Biometric evaluation

Abstract

The embodiments and models relate to imaging data processing methods, apparatus for processing imaging data, and computer program products configured to perform such image data processing methods. One embodiment provides an apparatus comprising at least one processor and at least one memory for storing instructions, wherein when an instruction is executed by the at least one processor, the apparatus causes at least measurement data relating to a feature of interest from imaging data representing a three-dimensional object containing the feature of interest, models the distribution of the measurement data as a superposition of a true distribution and a noise distribution of inaccurate measurements of the feature of interest, assumes a first shape for the true distribution and a second shape for the noise distribution, and estimates a probability distribution of measurements of the feature of interest based on the modeled measurement data. The configuration described herein provides a method for using or processing imaging data collected with respect to one or more features or elements of interest in a three-dimensional object so that an index of a typical “true” value of a dimension associated with a feature or element of interest can be determined.

Description

[Technical Field]

[0001] The present invention relates to an imaging data processing method, an apparatus for processing imaging data, and a computer program product. [Background technology]

[0002] Image processing is known. Generally, an imaging device or sensor captures an image of an object, providing image data representing that object. This image data can be re-examined to identify the elements of the object and the characteristics of those elements. While image processing techniques exist to assist in identifying the elements that make up an object and the characteristics of those elements, each of these techniques has its own drawbacks. Therefore, there is a desire to provide improved image processing techniques. [Overview of the project] [Means for solving the problem]

[0003] In some embodiments, though not necessarily all, the present invention provides a device comprising at least one processor and at least one memory for storing instructions, wherein when an instruction is executed by at least one processor, the device is caused to acquire measurement data relating to a feature of interest from imaging data representing a three-dimensional object containing the feature of interest, to model the distribution of the measurement data as a superposition of a true distribution of inaccurate measurements of the feature of interest and a noise distribution, to assume a first shape for the true distribution and a second shape for the noise distribution, and to estimate a probability distribution of the measured values ​​of the feature of interest based on the modeled measurement data. Thus, this embodiment recognizes that it is possible to derive an indication of the "true" value for a set of measurement data containing multiple measured values ​​of the feature of interest, based on the probability distribution of the measured values ​​constituting that set of measurement data. In other words, it is possible to determine an accurate indication of a feature or structure of interest appearing in imaging data from such imaging data. Such an approach can provide more reproducible and objective measurements of the feature of interest. Therefore, the apparatus can acquire measurement data relating to features of interest from imaging data representing a three-dimensional object containing the features of interest. The measurement data relating to features of interest may include "raw" data. The measurement data may also include pre-processed data. In either case, the measurement data can be modeled as having a measurement data distribution that is a superposition of the true distribution and noise distribution of inaccurate measurements of features of interest. The noise distribution may include aspects or elements of the measurement data that are not directly attributable to the measurements of interest, i.e., undesirable disturbances in the measurements. These disturbances may be attributable to a wide range of factors, such as interference. In this specification, noise is used to refer to anything that is not a feature of interest. The apparatus can be configured to assume a first shape for the true distribution and a second shape for the noise distribution. The first and second shapes may be different.The device can be configured to estimate the probability distribution of measurements of a feature of interest based on modeled measurement data. The device can start with initial inferences about first and second shapes and can be configured to adjust or iterate such initial inferences each time each measurement constituting the measurement data is processed.

[0004] According to some embodiments, the measurement data relates to the biometrics of features of interest from imaging data representing a three-dimensional object containing the features of interest. With respect to biometric features and associated measurements, the ability to evaluate the biometrics objectively rather than subjectively can have an impact, for example, when making clinical decisions or diagnoses, if the biometrics are commonly used.

[0005] According to some embodiments, the device is configured to estimate a measurement of a feature of interest based on a probability distribution of the measurement of that feature, which is estimated based on modeled measurement data, and this measurement is derived from a peak value derived from the estimated probability distribution. In other words, the true value of the measurement can be represented by the median measurement of the inferred probability distribution.

[0006] According to some embodiments, the device is configured to estimate measurements of a feature of interest based on a probability distribution of measurements of the feature of interest estimated based on modeled measurement data, and the measurements are derived from the median measurement of the estimated probability distribution.

[0007] According to some embodiments, the true distribution of inaccurate measurements of a feature of interest has a first shape modeled as a Gaussian distribution around the measurement. For example, other probability distributions for the true distribution may also be appropriate, based on the measurement or biometric characteristic or property being evaluated and / or the imaging technique used to acquire the imaging data. For example, the distribution may include a multivariate Gaussian projection, and the distribution may include a sum of Gaussian values ​​(e.g., estimating femoral length while avoiding competition with shorter arm or leg bones for a feature with several contributing estimators, of which only one is of interest). Furthermore, the distribution may be a sloped distribution in femoral estimation where the distribution of estimates is unlikely to be symmetrical.

