Multimodal clinical data placental implantation assessment system

Abstract

Disclosed is a placenta implantation sign evaluation device, comprising a collection unit, which collects and pre-processes a plurality of blood flow indexes and inputs a screening unit and collects a plurality of clinical indexes and inputs a fusion unit; the screening unit extracts effective blood flow indexes for placenta implantation sign typing from the plurality of blood flow indexes by using a first mode and / or a second mode, and inputs the effective blood flow indexes into the fusion unit; the fusion unit inputs the effective blood flow indexes and the plurality of clinical indexes into respective evaluators, and fuses placenta implantation sign typing results output by the evaluators to obtain placenta implantation sign evaluation results. Through the placenta implantation sign evaluation device, a plurality of blood flow indexes for accurately evaluating placenta implantation ultrasonic signs can be obtained, and the blood flow indexes are fused with clinical information to evaluate the development status of placenta implantation of a subject at different gestational stages.

Description

Technical Field

[0001] This application belongs to the field of placenta accreta detection technology, specifically, it relates to a multimodal clinical data placenta accreta assessment system. Background Technology

[0002] Placenta accreta is a pathological condition in which placental villi abnormally invade the myometrium or even penetrate the serosa. It can lead to fatal postpartum hemorrhage and infection, requiring premature termination of pregnancy to control bleeding, which increases the risk of complications in premature infants. Furthermore, placenta accreta can cause insufficient placental blood supply, restricting normal fetal growth. Placenta accreta and its related pregnancy symptoms can alter blood flow within the uterus and placenta, such as decreased endothelial cell count and impaired function in peripheral blood and umbilical cord blood in patients with preeclampsia, significantly higher uterine artery hemodynamic parameters than in normal pregnancies, and the use of peripheral vascular resistance, pulse wave, and other blood flow characteristics to predict gestational hypertension, as described in CN102058435B.

[0003] Previously, due to limitations in detection methods, the detection rate of indicators such as uterine blood flow was low. With the development of ultrasound imaging technology, techniques such as color Doppler energy imaging and three-dimensional ultrasound have been widely used for endometrial blood flow detection. The advent of three-dimensional ultrasound has further effectively compensated for the shortcomings of two-dimensional ultrasound. It can synthesize and analyze the image information of the coronal plane of the placenta through computer technology to provide a three-dimensional image, thereby effectively measuring the volumetric blood flow parameters of the endometrium, such as the vascular index (VI), flow index (FI), and vascular flow index (VFI), which are difficult to obtain by traditional detection methods. These blood flow parameters have the potential to analyze the development of placental implantation.

[0004] However, there is still a lack of methods to analyze or detect placenta accreta using hemodynamic parameters. Currently, the diagnosis of adhesive, accreta, and penetrating placenta accreta is still mainly based on manual or intelligent analysis of ultrasound images. This requires more human or computing resources, has higher requirements for image data, and the stability of the assessment may be difficult to guarantee.

[0005] Therefore, there is a need for an intelligent system that uses hemodynamic parameters to assess and classify placenta accreta. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this application is to provide a placenta accreta sign assessment device, which can acquire multiple blood flow indicators of the placenta from ultrasound images using an acquisition unit, extract effective blood flow indicators from blood flow indicators of multiple different vessels at different time periods of pregnancy that can be used to effectively distinguish between different types of placenta accreta populations and healthy populations using a screening unit, and combine with clinical parameters of other modalities to predict multiple signs of placenta accreta, thereby obtaining the assessment results of placenta accreta signs.

[0007] Specifically, this application relates to the following aspects:

[0008] According to one aspect of this application, a placenta accreta sign assessment device is provided, comprising: a collection unit for collecting and preprocessing multiple blood flow indicators and multiple clinical indicators, inputting the preprocessed multiple blood flow indicators into a screening unit, and inputting the multiple clinical indicators into a fusion unit; a screening unit for extracting effective blood flow indicators for placenta accreta sign classification from the multiple blood flow indicators using a first mode and / or a second mode, and inputting the effective blood flow indicators into the fusion unit; and a fusion unit for inputting the effective blood flow indicators and multiple clinical indicators into their respective evaluators, and fusing the placenta accreta sign classification results output by the evaluators to obtain a placenta accreta sign assessment result.

[0009] According to some embodiments of this application, the first mode includes: inputting multiple blood flow indicators into a classifier; adjusting the number and / or type of blood flow indicators input into the classifier in response to the classifier's evaluation index value not reaching a preset standard; and setting the currently input blood flow indicator into the classifier as a valid blood flow indicator in response to the classifier's evaluation index value reaching a preset standard.

[0010] According to some embodiments of this application, the first mode includes: inputting multiple blood flow indicators into a classifier, sorting the weight values ​​of the multiple blood flow indicators for placental implantation signs from low to high; sequentially removing blood flow indicators, retraining the classifier and sorting the multiple blood flow indicators from low to high based on the new weight values, and continuing the steps of classifier training, weight value sorting, and sequential removal of blood flow indicators in response to an increase in the classifier evaluation index value until the classifier evaluation index value no longer increases; in response to a decrease in the classifier evaluation index value, reintroducing the removed blood flow indicators and sequentially removing the next blood flow indicator, and continuing the steps of classifier training, weight value sorting, and evaluation index value comparison; and finally retaining the blood flow indicators as valid blood flow indicators.

