染色体异常检测模型。所述卷积神经网路学习分类器可为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
目标影像转换模块560是利用非监督式学习法分类器将目标染色体细胞分裂中期影像610中23对染色体进行排列，以取得目标染色体核型影像620。所述非监督式学习法分类器可为生成对抗神经网络。
目标初步分类模块570用以将目标染色体核型影像依据目标染色体条数进行分类。若目标染色体条数为46条，分类为染色体数目正常；若目标染色体条数为大于或小于46条，则分类为染色体数目异常。优选地，染色体数目异常可包含受试者的目标染色体为单倍体或多倍体。
目标特征选取模块580用以分析目标染色体核型影像后以得至少一个目标影像特征值。所述至少一个目标影像特征值可包含染色体大小、染色体位置或染色体形状。
比对模块590用以将目标影像特征值以所述染色体异常检测模型进行分析以得到目标影像特征值权重数据，并依据目标影像特征值权重数据判断受试者是否具有染色体结构异常或染色体拼凑型异常。优选地，染色体结构异常可包含受试者的目标染色体为染色体缺失、环状染色体、染色体转位、染色体倒转或染色体重复。
此外，非暂时性机器可读媒体500可还包含评估模块(图未绘示)，用以依据目标影像特征值权重数据进一步计算受试者具有染色体异常的风险值。
根据上述实施方式，以下提出具体试验例并配合附图予以详细说明。
<试验例>
一、参照数据库
本发明所使用的参照数据库为中国医药大学附属医院(China Medical University Hospital,CMUH)所搜集的回溯性去连结化的受检者临床内容，为经中国医药大学暨附属医院研究伦理委员会核准的临床试验计划，其编号为：CMUH107-REC3-151。前述的参照数据库包含30000笔受检者的参照染色体细胞分裂中期影像，且前述的参照染色体细胞分裂中期影像的所属受检者性别并无特别限制，年龄亦没有特别的区间。
二、本发明的染色体异常检测模型
本发明的染色体异常检测模型在取得参照数据库后，各参照染色体细胞分裂中期影像将利用参照影像转换模块，将各参照染色体细胞分裂中期影像以非监督式学习法分类器将各参照染色体细胞分裂中期影像中23对染色体进行排列，以得到多个参照染色体核型影像。
详细而言，由于目前的深度神经网路模型在运作上需要大量的训练数据(Training Data，即本发明的染色体异常检测模型的各参照染色体细胞分裂中 期影像)来达成稳定收敛及高度的分类准确率，倘若训练数据的数目不够充足将会使深度神经网路产生过拟合现象(Overfitting)而导致判断结果的误差值过高，致使深度神经网路模型的可信度较低。为了解决前述问题，本发明的染色体异常检测模型另包含影像前处理步骤，将各参照染色体核型影像进行进行黑白对比度校正，并将影像数值归一化，使影像数值介于0到1。
先进行初步分类步骤，以判断受检者是否具有染色体数目异常的状况，其是依据各参照染色体核型影像中的染色体条数进行分类。若染色体条数为46条，分类为染色体数目正常；若染色体条数为大于或小于46条，则分类为染色体数目异常。
接着，各参照染色体核型影像将以特征选取模块进行分析，以得至少一个影像特征值。详细而言，特征选取模块可进一步区别各参照染色体核型影像中的染色体大小、染色体位置或染色体形状的影像特征值。
接着，前述的影像特征值将通过卷积神经网路学习分类器进行训练而达到收敛，以得本发明的染色体异常检测模型。在本试验例中，染色体异常检测模型将应用于判断受试者是否具有染色体数目异常、染色体结构异常或染色体拼凑型异常。而卷积神经网路学习分类器可为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
请参照图5，其是绘示本发明的染色体异常检测模型的卷积神经网路学习分类器700的架构示意图。在图5的试验例中，卷积神经网路学习分类器700为Inception-ResNet-v2卷积神经网路，其包含多个卷积层(Convolution)、多个最大池化层(MaxPool)、多个平均池化层(AvgPool)以及多个级联层(Concat)，以对影像特征值进行训练与分析。