[0008] According to some embodiments, the measurements may include final measurements of the feature of interest. According to some embodiments, the measurements may include intermediate measurements of the feature of interest based on the current estimated probability distribution.

[0009] According to some embodiments, the noise distribution is due to data from misclassified features and has a second shape modeled as a uniform distribution across the search space. For example, other probability distributions for the noise distribution may also be appropriate, based on the measured value or biometric characteristic or property being evaluated and / or the imaging or image data processing technique used to acquire the imaging data.

[0010] According to some embodiments, the device is configured to estimate the probability distribution of measured values ​​of a feature of interest based on modeled measurement data by calculating the posterior probability that each measurement sample in the measurement data belongs to either the true distribution or the noise distribution.

[0011] According to some embodiments, the estimate of the true distribution is updated based on the likelihood that the measured sample of the measurement data is a true measurement.

[0012] According to some embodiments, the device is configured to perform Bayesian estimation regarding a true distribution and a probability distribution.

[0013] According to some embodiments, the relative weights of the true distribution and the noise distribution can be adjusted or updated. Such adjustment or update can be selected based on the nature of the imaging data.

[0014] According to some embodiments, the device is configured to estimate the probability distribution of the measured values of the features of interest based on the modeled measurement data by applying a Bayesian inference approach.

[0015] According to some embodiments, the device is configured to analyze the acquired measurement data and discard or downweight the measured values that deviate from a preselected range associated with the features of interest.

[0016] According to some embodiments, the step of analyzing the acquired measurement data includes filtering some of the measurement data from the set of data measurement values so as to remove, discard, or downweight the measurement values that represent unlikely measured values of the features of interest.

[0017] According to some embodiments, taking a and b as the lower and upper limits of a range selected according to the features of interest, the measured values that do not fall within the range [a, b] in the acquired measurement data are discarded without further processing.

[0018] According to some embodiments, instead of discarding the measured values in the measurement data, the measured values may be downweighted.

[0019] According to some embodiments, downweighting can depend on the magnitude or degree to which the measured value does not fall within the preselected range.

[0020] According to some embodiments, the measurement data includes biometric measurements relating to features of interest within ultrasound imaging data.

[0021] According to some embodiments, the apparatus is configured to acquire measurement data by analyzing one or more images of features of interest obtained from imaging data representing a three-dimensional object.

[0022] According to some embodiments, one or more images include standard frames obtained from ultrasound imaging data.

[0023] According to some embodiments, the device is configured to recognize one or more measurement anchors in an image containing features of interest, and to determine measurement values ​​of the features of interest from the one or more recognized measurement anchors.

[0024] According to some embodiments, the device is configured to use a neural network trained to recognize one or more measurement anchors within an image containing features of interest.

[0025] According to some embodiments, though not all, a method is provided that includes the steps of: obtaining measurement data relating to a feature of interest from imaging data representing a three-dimensional object containing the feature of interest; modeling the distribution of the measurement data as a superposition of a true distribution and a noise distribution of inaccurate measurements of the feature of interest; assuming a first shape for the true distribution and a second shape for the noise distribution; and estimating a probability distribution of measurements of the feature of interest based on the modeled measurement data.

[0026] According to some embodiments, the measurement data relates to the biometrics of the features of interest from imaging data representing a three-dimensional object containing the features of interest.

[0027] According to some embodiments, the method includes the step of estimating a measure of a feature of interest based on a probability distribution of the measure of the feature of interest estimated based on modeled measurement data, wherein the measure is derived from peak values ​​derived from the estimated probability distribution.

[0028] According to some embodiments, the device is configured to estimate a measurement of a feature of interest based on a probability distribution of measurements of the feature of interest estimated based on modeled measurement data, the measurement being derived from the median measurement of the estimated probability distribution.

[0029] According to some embodiments, the measurement may include a final measurement of the feature of interest. According to some embodiments, the measurement may include an intermediate measurement of the feature of interest based on the current estimated probability distribution.

[0030] According to some embodiments, the true distribution of inaccurate measurements of the feature of interest has a first shape modeled as a Gaussian distribution around the measured values.

[0031] According to some embodiments, the noise distribution may be due to data from misclassified features having a second shape, which is modeled as a uniform distribution across the search space.

[0032] According to some embodiments, this method includes the step of estimating the probability distribution of measurements of a feature of interest based on modeled measurement data by calculating the posterior probability that each measurement sample in the measurement data belongs to either the true distribution or the noise distribution.

[0033] According to some embodiments, the estimate of the true distribution is updated based on the likelihood that the measured sample of the measurement data is a true measurement.

[0034] According to some embodiments, this method includes the step of performing Bayesian estimations of the true distribution and the probability distribution.

[0035] According to some embodiments, the relative weights of the true distribution and the noise distribution can be adjusted or updated.

