Method and device for estimating gestational age
By using trained machine learning models, using ultrasound images to calculate the fetus' gestational age and its confidence evaluation, the problems of long scanning time and low efficiency in evaluation characteristics in the prior art are solved, and fast and accurate gestational age estimation and evaluation of multiple fetal characteristics are achieved.
Patent Information
- Application Number
- CN202380056251.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-07-28
- Filing Date
- 2023-07-26
- Publication Date
- 2025-04-18
AI Technical Summary
The prior art requires strict imaging criteria in estimating fetal gestational age, resulting in long scan times and lack of efficient methods in evaluating other key fetal characteristics.
Using trained machine learning models, fetal gestational age and its confidence assessment is calculated by ultrasound images and evaluated without strict imaging criteria, including assessment of multiple fetal characteristics.
The rapid and accurate estimate of fetal gestational age and other key fetal characteristics are achieved without the need for strict imaging standards, significantly reducing scanning time.
Smart Images
Figure CN119855552A_ABST
Abstract
Description
[0001] The present invention relates to methods and apparatus for estimating the gestational age of a fetus and optionally for assessing some other key aspects of pregnancy assessment.
[0002] The gestational age (GA) of the fetus is defined as the time elapsed since the start of the last menstrual period (LMP). GA is known to have many clinical applications, including due date estimation and growth monitoring. Although in some cases GA can be estimated by the mother's recollection of the date of her last menstrual period, this is often problematic, and therefore it is common practice to estimate GA based on ultrasound data.
[0003] Therefore, an estimate of fetal gestational age is a key output of routine obstetric scans and is obtained by taking certain measurements of the fetus. This is currently done in early pregnancy by measuring the crown-rump length (CRL), while later in pregnancy it is estimated based on a number of measurements, typically including head circumference, femur length and abdominal circumference.
[0004] Measuring the CRL is therefore a key component of current routine first trimester scans. This requires capturing images in the correct imaging plane, which is defined by certain properties well known to trained sonographers and defined internationally in various imaging standards, the key imaging standards being:
[0005] The top of the fetus's head and buttocks are visible
[0006] The fetal neck is in neutral flexion
[0007] The fetus is in a horizontal position relative to the ultrasound beam
[0008] In practice, obtaining such images is time consuming, with much time wasted waiting for the fetus to be in a suitable position for measurement, and a more time-efficient method of GA estimation is clinically desirable. It has been estimated that approximately one third of the total scan time in the first trimester is spent obtaining suitable CRL images.
[0009] During fetal screening, in addition to the assessment of GA, a number of other key clinical assessments are performed. While the details of these assessments vary across regions and scanning institutions, the following are typically assessed: the presence of multiple fetuses, and in this case, the chorionicity and amnioticity of these fetuses; fetal viability by confirming cardiac activity and estimating fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus having trisomy; estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; the presence of markers of genetic abnormalities such as increased nuchal fold thickness or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech, cephalic, etc.); placental location; the type of placental appearance; and an estimate of the amount of amniotic fluid.
[0010] Therefore, improved methods for assessing these fetal characteristics are also desired.
[0011] Embodiments of the present invention are directed to methods for obtaining reliable GA estimates throughout pregnancy that do not require as stringent imaging criteria as required in previously considered methods. Additional embodiments include assessment of other key fetal characteristics including, for example, those listed above.
[0012] The present invention is defined in the appended independent claims, to which reference should now be made. Further preferred features can be found in the appended dependent claims.
[0013] According to one aspect of the invention, a method for estimating the gestational age of a fetus is provided, the method comprising: obtaining at least one ultrasound image of at least a portion of the fetus; and calculating an estimate of the gestational age of the fetus and a corresponding confidence assessment for the estimate based on the image.
[0014] The method preferably comprises calculating the gestational age of the fetus and the associated confidence interval from any 2D ultrasound image of the fetus.This is in contrast to previously considered methods which require 2D images acquired from very specific planes, or in other cases a 3D volume.
[0015] The method may include calculating the gestational age of the fetus based on ultrasound images obtained at any stage during pregnancy.
[0016] In a preferred arrangement, the method does not require an ultrasound machine operator to view the or each ultrasound image.