详细而言，Inception-ResNet-v2卷积神经网路是基于ImageNet可视化数据数据库的大规模视觉辨识卷积神经网路，且ImageNet可视化数据数据库里面的影像数据皆为二维的彩色图像，因此公知的GoogLeNet卷积神经网路模型在其第一卷积层中具有RGB三通道的滤波器。然而，各参照染色体核型影像 的原始影像档案皆为三维的灰阶影像，是以本发明的染色体异常检测模型进一步将包含RGB三通道的滤波器的GoogLeNet卷积神经网路模型通过算术平均法而转换为单一通道，并将随机梯度下降法(Stochastic Gradient Descent，SGD)应用于本发明的染色体异常检测模型的预训练模型神经网路中，以优化其训练过程，其训练次数可为100期(Epochs)及采用96Mini-Batch Size的梯度下降法，并通过改变初始学习率(Learning Rates)以进行调变，其中学习率是对神经网路进行训练时控制权重(weight)和偏差(bias)变化的重要参数，是以本发明的染色体异常检测模型通过调整学习率的数值可进一步确保损失函数(Loss Function)可达稳定收敛。
在本发明的染色体异常检测模型对影像特征值进行训练的过程中，各参照染色体核型影像的影像特征值进行二层卷积层及一层最大池化层(MaxPool)处理，以将所提取的影像特征值进行最大输出，并再次重复前述的二层卷积层与一层最大池化层输出后，利用多个卷积层进行并行塔(parallel towers)训练，以完成影像特征值的初级训练(Inception)。
在完成前述的初级训练后，各参照染色体核型影像的影像特征值将进行10次(10×)、20次(20×)与10次(10×)的不同深度、不同阶层与不同方面的残差(Residual)模块训练，以对各参照染色体核型影像的影像特征值进行训练并达到收敛。详细而言，由于Inception-ResNet卷积神经网路在经过多个阶层的权重运算后，因为每一残差模块均对各标准化足内侧位X光影像数据的影像特征值进行不同的运算与判断，致使误差累积，因此Inception-ResNet卷积神经网路的训练将会把特定阶层的节点运算值拉回到该阶层的输入端再次进行运算，以防止卷积神经网路学习分类器700对前述的影像特征值进行多层的权重运算训练后发生梯度消失的退化现象，以及避免误差累积导致信息遗失，并可有效提升卷积神经网路学习分类器700的训练效率。
在完成深层且重复的残差模块训练后，将依序以一层卷积层、平均池化层、取代全局平均池化层(Global Average Pooling 2D，GloAvePool2D)以及线 性整流单元训练层(Rectified Linear Unit，ReLU)对收敛的影像特征值进行最终训练与处理，借以判断受试者的染色体异常情况。其中，平均池化层可先对完成残差模块训练的影像特征值进行计算，以求各影像特征值的平均值，取代全局平均池化层则可对卷积神经网路学习分类器700的整体网路架构进行正则化(Regularization)处理，防止卷积神经网路学习分类器700在追求低误差的训练模式下发生过拟合现象，而导致判断结果的误差值过高，最后，线性整流单元训练层则进一步对完成训练后的影像特征值进行激活，并输出目标影像特征值权重数据701，以进行后续的比对与分析。前述的线性整流单元训练层可避免足畸形检测模型输出的目标影像特征值权重数据701趋近于零或趋近于无限大，以利于后续比对步骤的进行，进而提升本发明的染色体异常检测模型的判断准确率。
接着，前述受试者的染色体异常状况判断结果将进一步整合于参照数据库中，以对本发明的染色体异常检测模型进行优化，进而使本发明的染色体异常检测模型的训练效果及判断准确度进一步提升。
