CN113456064A - Intelligent interpretation method for prenatal fetal heart monitoring signal - Google Patents
Intelligent interpretation method for prenatal fetal heart monitoring signal Download PDFInfo
- Publication number
- CN113456064A CN113456064A CN202110798011.9A CN202110798011A CN113456064A CN 113456064 A CN113456064 A CN 113456064A CN 202110798011 A CN202110798011 A CN 202110798011A CN 113456064 A CN113456064 A CN 113456064A
- Authority
- CN
- China
- Prior art keywords
- signal
- fetal heart
- fetal
- heart rate
- prenatal
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
- 210000002458 fetal heart Anatomy 0.000 title claims abstract description 99
- 238000012544 monitoring process Methods 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000001605 fetal effect Effects 0.000 claims abstract description 66
- 208000036029 Uterine contractions during pregnancy Diseases 0.000 claims abstract description 49
- 238000012545 processing Methods 0.000 claims abstract description 25
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 20
- 230000011218 segmentation Effects 0.000 claims abstract description 20
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 230000006835 compression Effects 0.000 claims abstract description 5
- 238000007906 compression Methods 0.000 claims abstract description 5
- 230000004927 fusion Effects 0.000 claims abstract description 4
- 230000002159 abnormal effect Effects 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 14
- 238000012217 deletion Methods 0.000 claims description 9
- 230000037430 deletion Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000013136 deep learning model Methods 0.000 abstract description 11
- 230000006870 function Effects 0.000 abstract description 3
- 210000003754 fetus Anatomy 0.000 description 14
- 238000010835 comparative analysis Methods 0.000 description 13
- 238000012795 verification Methods 0.000 description 11
- 230000036541 health Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000002427 irreversible effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000007621 cluster analysis Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 206010067484 Adverse reaction Diseases 0.000 description 1
- 208000001951 Fetal Death Diseases 0.000 description 1
- 206010055690 Foetal death Diseases 0.000 description 1
- 241000288105 Grus Species 0.000 description 1
- 206010021143 Hypoxia Diseases 0.000 description 1
- 230000006838 adverse reaction Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 231100000479 fetal death Toxicity 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010247 heart contraction Effects 0.000 description 1
- 230000001146 hypoxic effect Effects 0.000 description 1
- 238000002350 laparotomy Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 230000035935 pregnancy Effects 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1118—Determining activity level
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
- A61B5/02411—Detecting, measuring or recording pulse rate or heart rate of foetuses
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/43—Detecting, measuring or recording for evaluating the reproductive systems
- A61B5/4306—Detecting, measuring or recording for evaluating the reproductive systems for evaluating the female reproductive systems, e.g. gynaecological evaluations
- A61B5/4343—Pregnancy and labour monitoring, e.g. for labour onset detection
- A61B5/4356—Assessing uterine contractions
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/43—Detecting, measuring or recording for evaluating the reproductive systems
- A61B5/4306—Detecting, measuring or recording for evaluating the reproductive systems for evaluating the female reproductive systems, e.g. gynaecological evaluations
- A61B5/4343—Pregnancy and labour monitoring, e.g. for labour onset detection
- A61B5/4362—Assessing foetal parameters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2503/00—Evaluating a particular growth phase or type of persons or animals
- A61B2503/02—Foetus
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Physiology (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Gynecology & Obstetrics (AREA)
- Pregnancy & Childbirth (AREA)
- Reproductive Health (AREA)
- Cardiology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Pediatric Medicine (AREA)
- Dentistry (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
The invention discloses an intelligent prenatal fetal heart monitoring signal interpretation method, which comprises the steps of preprocessing a fetal heart rate signal, a uterine contraction pressure signal and a fetal movement signal to obtain a fusion multi-signal data set; then, carrying out segmentation processing to construct a data set to be classified; then sequentially inputting the embedding layer, the splicing layer, the bidirectional gating circulation unit layer and the full connection layer of the prenatal fetal monitoring intelligent interpretation model, and carrying out sigmoid function compression on the output result to obtain a classification discrimination result; according to the intelligent interpretation method, the uterine contraction pressure signal and the fetal movement signal are merged into the prenatal fetal monitoring intelligent interpretation model based on the bidirectional gating circulation unit, and compared with other deep learning models, the intelligent interpretation method is short in time consumption and has better classification and discrimination capability.
Description
Technical Field
The invention relates to a deep learning method, in particular to an intelligent prenatal fetal heart monitoring signal interpretation method which is used for carrying out intelligent classification and judgment on prenatal fetal condition assessment.