[0036] According to some embodiments, this method includes the step of estimating the probability distribution of measurements of a feature of interest based on modeled measurement data by applying a Bayesian inference approach.

[0037] According to some embodiments, this method includes the step of analyzing acquired measurement data and discarding measurements that fall outside a pre-selected range associated with the features of interest.

[0038] According to some embodiments, the step of analyzing acquired measurement data includes filtering the measurement data from the set of data measurements to remove measurements that are unlikely to represent the features of interest.

[0039] According to some embodiments, a and b are set as the lower and upper limits of a range selected according to the characteristics of the object of interest, and any measurements that do not fall within the range [a,b] in the acquired measurement data are discarded without further processing.

[0040] According to some embodiments, the measurement data includes biometric measurements relating to features of interest within ultrasound imaging data.

[0041] According to some embodiments, this method includes the step of obtaining measurement data by analyzing one or more images of features of interest obtained from imaging data representing a three-dimensional object.

[0042] According to some embodiments, one or more images include standard frames obtained from ultrasound imaging data.

[0043] According to some embodiments, this method includes the steps of recognizing one or more measurement anchors in an image containing a feature of interest, and determining a measurement value of the feature of interest from the one or more recognized measurement anchors.

[0044] According to some embodiments, the method includes the step of using a neural network trained to recognize one or more measurement anchors in an image containing features of interest.

[0045] According to some embodiments, this method includes the step of analyzing acquired measurement data and discarding measurements that fall outside a pre-selected range associated with the features of interest.

[0046] In some, though not all, embodiments, a computer program product is provided that, when run on a computer, can operate to acquire measurement data about a feature of interest from imaging data representing a three-dimensional object containing the feature of interest, model the distribution of the measurement data as a superposition of the true distribution and noise distribution of inaccurate measurements of the feature of interest, assume a first shape for the true distribution and a second shape for the noise distribution, and estimate the probability distribution of the measured values ​​of the feature of interest based on the modeled measurement data.

[0047] In some embodiments, though not necessarily all, the apparatus is provided, comprising: means configured to acquire measurement data relating to a feature of interest from imaging data representing a three-dimensional object containing the feature of interest; means configured to model the distribution of measurement data as a superposition of a true distribution and a noise distribution of inaccurate measurements of the feature of interest; means configured to assume a first shape for the true distribution and a second shape for the noise distribution; and means configured to estimate the probability distribution of measurements of the feature of interest based on the modeled measurement data.

[0048] In some embodiments, though not necessarily all, the apparatus is provided, comprising: a circuit configured to acquire measurement data relating to a feature of interest from imaging data representing a three-dimensional object containing the feature of interest; a circuit configured to model the distribution of measurement data as a superposition of the true distribution and noise distribution of inaccurate measurements of the feature of interest; a circuit configured to assume a first shape for the true distribution and a second shape for the noise distribution; and a circuit configured to estimate the probability distribution of measurements of the feature of interest based on the modeled measurement data.

[0049] Further specific preferred embodiments are described in the attached independent and dependent claims. Features of the dependent claims may be combined with features of the independent claims as appropriate, and may also be combined with features other than those expressly described in the claims.

[0050] When a device is described as being capable of operating to provide a function, it will be understood that this includes the features of the device that provide that function, or the features of the device that are adapted or configured to provide that function. [Brief explanation of the drawing]

[0051] Embodiments of the present invention will be further described below with reference to the accompanying drawings. [Figure 1A] ~ [Figure 1B] This figure includes the ultrasound frame from which the features of the object of interest were measured. [Figure 2] This figure shows a graph of the idealized distribution of measurements for measuring the characteristics of a structure, element, or object of interest, with true values ​​at 30 mm. [Figure 3] This diagram schematically shows some of the components of a configuration according to one possible embodiment. [Figure 4] This diagram schematically shows the steps of the method performed by the configuration. [Modes for carrying out the invention]

[0052] Before discussing the embodiments in more detail, we will first provide an overview.

[0053] The configuration provides a technique for measuring elements, parts, components, or features of a three-dimensional object that is captured and represented in image data. Generally, it is necessary not only to recognize those elements of a three-dimensional object in the image data, but also to identify or observe them in a specific direction or specific imaging plane through which the three-dimensional object is seen. For example, consider imaging a complex three-dimensional part, such as a mechanical assembly. It may be desirable not only to identify elements such as O-rings within that mechanical assembly, but also to identify the O-rings as seen in a plan view rather than a cross-sectional view. Similarly, consider imaging a complex three-dimensional part, such as the body of a human or animal. It may be desirable not only to identify heart valves, but also to identify the valves as seen in a cross-sectional view rather than a plan view. After identifying one or more elements that form a part of a three-dimensional object, the configuration provides a mechanism that can determine one or more dimensions of such elements.