[0011] According to some embodiments of this application, the second mode includes: performing intergroup analysis of multiple blood flow indicators for placenta accreta phenotype classification over multiple time intervals, to determine the effective blood flow indicator among the multiple blood flow indicators that show significant differences between patients with placenta accreta phenotype and healthy individuals.

[0012] According to some embodiments of this application, obtaining a placenta accreta assessment result by fusing the placenta accreta sign classification results output by the evaluator includes: inputting multiple effective blood flow indicators and multiple clinical indicators into a first evaluator and a second evaluator respectively to obtain a first classification result and a second classification result respectively; assigning a first weight and a second weight to the first classification result and the second classification result respectively according to the evaluation indicator values ​​of the first evaluator and the second evaluator, so as to obtain a placenta accreta sign assessment result by weighting the first classification result and the second classification result.

[0013] According to some embodiments of this application, obtaining placenta accreta typing results from the fusion evaluator includes: inputting multiple effective blood flow indicators and multiple clinical indicators into a first evaluator and a second evaluator respectively to obtain a first typing result and a second typing result respectively; inputting the first typing result and the second typing result into a third evaluator to train the third evaluator so that its evaluation indicator values ​​reach preset standards, and using the third typing result output by the third evaluator as the placenta accreta typing assessment result.

[0014] According to some embodiments of this application, signs of placenta accreta include: placental location, placental thickness, posterior placental hypoechoic band, vesicular line, placental fossa, blood flow at the base of the placenta, cervical sinuses and / or cervical morphology.

[0015] According to some embodiments of this application, multiple blood flow parameters include vascular index, blood flow index, vascular blood flow index, resistance index, pulsatility index and / or the ratio of end-systolic peak value to end-diastolic peak value, collected in seven time intervals during pregnancy: 11-16 weeks, 16-20 weeks, 20-24 weeks, 24-28 weeks, 28-32 weeks, 32-36 weeks and 38 weeks and above.

[0016] According to some embodiments of this application, the resistance index, pulsatility index, and end-systolic peak to end-diastolic peak ratio include the resistance index, pulsatility index, and end-systolic peak to end-diastolic peak ratio of the umbilical artery, the right uterine artery, and the left uterine artery, respectively.

[0017] According to another aspect of this application, a multimodal placenta accreta assessment system is provided, comprising: a sign prediction unit, which collects and fuses multiple clinical and blood flow indicators of the subject, predicts the subtype of each placenta accreta sign among the multiple placenta accreta signs of the subject based on the fused indicators, and inputs the prediction results into the assessment unit; and an assessment unit, which assigns a third weight to each placenta accreta sign among the multiple placenta accreta signs, receives the prediction results to weightedly fuse the multiple placenta accreta signs, and obtains the subject's placenta accreta assessment result.

[0018] According to some embodiments of this application, multiple blood flow parameters include: the left uterine artery pulsatility index and the ratio of the left uterine artery end-systolic peak value to the end-diastolic peak value during 11-16 weeks of gestation; the ratio of the umbilical artery end-systolic peak value to the end-diastolic peak value during 16-20 weeks of gestation; the left uterine artery pulsatility index and the ratio of the left uterine artery end-systolic peak value to the end-diastolic peak value during 24-28 weeks of gestation; the change in the umbilical artery resistance index between 16-38 weeks of gestation; the change in the ratio of the left uterine artery end-systolic peak value to the end-diastolic peak value between 11-32 weeks of gestation; and the changes in the endometrial volume vascular index and vascular blood flow index between 11-38 weeks of gestation.

[0019] According to some embodiments of this application, multiple clinical indicators include: age, parity, medical history information, and / or gestational week of delivery.

[0020] According to some embodiments of this application, the sign prediction unit: uses the acquisition unit of the aforementioned placenta accreta sign assessment device to acquire multiple clinical indicators and multiple blood flow indicators of the subject; and uses the fusion unit of the aforementioned placenta accreta sign assessment device to fuse multiple clinical indicators and multiple blood flow indicators, and uses an evaluator to predict the subtype of each placenta accreta sign among multiple placenta accreta signs based on the fused indicators.

[0021] The placenta accreta assessment device provided in this application can extract sensitive indicators of placenta accreta from various blood flow parameters of patients with known placenta accreta results. The extracted blood flow parameters and other clinical examination results are used to train the placenta accreta assessment device. After obtaining effective blood flow parameters and the assessment device, the multimodal placenta accreta assessment system provided in this application predicts the classification of various placenta accreta signs based on the input data provided by the new subjects. The placenta accreta result of the subject is calculated according to the importance of each placenta accreta sign to the assessment of the severity of placenta accreta. Thus, the assessment result of placenta accreta can be obtained automatically without in-depth analysis of ultrasound images, providing a stable and rapid early screening and assessment solution for placenta accreta from a new perspective. Attached Figure Description

[0022] Figure 1 The figure shows a schematic diagram of the structure of a placenta accreta sign assessment device according to an embodiment of this application.

[0023] Figure 2 The illustration shows a flowchart for screening effective blood flow indicators according to an embodiment of this application.

[0024] Figure 3 The illustration shows a first schematic diagram of blood flow index values ​​for different populations according to an embodiment of this application.

[0025] Figure 4The illustration shows a second schematic diagram of blood flow index values ​​for different populations according to an embodiment of this application.

[0026] Figure 5 The illustration shows a third schematic diagram of blood flow index values ​​for different populations according to embodiments of this application.