Background
Prenatal fetal monitoring is widely used for evaluating the health development condition of a fetus and timely treating the fetus before adverse reactions occur to the fetus. A fetal heart uterine contraction monitoring Chart (CTG) is an important tool for monitoring the prenatal health of a fetus, and records the change of the fetal heart rate of a pregnant woman during pregnancy and the time relation between the change and uterine contraction. Its role includes, but is not limited to, determining whether the fetus is hypoxic, determining whether delivery of the pregnant woman via caesarean section is required, and guiding other assessments of the fetus' health.
At this stage, the interpretation of fetal heart contraction monitoring Charts (CTG) is under the responsibility of the obstetrician. Interpretation is based on clinical experience derived primarily from antenatal fetal monitoring guidelines and the obstetrician himself. However, the existing fetal heart uterine contraction monitoring Chart (CTG) has inconsistent interpretation standards and inconsistent experience levels of obstetricians, so that the interpretation process is very subjective, interpretation results cannot be reproduced, ambiguity easily occurs, misdiagnosis is caused, and irreparable damage is caused to pregnant women and fetuses.
Therefore, numerous scholars at home and abroad begin to research on prenatal fetal monitoring intelligent classification and judgment by using a machine learning algorithm, the used data is mostly public fetal monitoring standard CTG data sets of a University of California Irvine mechanical learning knowledge base (UCI), and 21 pieces of clinical characteristic data are obtained by using fetal heart rate and uterine contraction signals to perform characteristic extraction for classification.
However, most of the automatic interpretation prenatal fetal heart and uterine contraction map models based on machine learning at the present stage are modeled by using features, and measurement errors which cannot be eliminated exist in feature extraction. Meanwhile, after feature extraction, the interpretation results obtained according to different antenatal fetal monitoring interpretation guidelines are also inconsistent. The model using signal modeling only considers the fetal heart rate signal, ignores the relation between the fetal heart rate signal and the uterine contraction pressure signal, does not deeply research the influence of the uterine contraction pressure signal and the fetal movement signal on the health condition of the fetus, and has a too simple signal preprocessing flow.
In general, the prenatal fetal monitoring machines used at home and abroad at present still cannot reach the intelligent level, the accuracy rate of the obtained suspicious discrimination is only 45-82%, the accuracy rate of the abnormal discrimination is only 66-94%, and the prenatal fetal monitoring machines cannot be applied to clinical work for a long time. Therefore, how to further improve the interpretation accuracy of the prenatal fetal monitoring intelligent model becomes a technical problem to be solved urgently in the field.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides an intelligent prenatal fetal heart monitoring signal interpretation method.
The invention adopts the following technical scheme:
an intelligent prenatal fetal heart monitoring signal interpretation method comprises the following steps:
s1: acquiring original CTG signal data comprising a fetal heart rate signal, a uterine contraction pressure signal and a fetal movement signal;
s2: preprocessing the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal to form a fusion multi-signal data set;
s3: carrying out segmentation processing on the fused multi-signal data set to obtain one-dimensional fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments with the signal lengths of d, and constructing a data set to be classified;
s4: inputting the data set to be classified into a pre-trained prenatal fetal monitoring intelligent interpretation model for classification and judgment, wherein the prenatal fetal monitoring intelligent interpretation model comprises an embedding layer, a splicing layer, a bidirectional gating circulation unit layer and a full connection layer;
respectively inputting the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal segment of which the signal lengths are d into the embedding layer to obtain the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal segment of a two-dimensional matrix (dxm), wherein m is the output dimension of the embedding layer;
inputting fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments of the two-dimensional matrix (d x m) into a splicing layer for splicing to obtain a two-dimensional matrix (d x 3 m) vector;
inputting the two-dimensional matrix (d multiplied by 3 m) vector into the bidirectional gating circulation unit layers of 2k units for splicing to obtain a one-dimensional output vector with the length of 2 k;
inputting the one-dimensional output vector with the length of 2k to a full-connection layer of n units, and carrying out sigmoid function compression on an output result to output the probability that the sample is a normal class and the probability that the sample is an abnormal class;
respectively setting corresponding class labels for the probabilities that the sample is in the normal class and the sample is in the abnormal class;
and comparing the probability that the sample is a normal class with the probability that the sample is an abnormal class, and selecting the class label corresponding to the higher probability as a classification judgment result.
Preferably, the preprocessing of step S2 includes performing interpolation or deletion processing on the fetal heart rate signal, the uterine contraction pressure signal, and the fetal movement signal, respectively.
Preferably, the preprocessing of step S2 further includes normalizing the interpolated or deleted fetal heart rate signal.
Preferably, the normalization process comprises calculating a baseline value of the fetal heart rate signal and subtracting the baseline value from the interpolated or deleted processed fetal heart rate signal to obtain a standard fetal heart rate signal.