[0054] For example, consider one specific case of 3D imaging where the dimensions of elements within the imaged 3D object may be significant. In prenatal ultrasound scanning, biometrics such as head circumference and femur length are routinely measured to assess fetal growth and development. Current clinical standards require ultrasound technicians to perform a small number of manual measurements of these biometrics during an ultrasound scanning session. Typically, the ultrasound technician performs such measurements by freezing the image plane containing the element of interest during live scanning, and then determining one or more dimensions of the element of interest within the frozen image plane using manually positioned calipers or alternative forms of measurement. Such an approach will generally be understood to involve the selection of a single view or image of the element, in other words, the frozen or other-user-selected image plane. After the single view is selected, the dimensions or measurements are determined based on the appearance of the element within that single view. In some cases, multiple measurements of the element of interest may be performed, but all such multiple measurements are performed on the same single captured frozen image plane. When multiple measurements are taken, it is common practice to select one of the measurements as the reported value, at least in the case of prenatal ultrasound imaging to assess fetal growth, for example.

[0055] Products and methods have been established that can assist ultrasound technicians in identifying views or planes from which meaningful measurements related to elements or features can be obtained. An example of such an assistive technology is described in International Publication No. 2018 / 060723(A1) (KING'S COLLEGE LONDON), which discloses an imaging method, an imaging apparatus, and a computer program product. The imaging method disclosed in that document includes the steps of receiving image data of a three-dimensional object and assigning confidence levels to at least a portion of the image frames of the image data using a machine learning algorithm, the confidence level indicating the likelihood that the image frame has a specified element imaged on a specified plane through the three-dimensional object. In this way, a specific element imaged in a desired manner can be identified from the image data of the three-dimensional object. After identifying the “optimal” plane containing the specified element based on the assigned confidence levels, measurements of the specified element on that optimal plane can be performed in the same manner as described above with respect to a single frame selected by the ultrasound technician.

[0056] While the degree of automation and assistance in ultrasound imaging is gradually increasing, the workflow in clinical practice remains largely unchanged to date. It is recognized that biometric acquisition methods relying on only one or a very small number of still images can lead to significant variability between different measurements.

[0057] The configurations described herein provide methods for using or processing imaging data collected with respect to a three-dimensional object containing one or more features or elements of interest, so that it is possible to determine an indication of a representative "true" value of a dimension associated with a feature or element of interest. According to some configurations, it is possible to provide an indication of the reliability of the determined true value.

[0058] For example, using a method such as that described in International Publication No. 2018 / 060723A1 (KING'S COLLEGE LONDON), multiple imaging planes in which elements or features of interest can be observed can be automatically identified. In the case of International Publication No. 2018 / 060723A1 (KING'S COLLEGE LONDON), these planes may include standard fetal imaging planes in which one or more elements, features, or structures of interest exist. Measurements of elements, features, or structures of interest within each identified imaging plane can be obtained.

[0059] Identifying multiple imaging planes, including elements, features, or structures of interest, can support, for example, the configuration described.

[0060] In embodiments using the method of International Publication No. 2018 / 060723A1 (KING'S COLLEGE LONDON), multiple imaging planes containing elements, features, or structures of interest can be identified from a block of three-dimensional imaging data, and hundreds of distinct measurements of the elements, features, or structures can be generated from these multiple imaging planes. For example, such relevant imaging data can be processed to enable the acquisition of automated biometric measurements related to elements or features of interest for each relevant frame / plane of the three-dimensional scanning imaging data.

[0061] In other words, techniques can be used to generate hundreds of distinct measurements in relation to a single block of 3D imaging data from a scan.

[0062] Multiple measurements taken regarding the dimensions of an element, structure, or feature of interest are likely to vary, which raises the question of how to meaningfully obtain the "true" biometric value from those measurements.

[0063] It will be understood that the constructive approach described in more detail below does not depend on the approach described in International Publication No. 2018 / 060723A1. Strictly speaking, it does not depend on the provision of imaging datasets or frames of imaging datasets, and requires only a series of measurements derived from imaging data following the described distribution. However, the constructed approach described generally utilizes a considerable number of measurements to be useful and meaningful. It will be understood that applying the constructed approach described to a small number of available measurements may yield representative "true" values, which may be very close to the chosen prior distribution or somewhat unnatural.

[0064] The configuration described relates to an apparatus and method configured to provide a mechanism for estimating a measurement, for example, the “true” value of a biometric, from a set of indicators of that measurement. The indicators of the measurement may include measurements obtained from imaging data, for example, medical imaging data. In particular, one embodiment of the configuration described relates to an apparatus and method configured to estimate the true value from a set of measurements of one or more elements, parts, components, or features of a three-dimensional object that has been imaged and represented in the image data. Thus, the approach can provide an improved mechanism for identifying and providing accurate measurements of elements, features, or structures from imaging data relating to a three-dimensional object.

[0065] The constructs described relate to Bayesian methods that support obtaining estimates of the “true” value of a biometric from a set of noisy measurements. Some of the constructs described can provide indices of confidence intervals for the determined “true” value.

[0066] Therefore, the general approach adopted by the configuration described is to determine the distribution of measured values. 1) The “true” distribution of inaccurate measurements of correct structures, elements, or features, which in some embodiments are modeled as a Gaussian distribution before and after true biometrics, and 2) In some embodiments, the “noise” distribution of metrics from misclassified structures is modeled as a uniform distribution across the search space. It operates by modeling it as a superposition of these two distributions.

[0067] It will be understood that the model selected for each distribution can be chosen depending on one or more characteristics or properties of the data stream representing a set of noisy measurements on which the approach is performed. Generally, for example, the following examples are given for Gaussian and uniform distributions to model true and noisy measurements, respectively, but it will be understood that the approaches described are easily extendable to other probability distributions, provided they are easy to handle with Bayesian methods.

[0068] According to the configuration, for each measured sample, its posterior probability of belonging to one of the aforementioned distributions is calculated. Then, based on the likelihood that the measured value is the true measured value, the estimate of the true distribution can be updated. According to some embodiments, it will be understood that the relative weights of the true distribution and the noise distribution can be updated in real time.

[0069] Therefore, the configuration can support the reporting of a central estimate of the truth value. This central estimate can be obtained in real time, along with the standard error for the variance of the true distribution and the number of observed samples. The reported central estimate provides a representative indication of the ground truth measurement and supports an accurate indicator of the measurement associated with the elements, structures, or features observed in the captured image data.

[0070] While the approach has been outlined, this document provides further details regarding the characteristics of various possible embodiments of such an approach.

[0071] Data selection and preprocessing In general, the methods described herein can be applied to both non-biological and biological applications. Generally, in the context of imaging biological materials, measurements of elements or features of interest in the biological domain are referred to as "biometrics." Possible applications of the methods are described in terms of exemplary biological applications.

[0072] As described above, a standard clinical protocol for routine prenatal ultrasound scanning involves locating and observing a defined set of standard planes and measuring one or more features or elements shown within those planes. Standard planes are used to observe the anatomical structures of the fetus in a standardized, and therefore equivalent, manner across scans. With respect to fetal imaging, each standard plane contains specific anatomical features imaged from a particular angle. In clinical practice, the ultrasound technician manually locates the standard planes and saves screenshots of those planes for report preparation.

[0073] As referenced above, imaging data can be acquired in unconventional and alternative ways. For example, instead of recording only one or more standard planes of interest, it may be possible to record the entire set of data acquired during an ultrasound session. Using systems such as Sononet[1], standard planes can be detected from imaging data captured during an ultrasound scan without manual input from the ultrasound technician. The method described in Sononet features a convolutional neural network (CNN) trained on thousands of video recordings of normal ultrasound scans. Sononet can process all frames of image data captured as a result of an ultrasound scan and output which standard plane (if any) each frame belongs to. Sononet has a classification accuracy of 85–95% depending on the specific standard plane under consideration.

[0074] Biometric measurement Several standard planes, such as the femoral plane for prenatal fetal scanning, include biometric measurements that must be taken during ultrasound scanning according to clinical protocols. Such biometric measurements may include, for example, fetal femoral length.

[0075] In clinical practice, an ultrasound technician takes a screenshot of the relevant standard plane and then manually measures the edges of the feature or structure. This is done only for images within the standard plane.

[0076] Figures 1A and 1B include ultrasound frames in which features of interest were measured. Figure 1A shows the measurement results of the object of interest obtained by the method described later for an image that was accurately classified as containing the fetal femur. Figure 1B shows the measurement results of the object of interest obtained by the method described later for an image that was incorrectly classified as containing the fetal femur.

[0077] To obtain such measurements from a set of frames of interest (e.g., frames identified by a system such as Sononet), one possibility is to use another neural network, e.g., U-Net[2], to predict measurements of features within the frames. For example, U-Net can be configured to predict endpoints of structures, elements, or features within frames labeled by an ultrasound technician (see Figure 1A). Such endpoint predictions facilitate the measurement of distances between endpoints to obtain biometric indications, including the distance between two endpoints. According to some embodiments of the method, separate neural networks can be trained to recognize endpoints or other measurement anchors with respect to each biometric of interest. For example, a suitable neural network can be trained with thousands of manual measurements of a biometric of interest.

[0078] Some configurations can be configured to identify features of interest and obtain associated measurements for multiple frames or planes. For example, with respect to the Sononet frames referenced above, one embodiment of the method can determine one or more measurements relating to features, elements, or structures of interest for all frames of ultrasound video classified as relating to a particular standard plane.

[0079] Sononet's standard plane classification can sometimes fail, misclassifying planes, which may result in the request or creation of measurements for images that do not belong to the standard plane.

[0080] Because biometric measurement neural networks trained using the method described are generally trained only on views known to be of the relevant standard plane, the output of a biometric neural network (when a non-standard image is presented) can be unpredictable. While a biometric measurement network can often still output measurements, these measurements may include substantially meaningless measurements from questionable endpoints (see Figure 1B).

[0081] Doubtful feature measurements can add noise to a set of measurements. However, it will be understood that it is possible to filter out such doubtful feature measurements from the set of measurements. In particular, measurements derived from misclassified images are more likely to be random than measurements from accurately classified images, which tend to cluster around or close to the true femur length. According to some embodiments, it is possible to filter a set of measurements concerning a structure, feature, or element to remove measurements that represent impossible or highly unlikely measurements of the feature, element, or structure of interest. For example, with respect to fetal femur length or similar biometrics, filtering can be performed to remove measurements from the input dataset that represent biologically impossible measurements.

[0082] In other words, a method can be implemented to automatically discard, without further processing, any measurements that fall outside the range [a,b] (where a and b are the lower and upper limits of the length range that includes the actual measurements) with respect to the structure, element, or feature of interest.

[0083] The configuration may be configured to probabilistically reject readings and / or to lower their weight (or exclude them) from the overall estimate of the measurements of the feature of interest.

[0084] Figure 2 graphically shows the ideal distribution of measurements in the measurement of a structure, element, or feature of interest, where the true value is 30 mm. The observed distribution of measurements is the sum of the measurement distribution and the noise distribution.

[0085] The structure recognizes that measurements of the same elements, features, or structures, such as biometric measurements in different frames of a standard plane, should differ only slightly and should generally be expected to cluster around (between) the mean.

[0086] According to some configurations, it can be assumed that the measurements of features, elements, or structures follow the following normal distribution: N(μ,σ 2 ) In the above equation, μ is the mean of the distribution (if unbiased, this should be the same as the "true" value for that measurement), σ 2 This is its variance (which depends on the precision of the measurement).

[0087] However, since standard plane detection methods are incomplete, some misclassified planes also return measurement values. Such planes do not contain elements, structures, or features of interest, and the measurement network is not trained on such data, so the output from the measurement values derived from such planes essentially contains random lengths within the range [a, b] (where a and b are the lower and upper bounds of the length range within which real measurement values fall). The noise of that type can potentially be modeled as a uniform distribution U(a, b).

[0088] Therefore, the observed measurement value distribution D o is, for example, a weighted sum of the measurement value distribution N(μ, σ 2 ) for correctly classified standard planes, and the noise distribution U(a, b) (relating to misclassification of structures or planes). Therefore, the observed distribution can be considered as follows. D o =P t N(μ, σ 2 )+(1 - P t )U(a, b) [1] In the above equation, P t represents the proportion of measurement values coming from the true measurement value distribution.

[0089] This equation contains several unknowns Pt, μ, σ 2 and the method attempts to estimate these unknowns by repeatedly using the Bayesian approach.

[0090] P t is not known in advance and usually depends on, for example, the measurement values of interest such as biometrics, and the proportion of measurement values from misclassifications determined to be within the bounds [a, b].[[]]

[0091] To initialize the Bayesian estimation, the estimated values are initialized with prior values. According to some methods, for all measurements, P t0An initial value of =0.5 can be selected. Regardless of the initial value chosen, it will be understood that the estimation process is configured to converge to the true value from any starting point that is neither 0 nor 1. The prior values ​​are also given weights W0, which are another configurable parameter that can vary depending on the chosen implementation of the method described. The estimation will proceed regardless of what W0 is set to, P t It should converge to the true value of P, but if W0 is set too high, the estimation may converge slowly, and if W0 is set too low, P t The estimation may fluctuate inappropriately.

[0092] Each new measurement x that forms part of the dataset i For this, it is possible to estimate the likelihood that it will be the "true" measurement T or noise T.

[0093] It is possible to apply Bayes' rules as follows:

number

[0094] Next, based on the calculated likelihood, P for a given measurement of an element, feature, or structure is calculated. t It is possible to update this. This can be done using a weighted sum.

[0095] x i If it is the i-th measurement of a particular element, structure, or feature, then the rule

number

[0096] Estimation of distribution parameters Using a similar method, the measured value distribution parameters μ and σ 2 It is possible to estimate this.

[0097] Prior estimates of these values ​​μ0 and σ0 2 However, an appropriate prior weight W μ,0 and W σ2,0 It is set together with this.

[0098] Next, each new measurement x i For this, the likelihood P(T|x) is the true measured value. i By weighting them accordingly, the estimated values ​​of the measurement distribution parameters can be fine-tuned as follows:

number

number

[0099] In the embodiments described, the noise distribution is assumed to be uniform, so there is no need to update the estimate for that term.

[0100] μ and σ 2 The estimate for μ can be updated in real time and independently for each measurement of the structure, element, or feature of interest. At any given moment, the estimate of μ represents the best estimate of the true value of the measurement of interest, assuming an unbiased set of measurements.

[0101] On the other hand, σ 2 The estimated value forms part of the calculation of the tolerance to assign to the measurement of interest. It will be understood that the tolerance is based on the standard error of the mean, as well as the number of measurements processed as part of the process.

[0102] While it has been explained that the use of a normal distribution for true measurements and a uniform distribution for noise is easy to handle for simple calculations and that this method gives good performance for the biometric data for which it was developed, it will be further understood that the schematic method described can also be applied to distributions under different assumptions.

[0103] Figure 3 schematically shows some components of a configuration according to one possible embodiment, and Figure 4 schematically shows the steps of the method performed by the configuration.

[0104] Figure 3 schematically shows some components of system 1000, which includes apparatus 2000 in several configurations. System 1000 includes an imaging device, such as an ultrasound scanner or MRI scanner 1100, configured to generate imaging data 1200 relating to images of a three-dimensional object containing or comprising elements or structures of interest. This data can be provided to apparatus 2000 in real time or by other means. Apparatus 2000 may include a computing device.

[0105] Device 2000 is, A means 2100 configured to acquire measurement data regarding the structure of the object of interest from imaging data representing a three-dimensional object containing the features of the object of interest, A means 2200 configured to model the distribution of measurement data as a superposition of the true distribution of inaccurate measurements of features of interest and the noise distribution, Means 2300 configured to assume a first shape for the true distribution and a second shape for the noise distribution, The system includes means 2400 configured to estimate the probability distribution of measured values ​​of a feature of interest based on modeled measurement data.

[0106] Figure 4 schematically shows the steps of the method performed by the configuration.

[0107] Device 2000 is, 4100: A step of obtaining measurement data regarding the features of the object of interest from imaging data representing a 3D object containing the features of the object of interest, 4200: A step of modeling the measurement data distribution as a superposition of the true distribution of inaccurate measurements of the features of interest and the noise distribution, 4300: A step of assuming a first shape for the true distribution and a second shape for the noise distribution, 4400: Can be configured to perform a method that includes the step of estimating the probability distribution of the measurement of interest based on modeled measurement data.

[0108] While exemplary embodiments of the present invention have been disclosed in detail herein with reference to the accompanying drawings, it will be understood that the present invention is not limited to exact embodiments, and that various changes and modifications can be made without departing from the scope of the invention as defined by the accompanying claims and equivalents.

[0109] It will be readily apparent to those skilled in the art that the steps of the various methods described above can be performed by a programmed computer. In this specification, some embodiments are also intended to encompass a program storage device, e.g., a digital data storage medium, which is machine- or computer-readable and encodes a machine-executable or computer-executable program of instructions, the instructions performing some or all of the steps of the methods described above. The program storage device may be, for example, a digital memory, a magnetic storage medium such as magnetic disks and magnetic tapes, a hard drive, or an optically readable digital data storage medium. These embodiments are also intended to encompass a computer programmed to perform the steps of the methods described above. In this specification, the term "non-temporary" refers to a limitation of the medium itself (i.e., tangible rather than signal-based) rather than a limitation of data storage persistence (e.g., RAM vs. ROM).

[0110] As used herein, the term "circuit" is, (a) Hardware-only circuit embodiments (such as embodiments in analog and / or digital circuits only), and (b) combinations of hardware circuits and software, for example (where applicable), (i) combinations of analog and / or digital hardware circuits and software / firmware, (ii) Any part of a hardware processor (including a digital signal processor), software, and memory that functions together to cause a mobile phone or a server or other device to perform various functions, (c) Hardware circuits and / or processors, such as microprocessors or parts of microprocessors, that require software (e.g., firmware) to operate, but may not be present when not required for operation. This can refer to one, two or more, or all of them.

[0111] This definition of circuit applies to all uses of the term herein, including in the claims. Further examples, as used herein, the term circuit also encompasses embodiments of a hardware circuit or processor (or more processors) or a portion of a hardware circuit or processor, as well as any accompanying software and / or firmware. The term circuit also encompasses, for example, baseband integrated circuits or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or network devices, as applicable to elements of specific claims.

[0112] While exemplary embodiments of the present invention have been described in the preceding paragraphs with reference to various examples, it should be understood that modifications can be made to the given examples without departing from the claimed scope of the invention.

[0113] The characteristics described above can also be used in combinations other than those explicitly stated.

[0114] While the functions were described by referring to specific features, those functions can also be implemented by other features, regardless of whether they are explicitly mentioned or not.

[0115] While the features have been described with reference to specific embodiments, those features may also be present in other embodiments, whether or not they are described.

[0116] While the above specification has sought to draw attention to those features of the present invention that are considered particularly important, it should be understood that the applicant seeks protection for any patentable features or combinations of features shown above and / or in the drawings, whether or not they were particularly emphasized.

[0117] References (1)Christian F. Baumgartner, Konstantinos Kamnitsas, Jacqueline Matthew, Tara P. Fletcher, Sandra Smith, Lisa M. Koch, Bernhard Kainz, and Daniel Rueckert. Sononet: Real-time detection and localisation of fetal standard scan planes in freehand ultrasound. IEEE Transactions on Medical Imaging, 36, 2017. (2)Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. volume 9351, 2015.

Claims

1. A device comprising at least one processor and at least one memory for storing instructions, wherein when an instruction is executed by the at least one processor, the device has at least Obtaining measurement data regarding the characteristics of the object of interest from imaging data representing a three-dimensional object containing the characteristics of the object of interest, The distribution of measurement data The true distribution of inaccurate measurements of the features of interest, and Noise distribution Modeling it as a superposition of, Assuming a first shape for the true distribution and a second shape for the noise distribution, To estimate the probability distribution of measured values ​​of the features of interest based on the modeled measurement data. The apparatus that causes the following to occur.

2. The apparatus according to claim 1, wherein the measurement data relates to the biometrics of the features of the object of interest from imaging data representing a three-dimensional object including the features of the object of interest.

3. The apparatus according to claim 1 or 2, configured to estimate a measured value of a feature of interest based on an estimated probability distribution of measured values ​​of the feature of interest based on the modeled measurement data, wherein the measured value is derived from the median value of the estimated probability distribution.

4. The apparatus according to any one of claims 1 to 3, wherein the true distribution of inaccurate measurements of the features of interest has a first shape modeled as a Gaussian distribution around the measured values.

5. The apparatus according to any one of claims 1 to 4, wherein the noise distribution is due to data from misclassified features having a second shape modeled as a uniform distribution across the search space.

6. The apparatus according to any one of claims 1 to 5, wherein for each measurement sample of the measurement data, the apparatus is configured to estimate the probability distribution of the measured values ​​of the feature of interest based on the modeled measurement data by calculating the posterior probability that the measurement sample belongs to either the true distribution or the noise distribution.

7. The apparatus according to claim 6, wherein the estimate of the true distribution is updated based on the likelihood that the measurement sample of the measurement data is a true measurement value.

8. The apparatus according to any one of claims 1 to 7, which can update the relative weights of the true distribution and the noise distribution.

9. The apparatus according to any one of claims 1 to 8, configured to estimate the probability distribution of measured values ​​of the feature of interest based on the modeled measurement data by applying a Bayesian inference approach.

10. The acquired measurement data is analyzed, and any measurements that fall outside the pre-selected range associated with the characteristics of interest are discarded. The apparatus according to any one of claims 1 to 9, configured as described above.

11. The apparatus according to any one of claims 1 to 10, configured to filter some measurement data from a set of data measurement values ​​so as to remove measurement values ​​that represent impossible measurements of the features of interest.

12. The apparatus according to claim 10 or 11, wherein any measurement values ​​in the acquired measurement data that fall outside the range [a, b] are discarded without further processing, and a and b are the lower and upper limits of a range selected based on the characteristics of the object of interest.

13. The apparatus according to any one of claims 1 to 12, wherein the measurement data includes biometric measurements relating to features of interest within ultrasound imaging data.

14. The apparatus according to any one of claims 1 to 13, configured to acquire measurement data by analyzing one or more images of features of an object of interest obtained from imaging data representing the three-dimensional object.

15. The apparatus according to claim 14, wherein the one or more images include a standard frame obtained from ultrasound imaging data.

16. The apparatus according to claim 14 or 15, configured to recognize one or more measurement anchors in an image containing the features of interest, and to determine a measurement value of the features of interest from the one or more recognized measurement anchors.

17. The apparatus according to claim 16, configured to use a neural network trained to recognize one or more measurement anchors in an image containing the features of the object of interest.

18. The steps include: obtaining measurement data relating to the characteristics of the object of interest from imaging data representing a three-dimensional object containing the characteristics of the object of interest; The distribution of measurement data The true distribution of inaccurate measurements of the features of interest, and Noise distribution The step of modeling it as a superposition, A step of assuming a first shape for the true distribution and a second shape for the noise distribution, A step of estimating the probability distribution of measured values ​​of the features of interest based on the modeled measurement data. A method that includes this.

19. A computer program product that, when executed on a computer, is operable to perform the method described in claim 18.

Images