[0027] Figure 6 The illustration shows a fourth schematic diagram of blood flow index values ​​for different populations according to embodiments of this application. Detailed Implementation

[0028] The present application is further illustrated below with reference to embodiments. It should be understood that the embodiments are only used to further illustrate and explain the present application and are not intended to limit the present application.

[0029] Unless otherwise defined, technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art. While similar or identical methods and materials may be applied in experimental or practical applications, materials and methods are described herein. In case of conflict, the definitions included herein shall prevail. Furthermore, materials, methods, and examples are for illustrative purposes only and are not intended to be limiting. The present application is further described below with reference to specific embodiments, but is not intended to limit the scope of the application.

[0030] definition

[0031] Signs of placenta accreta

[0032] Placental accretion can affect the location or structure of placenta on imaging or pathology, including placental location, placental thickness, shape of the hypoechoic band behind the placenta, shape of the bladder line, placental indentation, blood flow pattern at the base of the placental junction, cervical sinuses, and cervical morphology. The detection and classification of these signs are important bases for judging the pathological characteristics and development of placental accretion without direct intraoperative visualization. Based on the classification of different signs, i.e., the severity, the individual's type of placental accretion can be comprehensively assessed to distinguish whether it is a healthy state or a mild early adhesion type, or a more dangerous implantation or penetrating type.

[0033] Application Overview

[0034] As mentioned above, although many hemodynamic parameters have been used for the clinical detection or early screening of diseases during pregnancy, such as gestational hypertension, there is still a lack of methods or systems for interpreting signs of placenta accreta. Specifically, current ultrasound assessments of placenta accreta include blood flow-related parameters, but these are generally subjective, lacking consistency and stability, and are only used to assist in the interpretation of ultrasound images. Since current ultra-microscopic blood flow imaging technology can visually display the three-dimensional distribution of blood vessels in the placental implantation area, the extent of placental implantation, and its blood perfusion, and sensitively capture low-velocity perforating vessels within the implanted placenta, it can more realistically show the blood perfusion in the placental implantation area and accurately distinguish the boundary between the myometrium and the implanted placenta. Therefore, using blood flow parameters for intelligent assessment of placenta accreta is entirely feasible.

[0035] Based on the above considerations, this application provides a placenta accreta sign assessment device and a multimodal placenta accreta assessment system based on the device. The placenta accreta sign assessment device uses its acquisition unit to collect and preprocess multiple blood flow indicators from ultrasound images and multiple clinical indicators from medical records. The multiple blood flow indicators can reflect the severity of placenta accreta signs, while the multiple clinical indicators are supplementary information for placenta accreta classification from additional modalities. A screening unit is used to further obtain the preprocessed multiple blood flow indicators. The multiple clinical indicators are directly input into a fusion unit. After the effective classification indicators among the multiple blood flow indicators are extracted by the screening unit, the fusion unit supplements the general semantic features that may be lost in the blood flow parameters with easily obtainable clinical indicators, thereby improving the accuracy of placenta accreta classification.

[0036] In the screening unit, this application can use two different screening modes to extract effective blood flow indicators for placental implantation sign classification from the multiple blood flow indicators. The two modes can be used to screen some blood flow indicators that are suitable for the mode and then merge them to obtain redundant indicator information to improve the robustness of effective blood flow features. Alternatively, the two modes can be used to screen the same indicators and then take the intersection to obtain high-sensitivity blood flow indicators. After implementing different schemes, the extracted effective blood flow indicators are input into the fusion unit.

[0037] In the fusion unit, the effective blood flow indicators and multiple clinical indicators are input into their respective evaluators. The placental accretion sign classification results output by the evaluators are then weighted and fused at the decision layer to obtain the placental accretion sign assessment result. This effectively solves the problems of low stability and loss of key features in the assessment of signs from single-modality blood flow data. The multimodal placental accretion assessment system directly utilizes the extracted effective blood flow indicators to fuse multiple clinical indicators to obtain classification results for multiple placental accretion signs. Then, the weights of each placental accretion sign in assessing the severity of placental accretion are set and adjusted, finally yielding a placental accretion assessment result based on the classification results of multiple placental accretion signs. These methods are more objective and stable than traditional manual interpretation of placental accretion, and have lower data acquisition, processing, and training costs compared to image segmentation and classification schemes, thus possessing high practicality.

[0038] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0039] Exemplary System

[0040] Figure 1 The illustration shows a block diagram of a placenta accreta sign assessment device according to an embodiment of this application.

[0041] refer to Figure 1 The placenta accreta assessment device provided in this application includes the following components.

[0042] The acquisition unit 1 is used to acquire and preprocess multiple blood flow indicators and multiple clinical indicators. The preprocessed blood flow indicators are input into the screening unit, and the multiple clinical indicators are input into the fusion unit.

[0043] The acquisition unit of the device described in this application uses different extraction methods for different types of blood flow indicators. First, it needs to be clarified that the blood flow indicators selected for classifying placenta accreta include two main categories: volumetric blood flow parameters of the endometrium and specific vascular blood flow parameters obtained through blood flow monitoring and analysis. For different categories of blood flow indicators, the acquisition unit provides customized acquisition schemes:

[0044] For endometrial volume blood flow parameters, these include the aforementioned vascular index (VI), blood flow index (FI), and vascular blood flow index (VFI). VI reflects the density of blood vessel distribution within the region of interest (ROI) of the ultrasound image, such as the endometrium; its value is the ratio of the area of ​​colored pixels in the Doppler image to the area of ​​the ROI. FI represents blood flow intensity, and its value is the mean Doppler blood flow energy signal. VFI is a weighted fusion of the above two indicators. For clinical energy Doppler ultrasound equipment, the acquisition unit acquires the image from the equipment, performs binarization processing, integrates the blood flow signal and blood flow energy, and obtains the signal area (VI) and energy magnitude (FI). For color Doppler ultrasound images, the proportion of colored pixels (VI) and the mean signal intensity (FI) of the endometrial region can be directly calculated. For contrast-enhanced ultrasound, the area under the contrast agent's TIC curve (VI) and peak intensity (FI) are calculated by integration. For other commercially available clinical 3D ultrasound equipment, most of their accompanying software can directly output these three indicator values. This is how some of these blood flow indicators are obtained.