Preferably, the preprocessing of step S2 further includes performing sliding window segmentation on the interpolated or deleted fetal heart rate signal before performing normalization processing on the interpolated or deleted fetal heart rate signal, so as to obtain a segment of the fetal heart rate signal with a signal length not less than p.
Preferably, the segmentation process of step S3 includes synchronously performing sliding window segmentation on the preprocessed fetal heart rate signal, the preprocessed uterine contraction pressure signal and the preprocessed fetal movement signal.
The invention has the following beneficial effects:
firstly, the uterine contraction pressure signal and the fetal movement signal are merged into the prenatal fetal monitoring intelligent interpretation model based on the bidirectional gating circulation unit, and compared with other deep learning models, the intelligent interpretation method not only is short in time consumption, but also has better classification and discrimination capability, and can quickly and effectively provide auxiliary decision support for prenatal fetal monitoring for obstetrical medical care personnel.
Secondly, the invention not only carries out preprocessing according to the data characteristics of the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal, but also carries out sliding window segmentation processing on the preprocessed standard fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal, thereby realizing data enhancement, unifying the signal length and further improving the classification and discrimination performance of the intelligent interpretation model.
Thirdly, the prenatal fetal heart monitoring signal intelligent interpretation method provided by the invention uses signal modeling, does not need to extract the morphological characteristics of a fetal heart uterine contraction monitoring Chart (CTG), and effectively avoids the measurement error which can not be eliminated when the signal extracts the clinical morphological characteristics.
Fourthly, the prenatal fetal heart monitoring signal intelligent interpretation method provided by the invention can effectively reduce the workload of medical care personnel, effectively reduce the fetal death rate and the laparotomy yield, avoid unnecessary medical intervention, ensure the normal growth and development of the fetus and contribute to improving the quality of the birth population in China.
Drawings
Fig. 1 is a schematic algorithm flow diagram of the prenatal fetal monitoring intelligent interpretation model of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be further described with reference to the accompanying drawings and specific embodiments of the specification.
Example 1
The invention provides an intelligent prenatal fetal heart monitoring signal interpretation method, which comprises the following steps:
s1: acquiring original CTG signal data comprising a fetal heart rate signal, a uterine contraction pressure signal and a fetal movement signal;
s2: preprocessing the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal to form a fusion multi-signal data set;
the preprocessing comprises the step of carrying out interpolation or deletion processing on the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal respectively;
the preprocessing further comprises the step of carrying out standardization processing on the fetal heart rate signals subjected to interpolation or deletion processing;
the preprocessing further comprises the step of carrying out sliding window segmentation on the fetal heart rate signals subjected to interpolation or deletion processing before carrying out standardization processing on the fetal heart rate signals subjected to interpolation or deletion processing to obtain fetal heart rate signal segments with signal length not less than p; wherein p is 750;
the normalization process includes calculating a baseline value of the fetal heart rate signal and subtracting the baseline value from the interpolated or deleted fetal heart rate signal to obtain a standard fetal heart rate signal;
s3: carrying out segmentation processing on the fused multi-signal data set to obtain one-dimensional fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments with the signal lengths of d, and constructing a data set to be classified; wherein d is 1125;
and the segmentation processing comprises synchronously carrying out sliding window segmentation on the preprocessed fetal heart rate signal, the preprocessed uterine contraction pressure signal and the preprocessed fetal movement signal.
S4: inputting the data set to be classified into a pre-trained prenatal fetal monitoring intelligent interpretation model for classification and judgment, wherein the prenatal fetal monitoring intelligent interpretation model comprises an embedding layer, a splicing layer, a bidirectional gating circulation unit layer and a full connection layer, and is shown in figure 1;
respectively inputting the one-dimensional fetal heart rate signals, the uterine contraction pressure signals and the fetal movement signal segments with the signal lengths of d into the embedding layer to obtain the fetal heart rate signals, the uterine contraction pressure signals and the fetal movement signal segments of a two-dimensional matrix (dxm), wherein m is the output dimension of the embedding layer; wherein m is 4;
inputting fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments of the two-dimensional matrix (d x m) into a splicing layer for splicing to obtain a two-dimensional matrix (d x 3 m) vector;
inputting the two-dimensional matrix (d multiplied by 3 m) vector into the bidirectional gating circulation unit layers of 2k units for splicing to obtain a one-dimensional output vector with the length of 2 k; k is the number of GRU units in one direction, and 2k is the number of GRU units in the bidirectional gating circulation unit layer; wherein k is 512 and 2k is 1024;
inputting the one-dimensional output vector with the length of 2k to a full-connection layer of n units, and carrying out sigmoid function compression on an output result to output the probability that the sample is a normal class and the probability that the sample is an abnormal class; wherein n is 2;
respectively setting corresponding class labels for the probabilities that the sample is in the normal class and the sample is in the abnormal class; the label of the sample as a normal class is 0, and the label of the sample as an abnormal class is 1.