请再参照图6，其绘示本发明的染色体异常检测模型的卷积神经网路学习分类器800的架构示意图。在图6的试验例中，卷积神经网路学习分类器800为Inception V3卷积神经网路，其包含多个卷积层(Convolution)、多个平均池化层(AvgPool)、多个最大池化层(MaxPool)以及多个级联层(Concat)，并利用丢弃层(Dropout)、全连结层(Fully connected)和归一化层(Softmax)解决机器学习上过拟合的问题，以对影像特征值进行训练与分析。
单层的神经网路会因为参数过多，而导致机器学习上过拟合的问题。Inception V3卷积神经网路为基于大滤波器尺寸分解卷积网路的因式分解，以平行式参数降阶，既可解决过拟合的问题，又可通过增加网路深度，来增加参数的数目进而更近似原本欲近似的数学模型。
在本发明的染色体异常检测模型对影像特征值进行训练的过程中，各参照染色体核型影像的影像特征值分别进行一层平均池化层和一层卷积层；五 层卷积层；三层卷积层；一层卷积层运算后，将各组运算的特征矩阵数值以级联层迭合。之后再重复2次分别进行一层平均池化层和一层卷积层；五层卷积层；三层卷积层；一层卷积层运算后，并将各组运算的特征矩阵数值以级联层迭合。再分别进行一层最大池化层；三层卷积层；一层卷积层运算后，将各组运算的特征矩阵数值以级联层迭合。之后再重复4次分别进行一层平均池化层和一层卷积层；五层卷积层；三层卷积层和一层卷积层运算后，并将各组运算的特征矩阵数值以级联层迭合。再进行一层平均池化层、二层卷积层、一层全连结层和一层规一化层运算，运算的特征矩阵数值再重复2次分别进行一层平均池化层和一层卷积层；三层卷积层和一层级联层；二层卷积层和一层级联层；一层卷积层运算后，将各组运算的特征矩阵数值以级联层迭合。最后再进行一层平均池化层、一层丢弃层、一层全连结层和一层规一化层运算后，输出目标影像特征值权重数据801，以得到训练好的染色体异常检测模型。
接着，前述受试者的染色体异常状况判断结果将进一步整合于参照数据库中，以对本发明的染色体异常检测模型进行优化，进而使本发明的染色体异常检测模型的训练效果及判断准确度进一步提升。
请再参照图7，为本发明的染色体异常检测模型用于判断受试者的染色体异常的混淆矩阵。于图7的试验例中，建立染色体异常检测模型的卷积神经网路学习分类器为图6绘示的卷积神经网路学习分类器800来判断受试者的染色体是否异常，并将结果分为正常和异常。其中横轴为预测标签，纵轴为实际标签，可将混淆矩阵区分为真阳性(True Positive,TP)、真阴性(True Negative,TN)、伪阳性(False Positive,FP)和伪阴性(False Negative,FN)四部分，并依据TP、TN、FP和FN的数据计算本发明的染色体异常检测模型的正确率、灵敏度、特异度、阳性预测值和阴性预测值。其中正确率的计算方式为(TP+TN)/(TP+FP+TN+FN)，灵敏度的计算方式为TP/(TP+FN)，特异度的计算方式为 TN/(TN+FP)，阳性预测值的计算方式为TP/(TP+FP)，阴性预测值的计算方式为TN/(FN+TN)。
如图7的结果显示，TP区块的受试者数量为206人，TN区块的受试者数量为201人，FP区块的受试者数量为3人，FN区块的受试者数量为0人。经计算后，本发明的染色体异常检测模型用于判断受试者的染色体异常的预测结果如表一所示。
Figure PCTCN2019075693-appb-000001
由上述结果显见本发明的染色体异常检测模型可用以精准的判断受试者是否具有染色体异常状况，且染色体异常状况可包含染色体数目异常、染色体结构异常和染色体拼凑型异常。
借此，本发明的染色体异常检测系统可有效提升染色体异常检测的准确度与敏感度，并可缩短受试者是否具有染色体异常的评估时间，从原始影像输入到判读结果，平均只需0.1-1秒即可完成，使其运用更为广泛。
然本发明已以实施方式公开如上，然其并非用以限定本发明，任何本领域的技术人员，在不脱离本发明的精神和范围内，当可作各种的更动与润饰，因此本发明的保护范围当视权利要求所界定的为准。

Claims (13)

  1. 一种染色体异常检测模型，其特征在于，包含以下建立步骤：
    取得参照数据库，其中所述参照数据库包含多个参照染色体细胞分裂中期影像；
    进行影像转换步骤，其是利用非监督式学习法分类器将所述多个参照染色体细胞分裂中期影像中23对染色体进行排列，以得到多个参照染色体核型影像；
    进行初步分类步骤，其是依据所述多个参照染色体核型影像中的染色体条数进行分类，若所述染色体条数为46条，分类为染色体数目正常，若所述染色体条数为大于或小于46条，则分类为染色体数目异常；
    进行特征选取步骤，其是利用特征选取模块分析所述多个参照染色体核型影像后以得到至少一个影像特征值；以及
    进行训练步骤，其是将所述至少一个影像特征值通过卷积神经网路学习分类器进行训练而达到收敛，以得到所述染色体异常检测模型，其中所述染色体异常检测模型是用以判断受试者是否具有染色体结构异常或染色体拼凑型异常。
  2. 如权利要求1所述的染色体异常检测模型，其特征在于，所述非监督式学习法分类器为生成对抗神经网络。
  3. 如权利要求1所述的染色体异常检测模型，其特征在于，所述至少一个影像特征值包含染色体大小、染色体位置或染色体形状。
  4. 如权利要求1所述的染色体异常检测模型，其特征在于，所述卷积神经网路学习分类器为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
  5. 一种染色体异常检测方法，其特征在于，包含：
    提供如权利要求1所述的染色体异常检测模型；
    提供受试者的目标染色体细胞分裂中期影像；
    利用所述非监督式学习法分类器将所述目标染色体细胞分裂中期影像中23对染色体进行排列，以得到目标染色体核型影像；以及
    利用所述染色体异常检测模型分析所述目标染色体核型影像，以判断所述受试者是否具有染色体异常。
  6. 如权利要求5所述的染色体异常检测方法，其特征在于，所述染色体异常包含染色体数目异常、染色体结构异常或染色体拼凑型异常。
  7. 如权利要求6所述的染色体异常检测方法，其特征在于，所述染色体数目异常包含所述受试者的目标染色体为单倍体或多倍体。
  8. 如权利要求6所述的染色体异常检测方法，其特征在于，所述染色体结构异常包含所述受试者的目标染色体为染色体缺失、环状染色体、染色体转位、染色体倒转或染色体重复。
  9. 一种染色体异常检测系统，其特征在于，包含：
    影像撷取单元，用以取得受试者的目标染色体细胞分裂中期影像；以及
    非暂时性机器可读媒体，信号连接所述影像撷取单元，其中所述非暂时性机器可读媒体用以储存程序，当所述程序由处理单元执行时是用以判断所述受试者是否具有染色体异常，且所述程序包含：
    参照数据库取得模块，用以取得参照数据库，且所述参照数据库是由多个参照染色体细胞分裂中期影像所建立；
    参照影像转换模块，其是利用非监督式学习法分类器将所述多个 参照染色体细胞分裂中期影像中23对染色体进行排列，以得到多个参照染色体核型影像；
    参照初步分类模块，用以将所述多个参照染色体核型影像依据参照染色体条数进行分类，若所述参照染色体条数为46条，分类为染色体数目正常，若所述参照染色体条数为大于或小于46条，则分类为染色体数目异常；
    参照特征选取模块，用以分析所述多个参照染色体核型影像后以取得至少一个参照影像特征值；
    训练模块，用以将所述至少一个参照影像特征值通过卷积神经网路学习分类器训练达到收敛，以得到染色体异常检测模型；
    目标影像转换模块，其是利用所述非监督式学习法分类器将所述目标染色体细胞分裂中期影像中23对染色体进行排列，以得到目标染色体核型影像；
    目标初步分类模块，用以将所述目标染色体核型影像依据目标染色体条数进行分类，若所述目标染色体条数为46条，分类为染色体数目正常，若所述目标染色体条数为大于或小于46条，则分类为染色体数目异常；
    目标特征选取模块，其是用以分析所述目标染色体核型影像后以得至少一个目标影像特征值；及
    比对模块，其是用以将所述目标影像特征值以所述染色体异常检测模型进行分析以得目标影像特征值权重数据，并依据所述目标影像特征值权重数据判断所述受试者是否具有染色体结构异常或染色体拼凑型异常。
  10. 如权利要求9所述的染色体异常检测系统，其特征在于，所述非监 督式学习法分类器为生成对抗神经网络。
  11. 如权利要求9所述的染色体异常检测系统，其特征在于，所述至少一个参照影像特征值包含染色体大小、染色体位置或染色体形状，所述至少一个目标影像特征值包含染色体大小、染色体位置或染色体形状。
  12. 如权利要求9所述的染色体异常检测系统，其特征在于，所述卷积神经网路学习分类器为Inception-ResNet-v2卷积神经网路或Inception V3卷积神经网路。
  13. 如权利要求9所述的染色体异常检测系统，其特征在于，所述非暂时性机器可读媒体还包含：
    评估模块，用以依据所述目标影像特征值权重数据计算所述受试者具有所述染色体异常的风险值。
PCT/CN2019/075693 2019-02-21 2019-02-21 染色体异常检测模型、其检测系统及染色体异常检测方法 WO2020168511A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2019/075693 WO2020168511A1 (zh) 2019-02-21 2019-02-21 染色体异常检测模型、其检测系统及染色体异常检测方法
JP2021505288A JP2021531812A (ja) 2019-02-21 2019-02-21 染色体異常のテストモデル、そのテストシステム及び染色体異常のテスト方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/075693 WO2020168511A1 (zh) 2019-02-21 2019-02-21 染色体异常检测模型、其检测系统及染色体异常检测方法

Publications (1)

Publication Number Publication Date
WO2020168511A1 true WO2020168511A1 (zh) 2020-08-27

Family

ID=72143987

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/075693 WO2020168511A1 (zh) 2019-02-21 2019-02-21 染色体异常检测模型、其检测系统及染色体异常检测方法

Country Status (2)

Country Link
JP (1) JP2021531812A (zh)
WO (1) WO2020168511A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288706A (zh) * 2020-10-27 2021-01-29 武汉大学 一种自动化的染色体核型分析以及异常检测方法
CN112382384A (zh) * 2020-11-10 2021-02-19 中国科学院自动化研究所 特纳综合征诊断模型训练方法、诊断系统及相关设备
CN112487942A (zh) * 2020-11-26 2021-03-12 华南师范大学 染色体实例分割方法、系统和存储介质
CN114219786A (zh) * 2021-12-16 2022-03-22 易构智能科技(广州)有限公司 一种基于深度学习的染色体核型分析方法及系统
CN115375682A (zh) * 2022-10-24 2022-11-22 湖南自兴智慧医疗科技有限公司 一种染色体罗氏易位异常检测方法、系统及存储介质
CN115410050A (zh) * 2022-11-02 2022-11-29 杭州华得森生物技术有限公司 基于机器视觉的肿瘤细胞检测设备及其方法
CN117152147A (zh) * 2023-10-31 2023-12-01 杭州德适生物科技有限公司 一种在线染色体协同分析方法、系统及介质
WO2023240820A1 (zh) * 2022-06-17 2023-12-21 广州智睿医疗科技有限公司 一种染色体核型分析模块
CN118626878A (zh) * 2024-08-08 2024-09-10 绵阳梓兴食品科技有限公司 公猪精液溯源管理系统及方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108763874A (zh) * 2018-05-25 2018-11-06 南京大学 一种基于生成对抗网络的染色体分类方法及装置
CN109214285A (zh) * 2018-08-01 2019-01-15 浙江深眸科技有限公司 基于深度卷积神经网络与长短期记忆网络的摔倒检测方法
CN109285174A (zh) * 2017-07-19 2019-01-29 塔塔咨询服务公司 基于众包和深度学习的染色体分割和核型分析
CN109300111A (zh) * 2018-08-27 2019-02-01 杭州德适生物科技有限公司 一种基于深度学习的染色体识别方法
CN109344874A (zh) * 2018-08-31 2019-02-15 华侨大学 一种基于深度学习的染色体自动分析方法及系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6669796B2 (ja) * 2017-03-29 2020-03-18 Ｊｆｅテクノリサーチ株式会社 染色体異常判定装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109285174A (zh) * 2017-07-19 2019-01-29 塔塔咨询服务公司 基于众包和深度学习的染色体分割和核型分析
CN108763874A (zh) * 2018-05-25 2018-11-06 南京大学 一种基于生成对抗网络的染色体分类方法及装置
CN109214285A (zh) * 2018-08-01 2019-01-15 浙江深眸科技有限公司 基于深度卷积神经网络与长短期记忆网络的摔倒检测方法
CN109300111A (zh) * 2018-08-27 2019-02-01 杭州德适生物科技有限公司 一种基于深度学习的染色体识别方法
CN109344874A (zh) * 2018-08-31 2019-02-15 华侨大学 一种基于深度学习的染色体自动分析方法及系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
FAN, TAO: "Automatic Chromosome Recognition of Human with Fuzzy Hopfield Neural Network", MASTER THESIS, 15 June 2002 (2002-06-15), CN, XP55730315 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288706A (zh) * 2020-10-27 2021-01-29 武汉大学 一种自动化的染色体核型分析以及异常检测方法
CN112288706B (zh) * 2020-10-27 2022-06-24 武汉大学 一种自动化的染色体核型分析以及异常检测方法
CN112382384A (zh) * 2020-11-10 2021-02-19 中国科学院自动化研究所 特纳综合征诊断模型训练方法、诊断系统及相关设备
CN112487942A (zh) * 2020-11-26 2021-03-12 华南师范大学 染色体实例分割方法、系统和存储介质
CN114219786A (zh) * 2021-12-16 2022-03-22 易构智能科技(广州)有限公司 一种基于深度学习的染色体核型分析方法及系统
CN114219786B (zh) * 2021-12-16 2022-07-05 易构智能科技(广州)有限公司 一种基于深度学习的染色体核型分析方法及系统
WO2023240820A1 (zh) * 2022-06-17 2023-12-21 广州智睿医疗科技有限公司 一种染色体核型分析模块
CN115375682A (zh) * 2022-10-24 2022-11-22 湖南自兴智慧医疗科技有限公司 一种染色体罗氏易位异常检测方法、系统及存储介质
CN115410050A (zh) * 2022-11-02 2022-11-29 杭州华得森生物技术有限公司 基于机器视觉的肿瘤细胞检测设备及其方法
CN117152147A (zh) * 2023-10-31 2023-12-01 杭州德适生物科技有限公司 一种在线染色体协同分析方法、系统及介质
CN117152147B (zh) * 2023-10-31 2024-02-09 杭州德适生物科技有限公司 一种在线染色体协同分析方法、系统及介质
CN118626878A (zh) * 2024-08-08 2024-09-10 绵阳梓兴食品科技有限公司 公猪精液溯源管理系统及方法

Also Published As

Publication number Publication date
JP2021531812A (ja) 2021-11-25

Similar Documents

Publication Publication Date Title
TWI687937B (zh) 染色體異常檢測模型之建立方法、染色體異常檢測系統及染色體異常檢測方法
WO2020168511A1 (zh) 染色体异常检测模型、其检测系统及染色体异常检测方法
EP3633682A1 (en) Chromosome abnormality detecting model, detecting system thereof, and method for detecting chromosome abnormality
US11436493B2 (en) Chromosome recognition method based on deep learning
WO2020024127A1 (zh) 骨龄评估与身高预测模型、其系统及其预测方法
WO2021062904A1 (zh) 基于病理图像的tmb分类方法、系统及tmb分析装置
WO2020103683A1 (zh) 基于脑功能图谱的猴-人跨物种迁移进行精神疾病的个体化预测方法和系统
Chamidah et al. Fetal state classification from cardiotocography based on feature extraction using hybrid K-Means and support vector machine
TWI684997B (zh) 骨齡評估與身高預測模型之建立方法、骨齡評估與身高預測系統及骨齡評估與身高預測方法
CN108778287B (zh) 用于早产结果的早期风险评估的方法和系统
CN113658174A (zh) 基于深度学习和图像处理算法的微核组学图像检测方法
WO2023198224A1 (zh) 一种精神障碍类磁共振图像初步筛查模型构建方法
CN113782184A (zh) 一种基于面部关键点与特征预学习的脑卒中辅助评估系统
WO2021134513A1 (zh) 确定染色体非整倍性、构建分类模型的方法和装置
CN118285774A (zh) 一种基于双注意力多模态融合的产时胎监智能判别方法
Verma et al. A novel framework for fetal nuchal translucency abnormality detection using hybrid maxpool matrix histogram analysis
WO2023240820A1 (zh) 一种染色体核型分析模块
WO2023024524A1 (zh) 基于胎儿超声影像特征组学的染色体异常预测模型的构建方法及诊断设备
CN113593629B (zh) 基于半导体测序的降低无创产前检测假阳性假阴性的方法
CN110010246A (zh) 一种基于神经网络和置信区间的疾病智能诊断技术
CN115661101A (zh) 基于随机采样和深度学习的早产儿视网膜病变检测系统
Sumon et al. Using Deep Learning Systems for Imaging Methods for Recognising Brain Tumors
CN115223654A (zh) 检测胎儿染色体非整倍体异常的方法、装置及存储介质
CN114512231A (zh) 基于级联特征选择算法的唐氏综合征筛查系统
CN113033330A (zh) 一种基于轻型卷积神经网络的舌体姿态异常判别方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19916059

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021505288

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19916059

Country of ref document: EP

Kind code of ref document: A1