[0045] Blood flow parameters obtained by blood flow monitoring and analysis methods may include the Resistance Index (RI), which reflects the ratio of the change in blood flow velocity during the cardiac cycle to the peak velocity at end-systole. Increased peripheral circulatory impedance will accelerate the rate of decrease in RI, thus reflecting the relative increase in peripheral impedance due to poor cardiac function or deeper placental implantation. The Pulsatility Index (PI) represents the mean of the maximum frequency shift of the velocity waveform (peak at end-systole - peak at end-diastole) within a cardiac cycle, which reflects the ratio of the decrease in diastolic blood flow velocity to the average velocity, thus reflecting the vascular condition from another perspective. The ratio of peak at end-systole (S) to peak at end-diastole (D) (S / D) is the amplitude of the change in blood flow velocity during the cardiac cycle, which can reflect placental blood flow function. It is easy to see that the above three blood flow parameters are all based on the S and D parameters. In this application, these three parameters are located in the three vascular areas that will undergo significant changes during the development of placental implantation: the umbilical artery, the left uterine artery, and the right uterine artery. The S, D and blood flow velocity values ​​throughout the entire cardiac cycle are obtained by an ultrasound Doppler probe (or alternatively, arterial puncture) and sent to the acquisition unit for calculation of the corresponding parameters.

[0046] It is important to note that the manifestations of placenta accreta and the extent to which the aforementioned indicators reflect the degree of placenta accreta can vary across different time periods of pregnancy. Using the average values ​​across all time periods or early pregnancy values ​​instead of collecting indicators at different time points may not yield satisfactory results and could overlook the unique characteristics of placenta accreta at different time stages. Therefore, the blood flow parameters described in this application are collected separately at multiple different time periods throughout pregnancy. For example, if an individual pregnancy is divided into n time periods for collection, the collection unit will ultimately collect and calculate 12n blood flow indicators, including VI, FI, VFI, RI, PI, and S / D of the umbilical artery, RI, PI, and S / D of the right uterine artery, and RI, PI, and S / D of the left uterine artery within each of the n time periods. The screening unit will effectively extract the blood flow parameters for each time period to obtain the effective blood flow indicators during the critical window period for placenta accreta classification.

[0047] The optimal timeframe for collecting blood flow parameters is primarily considered to be after 10 weeks of pregnancy (i.e., the fetal period, before which the placenta is not fully established). Between 10 and 15 weeks, chorionic villus branching increases, the placenta thickens, and maternal-fetal circulation is established. After 15 weeks, placental ultrasound markers become clearly visible. After 20 weeks, uterine artery resistance decreases, and placental perfusion increases. After 28 weeks, the aforementioned blood flow parameters can clearly reflect placental structural aging or functional abnormalities. Therefore, it is preferable to divide the pregnancy into multiple time intervals of 5 weeks for collecting blood flow parameters, ensuring that the placenta differs morphologically and functionally within each time interval, and guaranteeing that the blood flow parameters cover different stages of placental development to reflect various possible developments of placenta implantation.

[0048] The data extracted by the acquisition unit from the clinical ultrasound equipment can be in DICOM format, or a dynamic sequence format or 3D format of DICOM, or video or image format data from Doppler or contrast imaging. By establishing a direct or indirect data transmission interface with the ultrasound equipment (e.g., typically USB, HDMI, or various data acquisition and storage cards such as TF cards or USB flash drives), relevant parameters in the DICOM or image frames, such as blood flow signal intensity, energy intensity, S-value, and D-value, are received and calculated to obtain multiple blood flow indicators required by the screening unit. Furthermore, the multiple clinical indicators acquired by the acquisition unit may include basic indicators highly correlated with placental diseases, such as the individual's age, parity, medical history, and / or gestational age at delivery, from which the blood flow parameters are derived. These indicators are usually easily obtained, and the acquisition unit can integrate existing technologies (e.g., basic networks of language models such as BioBERT and ClinicalBERT, or the TF-IDF algorithm) to directly extract word information of these indicators from individual medical records.

[0049] Before screening multiple blood flow indicators, the acquisition unit can perform various preprocessing operations, including outlier removal, missing value imputation, normalization, and labeling, to ensure the data quality of multiple blood flow indicators. Before inputting multiple clinical indicators into the fusion unit, the acquisition unit also needs to perform various preprocessing operations on these unstructured texts, including normalization, text cleaning to remove symbols and stop words, word segmentation correction, word embedding to generate word vectors, hot encoding, and labeling, so that these numerical or non-numerical text indicators can be directly recognized and processed by the evaluator, which will be described in detail below.

[0050] In other words, the signs of placenta accreta include: placental location, placental thickness, hypoechoic band behind the placenta, bladder line, placental fossa, blood flow at the base of the placenta, cervical sinuses and / or cervical morphology.