And comparing the probability that the sample is a normal class with the probability that the sample is an abnormal class, and selecting the class label corresponding to the higher probability as a classification judgment result.
According to the common knowledge in the field, the abnormal values and the missing values of the original fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal can be solved by interpolation or deletion of the missing segment. The fetal heart rate signal and the uterine contraction pressure signal have consistency, and when the fetal heart rate signal is preprocessed, interpolation or deletion processing can be synchronously performed on the uterine contraction pressure signal. The fetal movement signals are extracted point by point in the uterine contraction pressure signals and have consistency, so that interpolation or deletion processing can be synchronously performed on the fetal movement signals. The interpolation of the fetal heart rate signal and the uterine contraction pressure signal selects a median, and the fluctuation of the interpolated sequence is stable, so that the interpolation value of fetal movement is 0 when the fetal heart rate signal and the uterine contraction pressure signal are processed.
Since the signal length in the raw CTG signal data after preprocessing is changed from 20min (1500 points) to within the signal length range of 10min (750 points) to 20min (1500 points), the invention discards signals with a signal length less than 10min (750 points) by performing sliding window segmentation on the preprocessed fetal heart rate signals.
The invention performs sliding window segmentation by adopting the distance between the maximum points. Specifically, smooth noise reduction is carried out on the preprocessed fetal heart rate signal through an SG filtering (Savitzky-Golay) digital filter, and the time corresponding to an extreme point of the signal is not changed; then finding out a set of all maximum value points in the fetal heart rate signal, wherein the maximum value points are judged through a first-order difference of two adjacent points; then, taking the signal length of 10min as a section to perform window sliding treatment, namely taking the first point in the maximum value point set as an end point and taking the forward 10min as a first section, then starting sliding the window, taking the next maximum value point as an end point and taking the forward 10min as a new section, and performing circulation; and adding the head and the tail of the slid data to obtain a sliding segmentation result on the signal.
According to the common knowledge in the art, Baseline d is defined as the mean fetal heart rate at which amplitude levels within 5bpm are stable within 10 min. According to the invention, the extreme points of the fetal heart rate curve are subjected to cluster analysis, baseline points are extracted from all the extreme points, and the baseline points are averaged to obtain the baseline value of the fetal heart rate signal. Specifically, a set of all extreme points is found out, all extreme values are subjected to decentralization, and then the obtained decentralization extreme points are subjected to cluster analysis by adopting a K-Means algorithm; marking and distinguishing a baseline part and a non-baseline part of the obtained result, and then taking an average value t of time abscissas of all baseline points through the baseline partmMean value ƒ of fetal heart rate ordinatemAs baseline values (tm, ƒ) of the segment of the fetal heart rate signal, respectivelym) (ii) a Obtaining all fetal heart rate segmentation baseline values and then performing baseline fitting; and (4) putting all the baseline values and the fetal heart rate curve in the same coordinate system for data fitting, and directly adopting an interpolation method for data fitting to obtain the baseline value of the fetal heart rate signal.
In addition, the invention can also make the fetal heart rate signal curve closer to the baseline part by using Empirical Mode Decomposition (EMD) processing, reduce the number of extreme points and be beneficial to distinguishing the baseline part from the non-baseline part by using the extreme points.
It is noted that other methods may be used by those skilled in the art to pre-process the acquired raw CTG signal data, or to extract the baseline point of the fetal heart rate signal to obtain the baseline value of the fetal heart rate signal.
The prenatal fetal monitoring intelligent interpretation model comprises an embedded layer, a splicing layer, a bidirectional gating circulation unit layer (namely a BiGRU layer) and a full connection layer. The embedding layer carries out oneHot processing on each signal point of a one-dimensional fetal heart rate signal, a uterine contraction pressure signal and a fetal movement signal segment with the signal length of d, multiplies the result by a two-dimensional matrix W, and transforms the signal into a 4 x d two-dimensional matrix, wherein W is a 4-row d-column two-dimensional matrix, d is the dimensionality of oneHot, and the formula is oneHot (t) x W, and t belongs to (1, d). And the splicing layer splices the output results of the embedding layers one by one to form a two-dimensional signal with a result of 12 multiplied by d.