[0017] Preferably, the method comprises using a trained machine learning model to calculate an estimate of gestational age and / or a corresponding confidence assessment for the estimate.
[0018] Preferably, the machine learning model is generated by training on a representative set of ultrasound (US) images of a fetus, for each representative ultrasound image in the representative ultrasound image set, the GA at the time of imaging is known. Model training is preferably accomplished via a process known as supervised learning (which will be well known to those skilled in the art), whereby the model is configured to achieve the strongest possible (while also robust and generalizable) association between each image and its corresponding GA value. In the present invention, the model and supervised training method are preferably additionally configured to output a range of values in which the GA is expected to lie. Preferably, the probability that a known GA value falls outside of this range is low.
[0019] Preferably, the supervised machine learning process adopts the form of deep learning, wherein the model is an artificial neural network. The supervised learning method may include optimizing the parameters of the network via stochastic gradient descent to minimize the loss function. Preferably, the loss function should be constructed so that for regression tasks (including, for example, GA and fetal heart rate estimation), the loss function produces improved performance in terms of both the accuracy of the estimate and the possibility that the supplied range contains the true value. Now, by way of example, below is a class of loss functions with this property. This description is formulated in accordance with GA predictions, but the technician will recognize that it is also operational for other regression tasks.
[0020] instruct:
[0021] x = ultrasound image of at least a portion of the fetus
[0022] y = fetal GA
[0023] = Two outputs of the neural network, representing the predicted GA and the width of the confidence interval around it
[0024] A class of functions with the desired properties is described by
[0025] L=L reg +αL conf
[0026] in:
[0027] L reg Is applied to (or alternatively if the prediction task is performed in logarithmic space ) any non-decreasing function of any semi-norm
[0028] L conf is any function with the following properties:
[0029] For some non-decreasing functions And for applications (or ) of some non-decreasing function of the semi-norm, and for some constant
[0030]
[0031] α is the algorithm hyperparameter
[0032] In the preferred configuration, the loss function is defined as:
[0033]
[0034] in:
[0035] |x| is used to represent the absolute value of x
[0036] f + represents the softplus function f + (x) = ln(1 + e x )
[0037] α is the algorithm hyperparameter
[0038] For classification tasks (including, for example, prediction of: chorionicity and amnioticity; sex of the fetus; whether the pregnancy is ectopic; presence of fetal activity; risk of trisomy in the fetus; presence of pathology such as ancrania or gastroschisis; fetal presentation; placental position), it is also desirable that the model should both generate predictions and quantify the uncertainty inherent in the predictions. For such tasks, preferably, the loss function should be constructed to produce improved performance in terms of both the accuracy of the estimate and the quantification of the uncertainty inherent in the estimate. A loss function of this nature is now described, by way of example. The description is formulated in terms of predicting whether the pregnancy is single or multiple, but the skilled person will recognize that it is readily operational for other classification tasks.
[0039] instruct:
[0040] x = ultrasound of at least a portion of the contents of the uterus containing one or more fetuses
[0041] image
[0042] y = one-hot encoded variable indicating singleton or multiple pregnancy
[0043] K (in this case = 2) is the number of categories in the dependent variable
[0044] The output of the neural network, representing the concentration parameter of the Dirichlet distribution
[0045] Γ(·) is the well-known gamma function
[0046] is the digamma function
[0047] The loss (for a single example) is defined as:
[0048] L=L class +αL conf
[0049] in:
[0050]
[0051] And α is an algorithm hyperparameter.
[0052] Of course, the loss is summed over all examples in the data set. The skilled person will recognize that many other loss formulas can be used to achieve the same goal, including, for example, formulas based on mean squared error or Bayesian risk and formulas based on distributions other than Dirichlet.
[0053] After training such a model, one can obtain point estimates of class probabilities from it by computing, for example, the expectation of the resulting Dirichlet distribution:
[0054]
[0055] The predicted class can be obtained by identifying the class with the highest class probability:
[0056]
[0057] The uncertainty inherent in the predicted class can be estimated, for example, by calculating the variance associated with the predicted class:
[0058]
[0059] Preferably, the architecture of the neural network is a convolutional neural network, a visual transformer or a variant thereof. Preferably, the Adam optimizer will be used to perform stochastic gradient descent.