[0051] Furthermore, the multiple blood flow indicators include: vascular index, blood flow index, vascular blood flow index, resistance index, pulsatility index, and / or the ratio of end-systolic peak value to end-diastolic peak value, collected at seven time intervals during pregnancy: 11-16 weeks, 16-20 weeks, 20-24 weeks, 24-28 weeks, 28-32 weeks, 32-36 weeks, and 38 weeks and above. The resistance index, the pulsatility index, and the ratio of end-systolic peak value to end-diastolic peak value include the resistance index, pulsatility index, and ratio of end-systolic peak value to end-diastolic peak value for the umbilical artery, the right uterine artery, and the left uterine artery, respectively.

[0052] The placenta accreta assessment device also includes a screening unit 2, the process of which screens effective blood flow indicators as follows: Figure 2As shown, "Other" represents the collection of corresponding indicator data from patients with conditions different from placenta accreta, such as gestational diabetes mellitus (GDM), fetal growth restriction (TGR), and gestational hypertension (PIH). Effective blood flow indicators for placenta accreta sign classification are extracted from these multiple blood flow indicators using a first mode and / or (preferably either or, but with small sample sizes, the effective blood flow indicators obtained by taking the intersection are easily insufficient) a second mode. These effective blood flow indicators are then input into the fusion unit. Statistical methods based on inter-group difference analysis are well-suited for differentiating clinical medical diagnostic indicators. They strictly adhere to hypothesis testing, effectively avoid incorrect classifications being rejected, and are well-suited for small sample analyses. Machine learning methods, on the other hand, are more suitable for situations with a large number of features and nonlinear features. As can be seen from the above, the blood flow indicators output by the acquisition unit to the screening unit in this application can be 12n. For example, when the interval is n=7, the blood flow indicators will reach 84. In this case, using a machine learning classifier to extract effective blood flow indicators is a feasible solution. In addition, many classifiers (such as logistic regression and support vector machine) are interpretable and can also help medical personnel understand why the screened blood flow indicators can effectively classify placental implantation signs.

[0053] Therefore, the first and second modes can respectively employ machine learning classifiers and statistical analysis to process and extract effective indicators from the numerous blood flow indicators described in this application. Only blood flow indicators common to both modes can be retained, reducing the likelihood of false positive indicators being selected. This results in effective blood flow indicators with strong evidentiary support, ensuring accuracy in placental implantation classification within a limited range. Alternatively, all blood flow indicators selected by both modes can be retained to ensure the integrity of blood flow information and avoid missing weakly correlated blood flow features. Furthermore, effective blood flow features obtained from these two modes can be fused to balance the rigor and flexibility of the blood flow indicators. For example, the normalized feature importance values ​​(such as the feature weight coefficients of SVM) and significance values ​​(such as the significance level p) of the effective blood flow indicators selected by the first and second modes can be used as weight values ​​for the effective blood flow indicators, thus weighting and fusing these effective blood flow indicators.

[0054] Specifically, in one example of this application, the first mode employs a machine learning algorithm to train a model using blood flow indicators collected from the 11th week of pregnancy to fit the data and identify the classification patterns of indicators for placenta accreta signs. Multiple blood flow indicators are calculated and preprocessed in the acquisition unit before being input into the screening unit. Each of the multiple blood flow indicators has two labels: asymptomatic or with placenta accreta signs. An SVM classifier is configured in the screening unit, which includes a kernel trick to map the raw data to a high-dimensional space, possessing the ability to handle nonlinear features and exhibiting high training efficiency when the sample size is small. Since the nonlinearity between blood flow parameters and placenta accreta signs is strong, three endometrial volume blood flow parameters (VI, FI, and VFI) from different gestational periods, or parameters (RI, PI, and S / D) from three blood flow monitoring analysis methods, or all six parameters can be selected as the blood flow indicators to be screened and input into the SVM. The kernel function of SVM can be a Gaussian kernel (RBF) to deal with the complex nonlinear relationships between blood flow indicators, especially the nonlinear relationships at the classification boundary, or a polynomial kernel can be selected to set the degree of nonlinearity to determine the optimal hyperplane shape for separating blood flow indicators of two labels.

[0055] Then, the selected blood flow indicators are input into the SVM classifier, and the number of blood flow indicators is optimized to achieve the preset classification performance of the SVM, such as its accuracy (reflecting the overall classification performance of signs and asymptomatic cases), recall (reflecting the ability to identify signs), and precision (avoiding misclassification of asymptomatic cases as signs) reaching preset values. Multiple blood flow indicators in the training set can be split evenly, and each split indicator can be used to train the classifier and test its performance. The performance under different combinations of indicator quantity and / or type (such as indicators of endometrial volume and indicators of blood flow monitoring analysis) is recorded, and the optimal combination of blood flow indicators that meets or does not meet the target is obtained by comparison and set as the effective blood flow indicators. Effective blood flow indicators obtained in this way can avoid overfitting caused by the classifier overemphasizing a portion of the data, and can also ensure a certain level of classifier fitting efficiency when inputting small samples in medical scenarios.