In the bidirectional gated cyclic unit network with 2k GRU units in the bidirectional gated cyclic unit layer, the calculation formula at the layer is as follows:
and (4) updating the door: z is a radical oft=σ(W（z）xt+U（z）ht-1)
Resetting a gate: r ist=σ(W（z）xt+U（z）ht-1)
The current storage content is as follows: h is1´=tanh(Wxt+r⊙Uht-1)
Final memory (output) of current time step: h ist= zt⊙ht-1 +(1- zt ) ⊙ht´
It is worth to be noted that, in the bidirectional gating circulation unit layer, each GRU unit outputs a one-dimensional vector, and finally, the one-dimensional output vectors with the output length of 2k are combined and output, and then input into the fully-connected layer of the two units for calculation, and the calculation formula of the ith unit of the fully-connected layer is as follows: ƒci= Wi1* h1+ Wi2* h2+…Wik* hk+ Wik+1* h1´+ W ik+2* h2´+…Wi2k* hk´。hkDenotes the result of forward GRU, hk"indicates the result of reverse GRU. Then sigmoid compression is carried out on the two unit results, and probability sigmoid that the samples are normal is respectively output (ƒ)c1) And probability sigmoid that the sample is abnormal (ƒ)c2）。
Verification example 1
In order to verify the discrimination ability of the prenatal fetus, deep learning models of RNN, LSTM, GRU, BiRNN, BiLSTM, CNN and DNN are selected for comparative analysis, and the discrimination ability comparative analysis results of the deep learning models are shown in Table 1.
The Gated Recurrent Unit (GRU) described above is a variant of the Recurrent Neural Network (RNN). Long Short-Term Memory networks (LSTMs) are variants of RNNs. The Bidirectional Gated recycling Unit (BiGRU) related to the invention is formed by superposing two layers of GRUs, and the forward and reverse Bidirectional GRU units jointly determine the output at the same time.
And each deep learning model learns the same training set, then the same verification set is used for verifying and outputting a convergence curve of the model learning process, finally classification and judgment are carried out on the same test set, a label corresponding to the maximum value of the output probability is taken as an output result, and the label is divided into a normal class and an abnormal class. Verification example 1 the accuracy, precision, recall, specificity, F1 value, Kappa coefficient, MCC coefficient, AUC value and time of the above deep learning model of comparative analysis.
Accuracy, i.e. Accuracy (Accuracy), is the most common evaluation index in deep learning. Precision (Precision) indicates the proportion of the data of the positive class predicted to be correct. The Recall (Recall) represents the proportion of positive data that is predicted to be correct in the data that is actually positive. Specificity represents the predicted correct negative class data proportion of the data that is actually negative class. The positive class related to the accuracy, recall and specificity is the abnormal class of the invention, and the negative class is the normal class of the invention.
The F1 value (F1-score) represents a comprehensive index considering accuracy and recall, and the F1 value is more representative when the data are unbalanced. The Kappa coefficient indicates an index describing judgment of the consistency, and a larger value indicates a higher level of consistency. The MCC Coefficient (Matthews Correlation Coefficient) and the Mazis Correlation Coefficient are the most abundant key indexes for measuring the quality of the two-classifier model.
In addition, verification example 1 introduces Receiver Operating Characteristic Curve (ROC) to evaluate the performance of the deep learning model. To measure the ROC result, the Area of ROC was defined as AUC (Area Under dark) and ranged from [0,1 ]. When the value of AUC is larger, the classification effect of the model is better.
TABLE 1 comparative analysis results of discrimination ability of each deep learning model
The results in table 1 show that, compared with other deep learning models, the intelligent classification and discrimination performed by the bidirectional gating cycle unit (BiGRU) deep learning model in the embodiment 1 not only consumes short time, but also integrates various evaluation indexes, and the embodiment 1 has the optimal discrimination capability, and can quickly and effectively provide prenatal fetal monitoring aid decision support for obstetrical medical care personnel.
Verification example 2
In order to verify the influence of the data set difference to be classified on the discrimination capability of the intelligent interpretation method. Validation example 2 four control groups were set, control group a: the dataset to be classified contains only standard fetal heart rate signals (FHR); control group B: the data set to be classified only contains uterine contraction signals (UC); control group C: the data set to be classified contains fetal movement signals (FM) only; control group D: the dataset to be classified contains only the standard fetal heart rate and contraction combined signal (F-U). The remaining process steps for the four control groups remained the same as in example 1. Verification example 2 comparative analysis example 1 and four control groups showed accuracy, precision, recall, specificity, F1 values, Kappa coefficient, MCC coefficient, and AUC values, and the results of the comparative analysis are shown in table 2.