[0060] In a preferred arrangement, the method includes determining at least one relative dimension of one or more anatomical features of the fetus.
[0061] The method preferably includes determining a confidence value for the estimate of gestational age. The confidence value may include an expression of confidence that the estimated fetal gestational age is accurate. The method may include determining a confidence value for the estimate of gestational age expressed as an age range, wherein the narrower the range, the higher the confidence value.
[0062] The method may include obtaining a plurality of ultrasound images of the fetus. In a preferred configuration, the method includes determining GA estimates and corresponding confidence ranges by applying a machine learning model to the plurality of ultrasound images, and then applying numerical processing to the plurality of outputs obtained (including but not limited to averaging the outputs).
[0063] In a preferred arrangement, one or more of the following steps is performed by an electronic processor: determining relative size, calculating a gestational age estimate, and determining a confidence value.
[0064] Preferably, the step of determining at least one relative dimension within the image comprises determining a proportion of a portion of the image relative to at least one other portion of the same image. Alternatively or additionally, the step of determining at least one relative dimension within the image may comprise determining a proportion of a portion of the image relative to at least one portion of at least one other image.
[0065] The image (or each of the images) may include an image of at least one anatomical feature of the fetus.
[0066] The method may include generating a plurality of estimates, more preferably a plurality of estimates with a plurality of corresponding confidence values, and filtering the plurality of estimates to select only one or more estimates that meet a predetermined confidence value threshold. The method may include generating a plurality of estimates, more preferably a plurality of estimates with a plurality of corresponding confidence values, and ranking the estimates based on the confidence value.
[0067] The method may include instructing the operator to obtain one or more specific images of the fetus. In one example, the operator may be instructed from the outset to obtain images of certain regions of the fetus that are known (whether by reasoning derived from clinical expertise or via data analysis) to produce an accurate estimate of the GA. In another example, the operator may be dynamically instructed to obtain additional images while scanning if the images already acquired have a wide confidence range associated with them.
[0068] The method may include arithmetically processing, such as averaging, the estimate of gestational age obtained from the plurality of images.
[0069] The image may be selected by an operator during scanning.Alternatively or additionally, the image may comprise one or more sample frames automatically captured / obtained during a scanning operation.
[0070] The method may include a method of estimating biometric measurements including, for example, one or more of: crown-to-rump length (CRL), head circumference (HC), femur length (FL), abdominal circumference (AC), and transverse cerebellar diameter (TCD).
[0071] The present invention also includes a method for performing one or more clinical assessments on a fetus, the method comprising: obtaining an ultrasound image of a fetus; and processing the image using a trained machine learning model to determine one or more of the following items, including but not limited to: the presence of multiple fetuses, and in this case, the chorionicity and amnioticity of these fetuses; fetal vitality by confirming cardiac activity and estimating the fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus suffering from trisomy; the estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; markers of genetic abnormalities such as increased thickness of the nuchal fold or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, head presentation, etc.); placental location; category of placental appearance; and estimation of amniotic fluid volume.
[0072] According to another aspect of the present invention, there is provided an apparatus for estimating the gestational age of a fetus, the apparatus comprising: an ultrasound image capture device for obtaining an image of at least a portion of a fetus; and an electronic processing device configured to process the image and calculate an estimate of the gestational age of the fetus and a corresponding confidence assessment for the estimate.
[0073] Preferably, the processing means is configured to use a trained machine learning model to calculate the estimate and / or confidence assessment.
[0074] The device is preferably configured to capture and process a plurality of ultrasound images of the fetus.In a preferred configuration, the device is configured to determine at least one relative size of one or more anatomical features of the fetus.
[0075] The apparatus is preferably configured to determine a confidence value for the or each estimate of gestational age. The confidence value may comprise an expression of confidence that the estimated fetal gestational age is accurate. The apparatus is preferably configured to determine, in use, a confidence value for an estimate of gestational age expressed as an age range, wherein the narrower the range, the higher the confidence value.
[0076] The device may be configured to provide real-time feedback to the user on the confidence range of the GA estimates that have been obtained, so that the user can direct the scan to a more appropriate image if necessary.