[0056] Specifically, a recursive approach can be used to progressively eliminate blood flow indicators with low importance in distinguishing placenta accreta. The contribution can be the feature weight coefficients in an SVM classifier. After each classifier training iteration, the weight coefficients for each blood flow indicator are obtained, and the blood flow indicator with the lowest weight coefficient is removed. The classifier is retrained, and the weight coefficients are obtained repeatedly. Based on the new classifier's performance in classifying placenta accreta, it's determined whether to continue removing the blood flow indicator with the lowest weight coefficient. For example, if the classifier's performance improved after removing the blood flow indicator with the lowest weight coefficient in the previous iteration, then removing the blood flow indicator with the lowest weight coefficient in the current training iteration can be considered. If the classifier's performance decreased instead of improved after removing the blood flow indicator with the lowest weight coefficient, then the depth relationship between the removed indicator and other indicators needs to be considered. It's possible to re-include the indicator and remove the blood flow indicators ranked after it, then train the classifier to re-verify the changes in classification performance.

[0057] In this way, the classifier performance will not continue to improve after removing any remaining blood flow parameters, thus confirming that the remaining blood flow parameters are valid. Specifically, in this first mode, a minimum number of valid blood flow parameters is preferably set to avoid overfitting or sudden underfitting after a single deletion due to excessive removal of parameters. In one example, a classifier (RBF-SVM) constructed using the first mode for three endometrial volume blood flow parameters was found to have similar classification weights for VI, FI, and VFI, with no significant difference in predictive performance for placenta accreta signs, and the classifier achieved an 89% prediction rate on the test set. However, in this example, the number of positive samples (i.e., samples diagnosed with placenta accreta) in the training and test sets was too small, potentially leading to overfitting due to data sparsity. Therefore, a second mode is needed to further filter valid blood flow parameters from these small sample data.

[0058] That is, the first mode includes: inputting the multiple blood flow indicators into the classifier; in response to the classifier's evaluation index value not reaching the preset standard, adjusting the number and / or type of blood flow indicators input into the classifier; and in response to the classifier's evaluation index value reaching the preset standard, setting the blood flow indicator currently input into the classifier as the effective blood flow indicator.

[0059] The first mode includes: inputting the plurality of blood flow indicators into a classifier, sorting the placental implantation sign subtyping weight values ​​of the plurality of blood flow indicators from low to high; sequentially removing blood flow indicators, retraining the classifier and sorting the plurality of blood flow indicators from low to high based on the new weight values, and continuing the steps of classifier training, weight value sorting, and sequential removal of blood flow indicators in response to an increase in the classifier evaluation index value until the classifier evaluation index value no longer increases; and reintroducing the removed blood flow indicators in response to a decrease in the classifier evaluation index value and sequentially removing the next blood flow indicator, while continuing the steps of classifier training, weight value sorting, and evaluation index value comparison; and using the last retained blood flow indicator as the effective blood flow indicator.

[0060] In another example of this application, the second mode employs statistical intergroup analysis to screen effective blood flow indicators. This involves calculating the mean and variance of blood flow indicators from similar samples to observe the overall relationship between the indicators and symptoms. This mode is more stable and interpretable when the sample size of blood flow indicators is small. For example, intergroup analysis can be performed on the three indicators VI, FI, VFI, and RI and PI (S / D) of the umbilical artery, right uterine artery, and left uterine artery within different gestational time intervals.

[0061] Specifically, the mean and variance of similar blood flow parameters were calculated for each pregnancy period. The means were plotted as a bar chart, and the variance range was represented by line segments for comparison. The results for RI, PI, and S / D indices of the umbilical artery are as follows: Figure 3 As shown, the mean values ​​of RI, PI, and S / D indices of the umbilical artery all decreased with increasing pregnancy duration, and the variance also decreased. Among these, the RI index showed relatively small changes and exhibited abrupt variance changes, possibly due to subjects adjusting ultrasound equipment parameters during individual data collection. In the PI index data of the umbilical artery, the mean value decreased more significantly for those with signs of placenta accreta, while the decrease was less pronounced for asymptomatic cases. The S / D index of the umbilical artery showed the most significant changes, with the mean value decreasing the most for those with signs of placenta accreta, and the magnitude of the decrease being significantly different from that of the asymptomatic case.

[0062] Blood flow parameters of the right uterine artery include the right uterine artery RI value, right uterine artery PI value, and right uterine artery S / D value. The sample categories remain the same: placenta accreta and asymptomatic. The blood flow parameter data of the right uterine artery are shown in the figure below. Figure 4 As shown, the mean and variance of blood flow parameters in the right uterine artery gradually decreased over time, with the S / D ratio of the right uterine artery showing the largest decrease. Over time, the mean values ​​of all data related to the right uterine artery in symptom-related parameters remained consistently lower than the mean values ​​for normal pregnant women, and gradually decreased over time, but the rate of decrease was not significantly different.

[0063] The observation indicators for the left uterine artery include the RI value, PI value, and S / D value of the left uterine artery. The classification of the indicators is as described above. The blood flow index data of the left uterine artery are shown in the figure below. Figure 5 As shown, the decreasing trend of the left uterine artery data is similar to that of the right uterine artery data. However, it is worth noting that the S / D value of the left uterine artery in the symptom-related samples was significantly lower than that in the asymptomatic samples in the early stages of pregnancy (11-16 weeks), but in the late stages of pregnancy (28 weeks and beyond), it was basically consistent with that in the asymptomatic samples, and the magnitude of the change was smaller than that in the asymptomatic samples.

[0064] Finally, the data graphs for the three endometrial volume blood flow parameters are as follows: Figure 6 As shown, the decrease in symptom samples and VFI throughout the entire pregnancy was significantly lower than that in asymptomatic samples. In addition, for each time interval, except for 11-16 weeks and 28-32 weeks, the mean values ​​of the two types of samples were quite similar, which made the data differences small and the distinction difficult.

[0065] As shown above, the changes in blood flow parameters throughout pregnancy can be used to classify placenta accreta and asymptomatic symptoms. Effective blood flow parameters include the umbilical artery RI (reduction index) between 16 and 38 weeks of gestation, the left uterine artery S / D (short-to-long-distance ratio) between 11 and 32 weeks of gestation, and the changes in endometrial volume VI (volume index) and VFI (volume index fibrillation index) between 11 and 38 weeks of gestation. However, in reality, not all subjects can provide blood flow parameter changes over such a long time span, especially for early-stage pregnant women in their early gestation, where the aforementioned effective blood flow parameters are difficult or impossible to obtain effectively. Therefore, finding effective blood flow parameters within each gestational period is a more realistic approach, covering more subjects; that is, using the blood flow parameters within each gestational period as the classification for placenta accreta and asymptomatic symptoms. In this case, the left uterine artery PI and S / D ratio at 11-16 weeks of gestation; the umbilical artery S / D ratio at 16-20 weeks of gestation; and the left uterine artery PI and S / D ratio at 24-28 weeks of gestation can be selected as effective blood flow indicators. However, VI, FI, and VFI are difficult to be used as effective blood flow indicators in a single interval.

[0066] That is, the second mode includes: performing intergroup analysis on the multiple blood flow indicators to classify placenta accreta signs across multiple time intervals, so as to determine the effective blood flow indicator among the multiple blood flow indicators that shows a significant difference between patients with placenta accreta signs and healthy individuals.

[0067] The placenta accreta assessment device further includes a fusion unit 3, which inputs the effective blood flow index and the multiple clinical indices into their respective evaluators, and fuses the placenta accreta classification results output by the evaluators to obtain the placenta accreta assessment result. A weighted average method can be used to fuse the evaluators trained separately using the effective blood flow index and multiple processed clinical indices. The weight values ​​of each evaluator can be set based on the performance indicators of each classifier on its validation set, such as assigning weight values ​​according to the normalized values ​​of indicators such as accuracy, recall, precision, or AUC of each classifier, and then weighted and fused using the following formula:

[0068]

[0069] Where P(y|x) represents the predicted probability of placenta accreta and asymptomatic signs for a sample under multiple evaluators. The category with the highest probability is selected as the placenta accreta classification result for that sample.

[0070]

[0071] When using an SVM classifier as an evaluator, it can be subjected to equal-division regression to output classification probability values, and then weighted fusion can be performed; when using logistic regression as an evaluator, it can be directly weighted fusion based on its output probability values.

[0072] Furthermore, multimodal metric fusion can be achieved by training a new evaluator using evaluators of different modalities. Specifically, the outputs of evaluators trained on two modalities (preferably the probability values ​​of the classification results) are concatenated to form a completely new metric to train a new evaluator.

[0073] φ(x)=[P1(y|x),P2(y|x)]

[0074] (Formula 3)

[0075]

[0076] Specifically, the preferred process for concatenating the outputs of the evaluators trained on the two modalities involves splitting all effective blood flow indicators and clinical indicators into multiple groups and training their respective evaluators through cross-training. The probability values ​​output by the evaluators are then concatenated into a new indicator matrix φ(x). This avoids overfitting of the evaluators for both effective blood flow indicators and clinical indicators. It can be understood that the evaluator for effective blood flow features can directly use the machine learning classifier used for screening in the first mode. This machine learning classifier has already undergone cross-validation in the first mode and can be directly used as an evaluator to generate placental implantation classification probability values.

[0077] That is, the process of integrating the placenta accreta sign classification results output by the evaluator to obtain the placenta accreta sign assessment result includes: inputting the multiple effective blood flow indicators and the multiple clinical indicators into the first evaluator and the second evaluator respectively to obtain the first classification result and the second classification result respectively; assigning a first weight and a second weight to the first classification result and the second classification result according to the evaluation indicator values ​​of the first evaluator and the second evaluator respectively, so as to weight the first classification result and the second classification result to obtain the placenta accreta sign assessment result.

[0078] Furthermore, the process of integrating the placenta accreta sign typing results output by the evaluator to obtain the placenta accreta sign typing results includes: inputting the plurality of effective blood flow indicators and the plurality of clinical indicators into the first evaluator and the second evaluator respectively to obtain the first typing result and the second typing result respectively; inputting the first typing result and the second typing result into the third evaluator to train the third evaluator so that its evaluation indicator values ​​reach the preset standard, and the third typing result output by the third evaluator as the placenta accreta sign evaluation result.

[0079] In this way, the results of placental implantation sign classification, effective blood flow indicators required for classification, and combinations of clinical indicators fused with effective blood flow indicators are obtained through the above units.

[0080] Based on these results, this application further provides a multimodal placenta accreta assessment system, which includes a sign prediction unit and an assessment unit. The sign prediction unit is used to collect multiple clinical indicators and multiple blood flow indicators of the subject and fuse the multiple clinical indicators and multiple blood flow indicators. Based on the fused indicators, it predicts the classification of each placenta accreta sign among the multiple placenta accreta signs of the subject, and inputs the prediction results into the assessment unit. It is easy to understand that the sign prediction unit is the placenta accreta sign assessment device described above that has determined the required indicators. It may not include the screening unit of the placenta accreta sign assessment device, but it may also include a screening unit to periodically and dynamically update the effective blood flow indicators, realize the dynamic updating of the entire system, and automatically generate an automatic and accurate classification of the placenta accreta sign by processing the input effective blood flow indicators and clinical indicators.

[0081] The evaluation unit is used to assign a third weight to each of the multiple placenta accreta signs, receive the prediction results, and weight and fuse the multiple placenta accreta signs to obtain the subject's placenta accreta assessment result. It is understood that different placenta accreta signs have different degrees of severity of impact on the subject's health. The multiple placenta accreta signs output by the sign prediction unit can be used as new input features to train a new classifier. The severity of the subject's placenta accreta can be used as the classification label to obtain the weight coefficients of different placenta accreta signs. These weight coefficients are then used to weight the multiple placenta accreta signs, and the weighted result is used as the final placenta accreta severity assessment result.

[0082] In one example, current clinical assessments of placenta accreta severity are obtained by equally weighted summation of all signs of placenta accreta, meaning that all signs are considered to have a consistent or nearly consistent ability to reflect the severity of placenta accreta; therefore, all third weights can be set to 1. However, with the optimization of existing clinical diagnostic criteria for placenta accreta, this application can use the aforementioned assessment unit to adjust the third weights of different signs of placenta accreta in a timely manner to provide subjects with more accurate placenta accreta assessment results.

[0083] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0084] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0085] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0086] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0087] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A multimodal placenta accreta assessment system, including: The acquisition unit acquires and preprocesses multiple blood flow indicators and multiple clinical indicators, inputs the preprocessed multiple blood flow indicators into the screening unit, and inputs the multiple clinical indicators into the fusion unit; The screening unit extracts effective blood flow indicators for placenta accretion sign classification from the multiple blood flow indicators using the first mode and the second mode, and inputs the effective blood flow indicators into the fusion unit. The fusion unit inputs the effective blood flow indicators and the multiple clinical indicators into their respective evaluators, and fuses the placenta accretion sign classification results output by the evaluators to obtain the placenta accretion sign assessment results. This includes inputting the multiple effective blood flow indicators and the multiple clinical indicators into the first evaluator and the second evaluator respectively, and obtaining the first classification result and the second classification result respectively. The first and second classification results are input into the third evaluator to train the third evaluator so that its evaluation index values ​​reach the preset standard. The third classification result output by the third evaluator is used as the evaluation result of the placenta accreta sign. The sign prediction unit collects the subject's multiple clinical indicators and multiple blood flow indicators and uses the fusion unit to fuse the multiple clinical indicators and multiple blood flow indicators. Based on the fused indicators, the third evaluator predicts the subtype of each placenta accretion sign among the subject's multiple placenta accretion signs and inputs the prediction results into the evaluation unit. The evaluation unit assigns a weight to each of the multiple placenta accretion signs, receives the prediction results to weightedly fuse the multiple placenta accretion signs, and obtains the subject's placenta accretion evaluation result. The multiple blood flow indicators include vascular index, blood flow index, vascular blood flow index, resistance index, pulsatility index, and / or the ratio of end-systolic peak value to end-diastolic peak value collected in seven time intervals during pregnancy: 11-16 weeks, 16-20 weeks, 20-24 weeks, 24-28 weeks, 28-32 weeks, 32-36 weeks, and over 38 weeks. The resistance index, the pulsatility index, and the ratio of end-systolic peak value to end-diastolic peak value include the resistance index, pulsatility index, and ratio of end-systolic peak value to end-diastolic peak value of the umbilical artery, the right uterine artery, and the left uterine artery, respectively.

2. The multimodal placenta implantation assessment system according to claim 1, wherein, The first mode includes: The multiple blood flow indicators are input into the classifier. In response to the classifier's evaluation index value not reaching the preset standard, the number and / or type of blood flow indicators input into the classifier are adjusted. In response to the classifier's evaluation index value reaching the preset standard, the blood flow indicator currently input into the classifier is set as the effective blood flow indicator.

3. The multimodal placenta implantation assessment system according to claim 1, wherein, The first mode includes: The multiple blood flow indicators are input into the classifier, and the weight values ​​of the placental implantation signs of the multiple blood flow indicators are sorted from low to high. The blood flow indicators are removed sequentially, the classifier is retrained, and the multiple blood flow indicators are sorted from low to high based on the new weight values. In response to the increase of the classifier evaluation index value, the steps of classifier training, weight value sorting, and sequential removal of blood flow indicators are continued until the classifier evaluation index value no longer increases. In response to the decrease of the classifier evaluation index value, the removed blood flow indicators are reintroduced and the next blood flow indicator is removed sequentially. The steps of classifier training, weight value sorting, and evaluation index value comparison are continued. The last retained blood flow index is taken as the effective blood flow index.

4. The multimodal placenta accreta assessment system according to claim 1, wherein, The second mode includes: Intergroup analysis was performed on the multiple blood flow indicators for placenta accreta genotyping across multiple time intervals to determine the effective blood flow indicators that showed significant differences between patients with placenta accreta genotyping and healthy individuals.

5. The multimodal placenta accreta assessment system according to any one of claims 1-4, wherein, The signs of placenta accreta include: placental location, placental thickness, hypoechoic band behind the placenta, bladder line, placental fossa, blood flow at the base of the placenta, cervical sinuses and / or cervical morphology.

6. The multimodal placenta implantation assessment system according to claim 1, wherein, The multiple clinical indicators include: Age, parity, medical history and / or gestational week of delivery.

Images