TABLE 2 comparative analysis of the Performance of four control groups compared with example 1
The results in table 2 show that, compared with the above four control groups, the data set to be classified in example 1 includes a standard fetal heart rate signal, a uterine contraction pressure signal, and a fetal movement signal, and various evaluation indexes are compared comprehensively, so that the intelligent interpretation model in example 1 has the optimal classification discrimination capability.
Verification example 3
In order to verify the influence of the synchronous segmentation processing of the fused multi-signal data sets through a sliding window on the judgment capability of the intelligent interpretation method. The fused multi-signal dataset for control group E was not processed by sliding window segmentation, and the remaining method steps remained the same as in example 1. Verification example 3 comparative analysis example 1 and control group E were analyzed for accuracy, precision, recall, specificity, F1 values, Kappa coefficient, MCC coefficient, and AUC values, with the comparative analysis results shown in table 3.
TABLE 3 comparative analysis results of control group B and example 1
The results in table 3 show that, compared with the control group E, the fused multi-signal dataset is synchronously segmented by the sliding window, so that the accuracy, the recall ratio, the F1 value, the Kappa coefficient, the MCC coefficient, and the AUC value of the intelligent interpretation model of embodiment 1 reach higher degrees, which indicates that the classification discrimination performance of the intelligent interpretation model of embodiment 1 is improved, the probability of misjudging the abnormal type sample as the normal type sample can be effectively reduced, and the irreversible damage to the health of the pregnant woman and the fetus due to missing of the treatment opportunity is avoided.
Verification example 4
In order to verify the influence of the standardization processing of the fetal heart rate signals on the judgment capability of the intelligent interpretation method. The control group F was pretreated without normalization and the remaining method steps were identical to those of example 1. Verification example 4 comparative analysis example 1 and control group F were tested for accuracy, precision, recall, specificity, F1 values, Kappa coefficient, MCC coefficient, and AUC values, and the results of the comparative analysis are shown in table 4.
TABLE 4 comparative analysis results of control group A and example 1
The results in table 4 show that, compared with the control group F, the normalization processing is performed on the original fetal heart rate signal after the preprocessing, so that the recall rate, the F1 value, the Kappa coefficient, the MCC coefficient, and the AUC value of the intelligent interpretation model in example 1 reach higher degrees, which indicates that the classification discrimination performance of the intelligent interpretation model in example 1 is improved, the probability of misjudging the abnormal type sample into the normal type sample can be effectively reduced, and the irreversible damage to the health of the pregnant woman and the fetus due to missing of the treatment opportunity can be avoided.
Verification example 5
The judgment capability of the prenatal fetal monitoring intelligent interpretation model based on the bidirectional gating circulation unit is verified by adopting the confusion matrix as follows:
table 5 confusion matrix for example 1
| Prediction/truth | Is normal | Abnormal state |
| Is normal | 84.36% | 13.85% |
| Abnormal state | 15.64% | 86.15% |
The results in table 5 show that the accuracy of the abnormal sample in example 1 is the highest, which reaches 86.15%, and the accuracy of the normal sample reaches 84.36%, and meanwhile, the probability of misjudging the real abnormal sample as the normal sample is 13.85%, so that the possibility of misjudging the normal sample as the abnormal sample is reduced, and unnecessary interference to the fetus caused by over-birth detection can be avoided.
In conclusion, the intelligent interpretation method of the invention integrates the uterine contraction pressure signal and the fetal movement signal into the prenatal fetal monitoring intelligent interpretation model based on the bidirectional gating circulation unit, compared with other deep learning models, the time consumption is shorter, the classification discrimination capability is better, unexpected technical effects are obtained, the auxiliary decision support of prenatal fetal monitoring can be quickly and effectively provided for obstetrical medical care personnel, the probability of misjudging the abnormal type samples into the normal type samples is effectively reduced, and the irreversible damage to the health of the pregnant woman and the fetus caused by missing of the treatment opportunity is avoided.
Moreover, the invention not only carries out preprocessing according to the data characteristics of the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal, but also carries out sliding window segmentation processing on the preprocessed standard fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal, thereby realizing data enhancement, unifying the signal length, further improving the classification discrimination performance of the prenatal fetal monitoring intelligent interpretation model based on the bidirectional gating circulation unit and obtaining unexpected technical effects.
The above description is only for the preferred embodiment of the present invention, but the present invention is not limited to the embodiment, and those skilled in the art can equally modify or replace the concept of the present invention within the scope of the present invention.