[0077] The processing may be implemented directly on the ultrasound device or on separate hardware that receives the video feed from the ultrasound device and optionally displays the results on a separate monitor.
[0078] The image (or each of the images) may include an image of at least one anatomical feature of the fetus.
[0079] The present invention also includes a device for performing one or more clinical assessments on a fetus, the device comprising: a processor for processing an image using a trained machine learning model to determine one or more of the following items, including but not limited to: the presence of multiple fetuses, and in this case the chorionicity and amnioticity of these fetuses; fetal vitality by confirming cardiac activity and estimating the fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus suffering from trisomy; the estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; markers of genetic abnormalities such as increased thickness of the nuchal fold or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, head presentation, etc.); placental position; category of placental appearance; and estimation of amniotic fluid volume.
[0080] In another aspect, the present invention provides a computer program product on a computer-readable medium, comprising instructions that, when executed by a computer, cause the computer to perform a method for estimating the gestational age of a fetus or for performing one or more clinical assessments of the fetus, the method being in accordance with any statement herein.
[0081] According to any statement herein, the present invention also includes a program for causing an apparatus to perform a method for estimating the gestational age of a fetus or for performing one or more clinical assessments on a fetus.
[0082] The present invention may include any combination of features or limitations mentioned herein, except such combinations of features that are mutually exclusive or mutually inconsistent.
[0083] Preferred embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:
[0084] Figure 1 schematically illustrates an ultrasound image of a human fetus captured according to an embodiment of the present invention;
[0085] Figure 2 Schematic diagram showing the superimposition of the size calculation Figure 1 images;
[0086] Figure 3 Schematically illustrates a method for training a model for use with embodiments of the present invention;
[0087] Figure 4 Schematically shows an overview of a method according to an embodiment of the present invention;
[0088] Figure 5 A first embodiment of the method according to the invention is schematically shown;
[0089] Figure 6 A second embodiment of the method according to the invention is schematically shown;
[0090] Figure 7 A first embodiment of the device according to the invention is shown;
[0091] Figure 8 A second embodiment of the device according to the invention is shown; and
[0092] Fig. 9 Further embodiments of the present invention are shown.
[0093] In a developing fetus, different structures will develop at different rates and will change throughout pregnancy. In embodiments of the invention, the way in which various structures change over time - optionally the way in which different structures change relative to each other - can be used to estimate the age of the fetus using a trained model.
[0094] The current standard method for estimating GA is to obtain biometric measurements of the fetus and use the correlation between these biometric measurements and GA published in the clinical literature. These measurements must be obtained from imaging planes with specific properties, and obtaining such images and obtaining accurate measurements requires a high skill level and extensive training of the operator. If the process of accurately estimating GA can be greatly de-technicalized so that it can be performed by a wider population after minimal training, it will be very beneficial to both healthcare providers and patients. For example, this will enable assessments to be performed in a primary care setting. It would be further preferred if such a process could be performed without the operator having to view the ultrasound image at all. This is particularly preferred in cases where it is not desired to reveal the sex of the fetus (or this is illegal - as in India).
[0095] The required operator skill level can be reduced somewhat by using a 3D ultrasound probe, which, at least in principle, allows the burden of plane finding and measurement to be shifted to some extent from the operator to the algorithm. This is not a complete transfer, as acquiring a suitable 3D volume still requires a higher level of skill than the average person has. Moreover, in practice, the plane finding and measurement problem is very difficult. All methods we know of are either not as accurate as manual biometric measurements, or are impractical to use during routine scanning - in which case near real-time output is required - or both. Additionally, 3D ultrasound probes (and the type of ultrasound machines that can work with them) are significantly more expensive than 2D probes. Therefore, it is desirable to have a method for estimating a GA that takes only 2D ultrasound images as its input.
[0096] Figure 1 A portion of an ultrasound image I of a fetus F obtained according to an embodiment of the present invention is schematically shown.
[0097] Figure 2 Schematically illustrated are exemplary dimensions D1 in an image of a fetus F, which are determined as the image I is processed and represent distances between selected points. Using one or more such dimensions and particularly taking into account their size relative to the overall image, the processor uses a machine learning model that has been trained on a previously acquired dataset to calculate an estimate of the gestational age (GA) of the fetus and an expression of confidence in the prediction.
[0098] In some implementations, embodiments of the invention use deep learning techniques to automatically generate reliable estimates of gestational age directly from ultrasound images rather than via index biometric measurements. The method and associated apparatus are preferably capable of generating accurate estimates from a wide range of images without carefully acquiring specific images of the fetus or images with special properties or even without an operator having to view the images. Although the input ultrasound images do not contain information about the real-world scale of the anatomical structures present therein, they do contain information about the relative scale of their different regions, which can provide an understanding of fetal growth without requiring information about the real-world size of the pixels in the image.
[0099] The trained machine learning model is designed to produce an estimate of gestational age from any image of the fetus, without requiring images with certain properties related to the position of the fetus and the presence of certain anatomical features. Since some images presented to the model may not be suitable for this purpose, the model also reports the confidence of the prediction it made, which can be in the form of a prediction interval or range that optionally contains the actual age with the confidence expressed as a percentage. The prediction interval is dynamic and is based on the suitability of the input image for the task. On good or very suitable images, the interval is narrow, while on bad images, the interval is wide. The model predictions can then be filtered to include only those model predictions that determine the gestational age date with sufficiently high accuracy.
[0100] In some embodiments of the invention, the sonographer / operator may be guided to obtain certain images of the fetus that are known to be highly predictive of gestational age (e.g., a mid-sagittal view of the entire fetus, an axial section of the head), and combine these images as input to the model to improve its predictive accuracy.
[0101] The model output may be averaged over many images (e.g., a set selected by the sonographer during a scan and / or a sample of frames acquired in real time during a scan), where doing so has been found to improve the accuracy of predictions.
[0102] Although the main purpose of the present invention is to estimate gestational age, its embodiments may also be able to estimate biometric measurements (CRL, head circumference, etc.) that are currently commonly used as the basis for GA estimation and for growth monitoring.
[0103] It is a further object of the present invention to additionally output other predictions related to key aspects of the clinical assessment of early pregnancy, including but not limited to: the presence of multiple fetuses, and in this case the chorionicity and amnioticity of these fetuses; demonstration of fetal vitality by confirming cardiac activity and estimating fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus having trisomy; estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; the presence of markers of genetic abnormalities such as increased thickness of the nuchal folds or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, cephalic presentation, etc.); placental location; the category of placental appearance; and an estimate of the amount of amniotic fluid.
[0104] Depending on the output of the analysis of multiple video frames, small adjustments to the model configuration may be required, as described below. Representation vectors obtained preferably from the layer immediately preceding the prediction head of the model should be buffered for each frame. The prediction head should be adjusted to take as input the concatenation of some subset of these buffered representation vectors. Preferably, these adjustments are made not only during model training but also in the application of the model.
[0105] Embodiments of the present invention are primarily directed to accurately estimating gestational age during the first trimester of pregnancy without requiring the sonographer to obtain images of the CRL in perfect position, with the expected result being a significant reduction in the time required to complete the scan (e.g., approximately 7 minutes per scan). However, GA estimates for later stages of pregnancy can also be generated.
[0106] Go to Figure 3 , Figure 3 A method for training a machine learning model suitable for use with embodiments of the present invention is generally shown at 1000. At step 1010, a batch of gestational age (GA) data is supplied to a processor, which applies a loss function operation at step 1020. At step 1030, the loss gradient is calculated with respect to the selected model parameters. At step 1050, the data is fed into the machine learning model along with the corresponding batch of ultrasound images obtained at step 1040. Then, at step 1060, a machine learning model prediction (i.e., a machine learning model prediction of gestational age) is obtained, and then a cycle of n iterations is repeated, wherein the parameters are updated at each cycle. Finally, at step 1070, the trained machine learning model is completed.
[0107] Figure 4 An overview of the model inference process is shown generally at 2000. At step 2010, an ultrasound image is obtained. At step 2020, the image is supplied to the trained machine learning model. At steps 2030 and 2040, which may be substantially simultaneous, a GA prediction value and a corresponding GA confidence interval value are output.
[0108] Figure 5 A first embodiment of a method according to the present invention is generally shown at 3000. At step 3010, an ultrasound image is acquired. At step 3020, the image is fed into the model, and at step 3030, a predicted GA value and an associated confidence interval are calculated. At step 3040, it is determined whether the confidence interval is narrower than the value of the best confidence interval obtained. If not, at step 3050, the ultrasound operation is optionally directed to acquire an ultrasound image of a specific plane in the fetus. On the other hand, if it is determined at step 3040 that the value is narrower than the best value obtained, the predicted value and confidence interval are displayed at step 3060, and the register for the value of the best confidence interval is updated at step 3070.
[0109] Figure 6 A second embodiment of a method according to the present invention is generally shown at 4000. At step 4010, image acquisition instructions are provided to an ultrasound operator. Next, at step 4020, a number of ultrasound images are captured by the operator. Each ultrasound image is provided to a trained model at step 4030, and a predicted GA value and associated confidence interval are obtained at step 4040. The values are aggregated at step 4050, and the aggregated value is output at step 4060.
[0110] Figure 7 is a schematic representation of a first embodiment of an apparatus according to the invention. The apparatus is generally shown at 5000. An ultrasound scanning machine 5010 has an image receiving unit 5020 which receives acquired ultrasound images. The images are processed by a model in a processor 5030 and the GA predictions and confidence intervals are supplied to a GUI pipeline 5040. A display 5050 of the machine 5010 then displays the predictions and confidence intervals.
[0111] Figure 8A second embodiment of an apparatus according to the invention is schematically shown generally at 6000. In this embodiment, an ultrasound scanning machine 6010 acquires ultrasound images which are then supplied, for example via an HDMI or other link, to an external processing unit 6020. The external unit 6020 receives the images, processes them via the model, and outputs a GA estimate and confidence interval which are then presented on a separate display 6030 of the external unit.
[0112] Fig. 9 A further embodiment of the invention is shown generally at 7000 in which an ultrasound scanning machine 7010 acquires an ultrasound image which is then supplied via a camera interface C of a user device 7020, in this case a handheld device such as a cellular telephone. The user device 7020 is optionally capable of displaying the or each ultrasound image on its display 7030. The processor of the device 7020 performs the following tasks using software such as may be provided in an application: receiving the or each ultrasound image 7040; processing the or each image 7050; and predicting a GA and confidence calculation 7060.
[0113] Given that the above embodiments use the example of gestational age estimation, the skilled person will understand that, as an alternative or in addition, there may be different types of assessments, such as (but not limited to): the presence of multiple fetuses, and in this case the chorionicity and amnioticity of these fetuses; demonstration of fetal vitality by confirming cardiac activity and estimating the fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus having trisomy; estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; the presence of markers of genetic abnormalities such as increased thickness of the nuchal folds or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, head presentation, etc.); placental position; the category of placental appearance; and an estimate of the amount of amniotic fluid.
[0114] While in the foregoing description attention has been paid to those features of the present invention regarded as particularly important, it should be understood that the applicants claim protection for any patentable feature or combination of features mentioned herein and / or shown in the accompanying drawings, whether or not particular emphasis has been placed thereon.
Claims
1. A method for estimating the gestational age of a fetus, the method comprising: obtaining at least one ultrasound image of at least a portion of the fetus; and calculating a gestational age estimate for the fetus and a corresponding confidence assessment for the estimate.
2. The method according to claim 1, wherein: The gestational age estimate and / or the confidence assessment are generated by a machine learning model.
3. The method according to claim 2, wherein: The machine learning model is generated by training a representative set of ultrasound images of the fetus, and for each representative ultrasound image in the representative ultrasound image set, the gestational age at the time of imaging is known.
4. The method according to claim 2 or 3, wherein: Model training is done via supervised learning, whereby the model is configured to achieve the strongest possible (while also robust and generalizable) association between each image and its corresponding gestational age value.
5. The method according to claim 4, wherein: The model and / or supervised training method is additionally configured to output a range of values within which the gestational age is expected to lie.
6. The method according to claim 4 or 5, wherein: The supervised machine learning process takes the form of deep learning, where the model is an artificial neural network.
7. The method according to claim 6, wherein: Supervised learning methods involve optimizing the parameters of the network via stochastic gradient descent to minimize a loss function.
8. The method according to claim 7, wherein: The loss function is constructed to produce improved performance on the accuracy of the predictions and the corresponding confidence estimates.
9. The method according to claim 6, wherein: The architecture of the neural network is a convolutional neural network, a visual transformer or a variation thereof.
10. A method according to any one of the preceding claims, wherein: The method includes determining at least one relative dimension of one or more anatomical features of the fetus.
11. A method according to any one of the preceding claims, wherein: The method includes generating a plurality of estimates having a plurality of corresponding confidence values and filtering the plurality of estimates to select only one or more estimates that satisfy a predetermined confidence value threshold.
12. A method according to any one of the preceding claims, wherein: The method includes generating a plurality of estimates having a plurality of corresponding confidence values; and ranking the estimates based on the confidence values.
13. A method according to any one of the preceding claims, wherein: The method includes instructing an operator to obtain one or more specific images of the fetus.
14. The method according to claim 13, wherein: The operator is instructed to acquire images of certain regions of the fetus that are known (either by reasoning derived from clinical expertise or via data analysis) to produce an accurate estimate of gestational age.
15. The method according to claim 13 or 14, wherein: In the event that images already acquired have wide confidence ranges associated with those images, the operator is dynamically instructed to acquire additional images while scanning.
16. A method according to any one of the preceding claims, wherein: The method comprises arithmetically processing, such as averaging, the estimate of gestational age obtained from a plurality of images.
17. A method according to any one of the preceding claims, wherein: The method comprises a method of estimating biometric measurements including one or more of a crown-to-rump length, CRL, and a head circumference.
18. A method according to any one of the preceding claims, wherein: The method includes calculating a gestational age estimate and a confidence value based on any ultrasound image of the fetus.
19. A method according to any one of the preceding claims, wherein: The method includes calculating a gestational age estimate and / or a confidence value based on fetal images obtained at any stage during pregnancy.
20. A device for estimating the gestational age of a fetus, the device comprising: an ultrasound image capture device for obtaining an image of at least a portion of the fetus; as well as An electronic processing device is configured to process the images and calculate a gestational age estimate for the fetus and a corresponding confidence assessment for the estimate.
21. The apparatus according to claim 20, wherein: The processing device is configured to calculate the estimate and / or the confidence assessment using a trained machine learning model.
22. The apparatus according to claim 20 or 21, wherein: The device is configured to provide real-time feedback to the user on the confidence range of the gestational age estimate that has been obtained, so that the user can direct the scan to a more appropriate image if necessary.
23. Apparatus according to any one of claims 20 to 22, wherein: The processing is performed directly on the ultrasound device or on a separate device, which receives a video feed from the ultrasound device or captures a copy of one or more of the ultrasound images via a camera of the device, and the results are optionally displayed on a separate monitor.
24. A method of performing one or more clinical assessments on a fetus, the method comprising: obtaining an ultrasound image of the fetus; and using the trained machine learning model to process the images to determine one or more of the following items, including but not limited to: the presence of multiple fetuses and, in this case, the chorionicity and amnioticity of these fetuses; evidence of fetal vitality by confirming cardiac activity and estimating the fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus having trisomy; estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; markers of genetic abnormalities such as increased thickness of the nuchal folds or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, head presentation, etc.); placental position; category of placental appearance; and estimation of amniotic fluid volume.
25. A device for performing one or more clinical assessments on a fetus, the device comprising a processor for processing an image using a trained machine learning model to determine one or more of the following items, including but not limited to: the presence of multiple fetuses and, in this case, the chorionicity and amnioticity of these fetuses; evidence of fetal vitality by confirming cardiac activity and estimating the fetal heart rate; whether the pregnancy is ectopic; the presence of fetal activity; the risk of the fetus having trisomy; estimated fetal weight and body composition; fetal growth and development, including brain growth and development; the presence of pathology such as ancrania, gastroschisis, spina bifida, or anterior wall coloboma; markers of genetic abnormalities such as increased thickness of the nuchal fold or nasal bone dysplasia; the presence of risk factors for adverse outcomes such as stillbirth; the sex of the fetus; fetal presentation (e.g., breech presentation, cephalic presentation, etc.); placental location; category of placental appearance; and an estimate of the amount of amniotic fluid.