Claims (6)
1. An intelligent prenatal fetal heart monitoring signal interpretation method comprises the following steps:
s1: acquiring original CTG signal data comprising a fetal heart rate signal, a uterine contraction pressure signal and a fetal movement signal;
s2: preprocessing the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal to form a fusion multi-signal data set;
s3: carrying out segmentation processing on the fused multi-signal data set to obtain one-dimensional fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments with the signal lengths of d, and constructing a data set to be classified;
s4: inputting the data set to be classified into a pre-trained prenatal fetal monitoring intelligent interpretation model for classification and judgment, wherein the prenatal fetal monitoring intelligent interpretation model comprises an embedding layer, a splicing layer, a bidirectional gating circulation unit layer and a full connection layer;
respectively inputting the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal segment of which the signal lengths are d into the embedding layer to obtain the fetal heart rate signal, the uterine contraction pressure signal and the fetal movement signal segment of a two-dimensional matrix (dxm), wherein m is the output dimension of the embedding layer;
inputting fetal heart rate signals, uterine contraction pressure signals and fetal movement signal segments of the two-dimensional matrix (d x m) into a splicing layer for splicing to obtain a two-dimensional matrix (d x 3 m) vector;
inputting the two-dimensional matrix (d multiplied by 3 m) vector into the bidirectional gating circulation unit layers of 2k units for splicing to obtain a one-dimensional output vector with the length of 2 k;
inputting the one-dimensional output vector with the length of 2k to a full-connection layer of n units, and carrying out sigmoid function compression on an output result to output the probability that the sample is a normal class and the probability that the sample is an abnormal class;
respectively setting corresponding class labels for the probabilities that the sample is in the normal class and the sample is in the abnormal class;
and comparing the probability that the sample is a normal class with the probability that the sample is an abnormal class, and selecting the class label corresponding to the higher probability as a classification judgment result.
2. The intelligent prenatal fetal heart monitoring signal interpretation method as defined in claim 1, wherein: the preprocessing of the step S2 includes performing interpolation or deletion processing on the fetal heart rate signal, the uterine contraction pressure signal, and the fetal movement signal, respectively.
3. The prenatal fetal heart monitoring signal intelligent interpretation method according to claim 2, characterized in that: the preprocessing of step S2 further includes normalizing the interpolated or deleted fetal heart rate signal.
4. The prenatal fetal heart monitoring signal intelligent interpretation method according to claim 3, characterized in that: the normalization process includes calculating a baseline value of the fetal heart rate signal and subtracting the baseline value from the interpolated or deleted fetal heart rate signal to obtain a standard fetal heart rate signal.
5. The prenatal fetal heart monitoring signal intelligent interpretation method according to claim 3, characterized in that: the preprocessing of the step S2 further includes performing sliding window segmentation on the interpolated or deleted fetal heart rate signal before performing normalization processing on the interpolated or deleted fetal heart rate signal, so as to obtain a segment of the fetal heart rate signal with a signal length not less than p.
6. The intelligent prenatal fetal heart monitoring signal interpretation method as defined in claim 1, wherein: the segmentation processing of the step S3 comprises the step of synchronously segmenting the preprocessed fetal heart rate signals, the preprocessed uterine contraction pressure signals and the preprocessed fetal movement signals by a sliding window.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110798011.9A CN113456064B (en) | 2021-07-15 | 2021-07-15 | Intelligent interpretation method for prenatal fetal heart monitoring signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110798011.9A CN113456064B (en) | 2021-07-15 | 2021-07-15 | Intelligent interpretation method for prenatal fetal heart monitoring signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113456064A true CN113456064A (en) | 2021-10-01 |
| CN113456064B CN113456064B (en) | 2024-04-16 |
Family
ID=77880341
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110798011.9A Active CN113456064B (en) | 2021-07-15 | 2021-07-15 | Intelligent interpretation method for prenatal fetal heart monitoring signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113456064B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114159039A (en) * | 2021-11-12 | 2022-03-11 | 广州三瑞医疗器械有限公司 | Intelligent antenatal fetal heart monitoring model |
| CN114724720A (en) * | 2022-06-10 | 2022-07-08 | 北京大学第三医院（北京大学第三临床医学院） | Prenatal electronic fetal heart monitoring automatic identification system based on deep learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1993025971A1 (en) * | 1992-06-09 | 1993-12-23 | University Of Plymouth | Medical signal analyzer |
| CN102210586A (en) * | 2011-06-03 | 2011-10-12 | 泰安市迈迪医疗电子有限公司 | Automatic analysis method for fetus monitoring device |
| US20140378855A1 (en) * | 2013-06-25 | 2014-12-25 | The Research Foundation For The State University Of New York | Apparatus and method for feature extraction and classification of fetal heart rate |
| CN107997741A (en) * | 2018-01-08 | 2018-05-08 | 北京大学 | A kind of coupling analytical method of Fetal Heart Rate and uterine contraction signal |
| CN108836307A (en) * | 2018-05-14 | 2018-11-20 | 广东工业大学 | A kind of intelligent ECG detection device, equipment and mobile terminal |
| CN109567868A (en) * | 2018-10-15 | 2019-04-05 | 广东宝莱特医用科技股份有限公司 | A kind of CTG fetal rhythm methods of marking and system |
| CN112971799A (en) * | 2021-02-04 | 2021-06-18 | 北京理工大学 | Non-stimulation fetal heart monitoring classification method based on machine learning |
-
2021
- 2021-07-15 CN CN202110798011.9A patent/CN113456064B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1993025971A1 (en) * | 1992-06-09 | 1993-12-23 | University Of Plymouth | Medical signal analyzer |
| CN102210586A (en) * | 2011-06-03 | 2011-10-12 | 泰安市迈迪医疗电子有限公司 | Automatic analysis method for fetus monitoring device |
| US20140378855A1 (en) * | 2013-06-25 | 2014-12-25 | The Research Foundation For The State University Of New York | Apparatus and method for feature extraction and classification of fetal heart rate |
| CN107997741A (en) * | 2018-01-08 | 2018-05-08 | 北京大学 | A kind of coupling analytical method of Fetal Heart Rate and uterine contraction signal |
| CN108836307A (en) * | 2018-05-14 | 2018-11-20 | 广东工业大学 | A kind of intelligent ECG detection device, equipment and mobile terminal |
| CN109567868A (en) * | 2018-10-15 | 2019-04-05 | 广东宝莱特医用科技股份有限公司 | A kind of CTG fetal rhythm methods of marking and system |
| CN112971799A (en) * | 2021-02-04 | 2021-06-18 | 北京理工大学 | Non-stimulation fetal heart monitoring classification method based on machine learning |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114159039A (en) * | 2021-11-12 | 2022-03-11 | 广州三瑞医疗器械有限公司 | Intelligent antenatal fetal heart monitoring model |
| CN114724720A (en) * | 2022-06-10 | 2022-07-08 | 北京大学第三医院（北京大学第三临床医学院） | Prenatal electronic fetal heart monitoring automatic identification system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113456064B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110188836B (en) | Brain function network classification method based on variational self-encoder | |
| CN113456064A (en) | Intelligent interpretation method for prenatal fetal heart monitoring signal | |
| CN111738363B (en) | Alzheimer disease classification method based on improved 3D CNN network | |
| CN112446891A (en) | Medical image segmentation method based on U-Net network brain glioma | |
| CN112426160A (en) | Electrocardiosignal type identification method and device | |
| CN113274031A (en) | Arrhythmia classification method based on deep convolution residual error network | |
| CN113657449A (en) | Traditional Chinese medicine tongue picture greasy classification method containing noise labeling data | |
| CN112529886A (en) | Attention DenseUNet-based MRI glioma segmentation method | |
| CN114224288B (en) | Microcapsule neural network training method and equipment for detecting epileptic brain electrical signals | |
| CN111696670A (en) | Intelligent prenatal fetus monitoring interpretation method based on deep forest | |
| CN114037001A (en) | Mechanical pump small sample fault diagnosis method based on WGAN-GP-C and metric learning | |
| US6941288B2 (en) | Online learning method in a decision system | |
| CN113610147A (en) | Multi-potential subspace information fusion earthquake short-term prediction method based on LSTM | |
| CN112863650A (en) | Cardiomyopathy identification system based on convolution and long-short term memory neural network | |
| CN117132849A (en) | Cerebral apoplexy hemorrhage transformation prediction method based on CT flat-scan image and graph neural network | |
| CN115115038B (en) | Model construction method based on single lead electrocardiosignal and gender identification method | |
| CN116543154A (en) | Medical image segmentation method based on multi-level semantic features | |
| CN111582440A (en) | Data processing method based on deep learning | |
| Yang et al. | Unsupervised clustering and analysis of contraction-dependent fetal heart rate segments | |
| CN114159039A (en) | Intelligent antenatal fetal heart monitoring model | |
| CN113080847B (en) | Device for diagnosing mild cognitive impairment based on bidirectional long-short term memory model of graph | |
| CN114723937A (en) | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image | |
| Mendis et al. | The Effect of Fetal Heart Rate Segment Selection on Deep Learning Models for Fetal Compromise Detection | |
| CN115644927A (en) | Multi-center fetal heart monitoring intelligent interpretation method based on maximum and minimum entropy semi-supervised field self-adaption | |
| CN115618284A (en) | Multi-center fetal heart monitoring intelligent interpretation method based on unsupervised field self-adaption |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |