Fetal Heart Rate Monitoring

Abstract

The present invention relates to an ultrasound fetal heart rate monitoring system. An ultrasound probe placed on the maternal abdomen provides sensor data. The ultrasound probe first operates in an imaging mode to acquire an ultrasound image from the ultrasound probe. A trained object detection model is applied to the ultrasound image to determine whether the fetal heart is located within a target area of ​​the ultrasound image that can be addressed by the Doppler mode of the ultrasound probe. If the fetal heart is located within the target area, the ultrasound probe is switched to a Doppler mode, a Doppler ultrasound signal is acquired from the ultrasound probe, and the fetal heart rate is calculated from the Doppler ultrasound signal.

Description

[Technical field]

[0001] The present invention relates to an ultrasound fetal heart rate monitoring system and a corresponding computer-implemented ultrasound fetal heart rate monitoring method.The present invention further relates to a computer readable medium. [Background technology]

[0002] Complications during delivery can affect the safety of both the mother and the newborn. Worldwide, there are approximately 500,000 maternal deaths each year due to complications during delivery. Approximately 7 million women suffer major long-term problems after giving birth, and 50 million women face adverse health consequences after giving birth.

[0003] The fetal heart rate is important information for assessing the well-being of the fetus during labor. Currently, fetal monitors based on Doppler ultrasound are typically used for this purpose. Using the Doppler effect, the fetal heart rate can be calculated based on the measured frequency shift. The fetal heart rate is measured using multiple ultrasound transducer elements based on the power of the Doppler signal received at these elements.

[0004] Unfortunately, current fetal monitoring is complicated and time-consuming to use, especially since the fetal monitor transducer is (re)positioned manually. It is important for the reliability of the fHR (fetal heart rate) measurement that the fetal heart is placed in the ultrasound beam. In clinical practice, the clinician palpates the maternal abdomen to identify the fetal position, and then fixes the ultrasound transducer to the maternal abdomen where the best fHR signal can be obtained. Finding the optimal position of the transducer is done based on listening to the strength of the Doppler sound output and / or based on signal quality indicators. Displacement of the ultrasound transducer or displacement of the fetal heart out of the ultrasound beam can cause the fHR signal to be lost. Thus, obstetricians are often required to repeat the tedious procedure of positioning the ultrasound transducer to avoid long-term loss of the fHR signal. Moreover, it is disadvantageous that the person using the equipment needs to be well trained, especially since there is a risk that the maternal heart rate (e.g., captured from the umbilical cord or other maternal vessels) will be selected as the fetal heart rate by the fetal monitor.

[0005] European patent EP 3454758 B1 discloses a method for determining the optimal position of an ultrasound transducer during fetal health monitoring. An ultrasound device is used that has a central piezoelectric element and additional piezoelectric elements arranged in a circular configuration around the central piezoelectric element. The device supports a position assistance mode and a fetal heart rate mode. In the position assistance mode, one element of the transducer is used to generate an ultrasound signal, and ultrasound echoes of this signal are received at each element to determine a respective Doppler signal. The user is informed which transducer has the strongest Doppler signal strength, which assists the user in repositioning the device so that the central element has the strongest signal. In the fetal heart rate mode, all piezoelectric elements are used, both transmitting and receiving, to generate an overall Doppler signal of the fetal heart rate. Summary of the Invention [Problem to be solved by the invention]

[0006] For example, it would be beneficial to provide a system for fetal heart rate monitoring that better assists in ensuring correct placement of an ultrasound probe being used for an inexperienced user, such as during home monitoring. [Means for solving the problem]

[0007] According to a first aspect of the present invention there is provided an ultrasound fetal heart rate monitoring system and corresponding method as defined respectively by claims 1 and 14. According to a further aspect of the present invention there is provided a computer readable medium as defined by claim 15.

[0008] Various aspects relate to monitoring the fetal heart rate by acquiring Doppler ultrasound signals using an ultrasound probe. The Doppler ultrasound signal exhibits a frequency shift in the measurement volume. Based on this frequency shift, the fetal heart rate can be determined. In particular, as known per se, the fetal heart rate can be determined by determining the speed and / or direction in which the heart walls move. Most often, Doppler ultrasound signals are acquired in the so-called pulse wave (PW) Doppler mode. In PW Doppler mode, multiple pulses are transmitted and the relative phase change of these transmitted pulses can be used to derive a frequency shift and thereby said speed and / or direction. As an alternative to using the PW Doppler method, it is also possible to use the continuous wave (CW) Doppler method. In CW Doppler, sound waves are transmitted and received continuously, making it possible to record their respective speeds along a defined path.

[0009] To ensure that the Doppler ultrasound signal actually represents a measurement of the fetal heart, and not, for example, the pulsation of the maternal blood vessels, the inventors of the present application envisage using an ultrasound probe that can also be operated in an imaging mode to acquire ultrasound images. In particular, unlike the probes typically used for Doppler ultrasound, this probe may comprise a phased array of transducers, for example a 1D or 2D phased array. The ultrasound image data is input to an object detection model to determine whether the fetal heart is located within a target area of ​​the ultrasound image that can be handled by the Doppler mode of the ultrasound probe. The imaging mode can for example be the so-called B-mode (brightness mode), also known as 2D mode. For example, a one-dimensional phased array of transducers can be used to scan a plane through the measurement volume, thereby acquiring a 2D ultrasound image, in other words a 2D ultrasound scan frame.

[0010] Although ultrasound images are known per se, they are not currently widely used during active labor because their operation, including data acquisition and interpretation, generally requires a high level of training effort and the current form factor is not suitable for monitoring purposes. However, the present inventors have discovered that ultrasound images can still be advantageously used in the context of Doppler-based fetal heart rate monitoring by using ultrasound images to determine whether the fetal heart is located within a target region.

[0011] The trained object detection model is applied to an ultrasound image acquired in imaging mode. The object detection model is configured to locate the fetal heart in the ultrasound image, e.g., to output the location of the fetal heart in the ultrasound image if detected, or to output that the fetal heart was not detected at all. If the detected location falls within a particular target region, the ultrasound probe is switched to a Doppler mode to measure the fetal heart rate, e.g., by using one of the transducers of a phased array that is also used for imaging mode as described above.

[0012] The provided approach has several advantages. By using a trained object detection model applied to the ultrasound images, the fetal heart is reliably located. As a result, when the fetal heart is recognized, the fetal heart rate measured when switching to Doppler mode is more likely to be measured accurately. By using ultrasound images, the imaging capabilities of ultrasound sensors can be exploited. With this newly available visual information, state-of-the-art object detection models, such as convolutional neural networks, can be utilized. Object detection techniques can provide excellent performance and can be performed in real-time or near real-time on common hardware. One reason for the improved accuracy is that the object detection model operates on the ultrasound images and not on the Doppler signals. This means that the model can use other parts of the maternal abdomen and / or fetal anatomy to localize the fetal heart, even during, for example, the ongoing stages of labor. For example, the fetal spine or fetal femur can be seen in the ultrasound images, unlike the Doppler data. Even if the object detection model is only trained to recognize the fetal heart, it can also implicitly use this other information.

[0013] Also, since the recognition is visual and not based on the Doppler signal, the risk of confusing the fetal heart with other sources of Doppler signal generation is eliminated. Relying on the strongest Doppler signal strength does not, by itself, guarantee that the fetal heart is being measured, and does not guarantee that another anatomical structure that generates a Doppler shift in the desired baseband is not being measured. Since the visual recognition is independent of the Doppler shift generated by the fetal heart itself or any other blood vessels, its accuracy and therefore its safety are also improved.

[0014] Another advantage is that localization can be obtained even if the fetal heart is not at the focus of the beam. For example, the measurement volume of an ultrasound probe in Doppler mode can be a cylindrical region about 10 cm wide. Providing guidance based on Doppler ultrasound is effective when the fetal heart is close to that measurement volume, for example within 20 cm. Furthermore, Doppler-based techniques have a particularly high risk of recognizing other anatomical sites than the fetal heart or otherwise providing ambiguous guidance, whereas image-based localization remains effective. For example, image-based localization can be effective based on first placing the probe at any location on the abdomen. For example, the probe is placed in two positions within a relatively short period of time, and then these positions inform the best position to measure the fetal heart, even if they are not where the fetal heart itself is located.

[0015] Furthermore, the use of image detection models applied in imaging mode also has many additional advantages related to the fact that while in imaging mode, the system can output ultrasound images and / or locations, which can be used in a variety of ways, as described elsewhere, for example, to detect and provide guidance to other parts of the fetal anatomy, even when the fetal heart is not visible, to indicate said images and / or locations, and / or to focus Doppler ultrasound.

[0016] Use of the Doppler mode to measure fetal heart rate has a number of advantages over using the probe only in imaging mode. One advantage is the use of current reliable measurement technology for the Doppler mode. In addition, the Doppler mode can use a lower amount of power than the imaging mode. For example, use in the Doppler mode can use a lower power budget of 20 mW / cm 2The imaging mode uses less than 20 mW / cm while the imaging mode uses more. As a result of the reduced power consumption, the system can be more suitable for continuous monitoring, e.g., the system can be used for monitoring for at least 1 hour, at least 4 hours, or at least 10 hours. In particular, during such long periods of use, less power, e.g., 20 mW / cm 2 Using less power helps prevent damage to bone and skin. Lower power also reduces heat generated by the device, making it feel more comfortable on the abdomen and improving the lifespan of the device's components.

[0017] The same transducer can be used to capture both the ultrasound image and the Doppler ultrasound signal, e.g., the Doppler ultrasound signal can be captured by one of the transducers that is also used for imaging. In this way, a relatively small amount of hardware is sufficient, and since the transducer used in both cases is the same, the location of the ultrasound image in the imaging mode is guaranteed to match the area that the ultrasound sensor can cover in the Doppler mode.

[0018] The target area for switching to Doppler mode can coincide with the entire ultrasound image, but is preferably a sub-area of ​​this entire ultrasound image, for example a sub-area excluding the border area of ​​the ultrasound image, for example a sub-area excluding at most or at least 10%, at most or at least 20%, or at most or at least 30% of the imaged area. In this way, when the probe is switched to Doppler mode, then moved slightly so that the signal is lost, and then switched to imaging mode, the fetal heart is prevented from being out of range of the ultrasound sensor. The area that can be addressed by the ultrasound probe in Doppler mode often coincides with the area captured by the ultrasound image, for example, the entire image is addressable by Doppler mode. Also, in such a case, the target area where the device switches to Doppler mode is typically a sub-area of ​​this entire area that is imaged and / or Doppler measured.

[0019] Optionally, when in Doppler mode, a signal quality indicator of the Doppler ultrasound signal may be determined, for example, using known techniques. If this signal quality indicator does not meet a predefined quality threshold, the ultrasound probe may be returned to imaging mode, and the object detection model may be applied again iteratively, for example, until the fetal heart is located. Thus, the measured fetal heart rate is further guaranteed to be accurate, i.e., in imaging mode, the fetal heart rate is not measured until the fetal heart is within the desired target region, and in Doppler mode, said signal quality indicator may ensure sufficient quality for accurate measurement of the heart rate. The quality threshold is calibrated such that at least when the fetal heart is within the target region, the signal quality is generally sufficient. However, it is possible to have a stricter definition of the target region, for example, the signal quality may be sufficient even outside the target region. However, the opposite situation, where the fetal heart is within the target region but the signal quality is insufficient, is usually undesirable.

[0020] In either case, the quality threshold is preferably calibrated so that it is approximately in the region captured by the ultrasound image, including, for example, a safety margin. In this way, when the threshold is reached, it is possible to automatically switch to an imaging mode for recognizing the fetal heart, and possibly automatically switch back to Doppler mode, without the need for repositioning the probe, for example using an updated position where the Doppler ultrasound beam is focused by beam steering.

[0021] Optionally, the signal quality indicator may be output in a manner perceptible to the user. In this way, the user is informed whether the probe is in a good position on the abdomen, and the user can attempt to move the probe to improve the signal quality. The use of a visual indicator on the ultrasound probe itself, for example one or more lights, is particularly convenient for the user.

[0022] Optionally, the ultrasound probe comprises one or more transducers in the form of piezoelectric transducers and / or capacitive ultrasonic transducers (CMUTs), for example forming a phased array of multiple transducers, for example a linear phased array, allowing for the generation of 2D ultrasound images, or also forming a grid of transducers, for example a matrix array, allowing for the generation of 3D ultrasound images.

[0023] Optionally, the Doppler ultrasound beam may be focused on the placed location. In other words, a window for Doppler measurements, including depth, can be set according to the placed location. Various ultrasound probes, particularly those based on transducer arrays, support the use of beamforming to focus the Doppler ultrasound beam on a specific location. To determine the location of the fetal heart, focusing is particularly advantageous in combination with the use of object detection, since the ultrasound beam is focused on that specific location. For example, the fetal heart does not need to be placed in a cylinder directly under the ultrasound probe, as in some prior art cases. This makes the ultrasound probe particularly easy to use, for example with little or no training, since less precision is required for the placement of the ultrasound probe.

[0024] Many advantages are provided compared to using depth selection. Focusing the Doppler ultrasound beam can include setting not only the depth but also the horizontal and vertical positions perpendicular to this depth. The advantage is that only setting the depth does not address the situation where there are multiple signal sources in this depth region. Furthermore, whereas depth selection requires a good initial signal to focus on the depth, the provided technique can provide guidance information even when there is no fetal heart in the imaged region. The provided technique also improves on the slow and iterative window setting of depth selection, and does not have the problem of the window being completely open when the signal-to-noise ratio deteriorates.

[0025] Optionally, the ultrasound image and / or the location of the fetal heart in this ultrasound image is shown on the display. For example, the ultrasound image or a graphical image of the maternal abdomen, e.g., an avatar, may be shown with a box, cross or the like used to indicate the detected location. This can serve several purposes. It can aid in ultrasound probe guidance, as the user can see on the screen where the fetal heart is located relative to the image produced by the ultrasound probe and also relative to the probe itself.

[0026] Optionally, for example, showing a graphical image instead of the ultrasound image itself may allow easier interpretation and is also preferred when, for example, for legal reasons, the user is not allowed to see and / or interpret the ultrasound image itself, or when, for example, the mother does not want this image to be seen. The ultrasound image is processed in the background, but is not stored or displayed to the user. Optionally, a target area for switching to Doppler mode is also visualized, providing guidance to the user when switching to Doppler mode is anticipated.

[0027] When showing the ultrasound image itself, displaying the location where the fetal heart was detected can also be used as a reliability mechanism by allowing the user to check whether the system has correctly identified the fetal heart. For example, when the system switches to Doppler mode and / or uses the identified fetal heart to focus the Doppler ultrasound beam, the user can verify whether this is based on a correct identification of the heart.

[0028] Optionally, the object detection model is configured to localize one or more fetal body parts of the fetus in addition to the fetal heart. These body parts may include the fetal head, the fetal spine and / or the fetal femur. The fact that an ultrasound image is used instead of a Doppler signal allows these body parts to be recognized as well. Outputting these additional localizations is advantageous, as this information can make it easier to correctly position the ultrasound probe, especially in situations where the fetal heart itself cannot be seen in the image.

[0029] One way to output these additional localizations is to show them on a display, for example an ultrasound image can be shown along with the localizations of any fetal body parts that the object detection model is configured to localize. These localizations can help a user, particularly an inexperienced user such as a nurse or midwife, to determine the orientation of the fetus and thereby position the ultrasound probe for proper operation.

[0030] Another way of using the localization of at least the fetal heart, but possibly also the localization of any other fetal body parts that the object detection model localizes, is to determine guidance information for guiding the ultrasound probe to the fetal heart. This guidance information can be output to a display. Interestingly, this guidance information can suggest moving the probe to a different location on the maternal abdomen, rather than simply suggesting a rotation or change in angle of the probe relative to the maternal abdomen.

[0031] In particular, based on localizing one or more of the fetal head, fetal spine, and fetal femur. In particular, if the fetal spine is detected but not the fetal head, guidance information can be determined for guiding the ultrasound probe to the expected position of the fetal head. This has the advantage that the fetal spine can be oriented in any position in the uterus, making successful placement less dependent on the initial placement of the probe. Furthermore, the position of the fetal spine provides very useful information for the overall anatomy of the fetus. Given the position of the spine, the position of the fetal heart can be suggested and placement of the ultrasound probe to capture the fetal heart is possible. Locating the fetal head is also useful because this fetal head can be recognized with a fairly high degree of reliability, and guidance to the fetal heart can be suggested with a particularly high degree of reliability when the ultrasound probe is placed at the fetal head position and both the fetal head and fetal spine are detected. Guidance information for guiding the ultrasound probe to the expected position of the fetal heart is then determined.

[0032] To determine the guidance information, for example, an algorithm may be used that internally reconstructs the fetal geometry based on said localization. It is also possible to use a machine learning enabled guidance model, for example, trained to output guidance based on a labeled dataset.

[0033] The guidance information can be shown in various ways, for example visualized as suggested positions on a grid, e.g., a 3x3 grid or the like, that is extended onto the image of the maternal abdomen. Examples are given herein. The guidance information can be combined with a determined fetal anatomical structure visualization and / or an actual image from an ultrasound probe, if desired. The guidance information may be displayed without displaying (and possibly storing) an actual ultrasound image, which, as mentioned elsewhere, can make the visualization easier to interpret and is useful in situations where displaying an ultrasound image is undesirable.

[0034] Guidance based on localization of body parts is a particularly effective guidance mechanism with the added advantage that the overall position of the fetus can be determined, and not just the position of the fetal heart itself. Moreover, this kind of guidance is familiar to people accustomed to clinical practice, where the person placing the ultrasound probe typically follows a protocol in which the clinician palpates the maternal abdomen according to a certain pattern, for example, upper → lower → left → right, in order to reveal the fetal anatomy. Protocols as used today can take 15-20 minutes in practice, even for experienced people. The guidance techniques provided allow to significantly reduce this time, allowing the probe to be used even by inexperienced or unqualified users (nurses, midwives) during antenatal care. The system, upon detecting the spine, aims to guide the probe first to the head and then to the heart, but in the meantime, it is possible to switch directly to Doppler mode on detecting this, if a sufficiently reliable heart has already been detected. However, it is also possible to detect only the fetal heart and / or switch to Doppler mode after both the spine and the head have been detected, for example, in order to allow a complete determination of the fetal orientation before switching modes.

[0035] Optionally, the orientation of the ultrasound probe is used to determine the guidance information. For example, the ultrasound probe may have an inertial measurement unit (IMU) that provides an orientation angle of the ultrasound probe. This orientation may be input to an object detection model to determine the position of the fetal heart and / or other body parts and may be used to determine the guidance information. Interestingly, given the sphere-like shape of the maternal abdomen, the orientation of the ultrasound probe can provide a fairly accurate estimate of the position of the ultrasound probe on said abdomen, and is used not only as an input representing the current position of the probe but also as an input to the object detection model to determine the guidance, among other things.

[0036] The above aspects relate to systems and methods that use machine-learnable object detection and / or guidance models. Systems for training object detection and / or guidance models for these uses, as well as corresponding computer-implemented methods, are also envisioned. The training can be performed using techniques known per se. For example, gradient-based training, such as stochastic gradient descent, can be performed using an Adam optimizer as disclosed in Kingma and Ba, “Adam: A Method for Stochastic Optimization” (available at https: / / arxiv.org / abs / 1412.6980, incorporated herein by reference). As is known, such optimization methods are heuristic and / or reach local optima.

[0037] It will be appreciated by those skilled in the art that two or more of the above-described embodiments, implementations and / or optional aspects of the invention may be combined in any manner deemed useful.

[0038] Modifications and variations of any of the systems and / or any of the computer readable mediums correspond to the described modifications and variations of the corresponding computer implemented methods and may be implemented by those skilled in the art based on this specification. [Brief description of the drawings]

[0039] These and other aspects of the invention will become more apparent and will be elucidated with reference to embodiments thereof, illustrated by way of example only with reference to the following description and the accompanying drawings, in which: [Figure 1] FIG. 1 shows an ultrasound fetal heart rate monitoring system. [Figure 2A] FIG. 2A shows a detailed example of a method for monitoring fetal heart rate. [Figure 2B] FIG. 2B shows a detailed example of how an ultrasound probe can be switched from imaging mode to Doppler mode. [Diagram 3]FIG. 3 shows a detailed example of locating the fetal heart. [Figure 4A] FIG. 4A shows a detailed example of determining guidance information. [Figure 4B] FIG. 4B shows a detailed example of determining guidance information. [Figure 4C] FIG. 4C shows a detailed example of determining guidance information. [Diagram 5] FIG. 5 illustrates a computer-implemented ultrasound fetal heart rate monitoring method. [Figure 6] FIG. 6 illustrates a computer readable medium having data.

[0040] It should be noted that the figures are merely schematic and are not drawn to scale. In the drawings, elements that correspond to elements already described may have the same reference numbers. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0041] FIG. 1 shows an ultrasound fetal heart rate monitoring system 100 .

[0042] The system 100 comprises a data interface 120 for accessing model data 030 representing a trained object detection model. The object detection model is configured to localize the fetal heart in an ultrasound image. The model data 030 may comprise a set of trained parameters, for example at least 10000 or at least 100000 parameters. For example, the model may be a neural network. A neural network is also known as an artificial neural network. In particular, the model may be a deep neural network or a convolutional neural network. In case of a neural network, the set of parameters 030 comprises weights of the nodes of the neural network. For example, the number of layers of the model may be at least 5 or at least 10 and the number of nodes and / or weights may be at least 1000 or at least 10000. The data interface 120 is for accessing additional information as described herein, for example model data representing a trained guidance model.

[0043] The model data representing the object detection model and / or the guidance model is obtained, for example, by training the model by system 100 itself or by another system having a similar hardware architecture, in particular by a system having the data interface and processor system described herein.

[0044] For example, as also shown in FIG. 1, the data interface 120 may be constituted by a data storage interface that can access the data 030 from a data storage 021. For example, the data storage interface 120 may be a memory interface or a persistent storage interface, such as a hard disk or SSD interface, but also a personal LAN or WAN network interface, such as a Bluetooth, Zigbee or Wi-Fi interface, or an Ethernet or fiber optic interface. The data storage 021 may be an internal data storage of the system 100, such as a hard drive or SSD, but also an external data storage, such as a network-accessible data storage. In some embodiments, the respective data are each accessed from different data storages, for example, via different subsystems of the data storage interface 120. Each subsystem may be of the type as described above for the data storage interface 120.

[0045] The system 100 may further include a sensor interface 160 for acquiring sensor data 124 from an ultrasound probe 071. The ultrasound probe 071 is configured to be placed on the maternal abdomen. The ultrasound probe 071 is operable in an imaging mode and a Doppler mode. Typically, the ultrasound probe 071 operates in an imaging mode or a Doppler mode, but not in both modes simultaneously. The ultrasound probe 071 includes one or more transducers. For example, the transducer may be a piezoelectric transducer, or the transducer may be a CMUT transducer. The use of a CMUT transducer is preferred due to its lower manufacturing cost.

[0046] The transducers may be arranged in a configuration suitable for 2D and / or 3D ultrasound imaging phased arrays, for example, a 1D or 2D phased array configuration. For example, the number of transducers in a phased array can be at least 2, at least 10, or at least 50. For example, the transducers may be arranged in a 1D phased array configuration, for example, a linear phased array having at least 2, at least 10, or at least 50 transducers. As another example, the transducers may be arranged in a 2D phased array configuration, for example, a matrix phased array. A matrix phased array can have at least 10 or at least 50 transducers in both directions, for example, arranged in a rectangular pattern. Such a transducer pattern, while used for imaging, can also be used in Doppler mode, for example, by using one specific transducer to generate one Doppler signal, or using multiple individual transducers to generate multiple individual Doppler signals. In particular, the transducers are typically not arranged in a circular array of piezoelectric elements with one piezoelectric element in the center, as this arrangement does not work well for imaging. The ultrasound probe 071 may, for example, operate at a frequency of at least 1 MHz and / or at most 5 MHz.

[0047] In the imaging mode, the ultrasound probe 071 can provide sensor data 124 representing an ultrasound image. The ultrasound image can represent the amplitude of the echoes of the ultrasound signal transmitted by the transducer. In particular, the imaging mode can be the so-called B-mode. For example, a one-dimensional phased array can be used to obtain a two-dimensional image. It is also possible to obtain a three-dimensional image using a two-dimensional phased array. It is also possible to use a one-dimensional phased array to obtain a three-dimensional image, as disclosed, for example, in M. N. Senlik and H. Koymen, "Radiation Impedance of an Array of Circular Capacitive Micromachined Ultrasonic Transducers", IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 57, no. 4, April 2010.

[0048] In Doppler mode, a Doppler ultrasound signal is obtained that represents the velocity of the measurement volume. For example, the Doppler mode may be a pulsed (PW) Doppler mode, in which ultrasound waves are alternately transmitted and received, making it possible to measure the velocity at a specified depth. The Doppler ultrasound beam is focused at a specific location by setting the direction at which the signal is transmitted, for example by using beamforming. It is also possible to use a continuous wave (CW) Doppler mode. This mode typically does not allow the selection of a specific depth, although the direction of the signal is still controlled.

[0049] Typically, imaging mode and Doppler mode both operate with the same set of transducers but not simultaneously, since each mode uses a different ultrasound signal. For example, in a B-type imaging mode, a beam can be transmitted by each transducer with a respective phase difference to scan in a particular direction. In a Doppler mode, for example, a pulse can be transmitted by only one transducer or the same pulse can be transmitted by multiple transducers at the same time. Thus, the ultrasound probe 071 is configured to operate in an imaging mode or a Doppler mode via a control signal 124 transmitted from the system 100 to the ultrasound probe 071 via the sensor interface 160.

[0050] The Doppler ultrasound beam may be focused on a particular location, specifically where the fetal heart is located by an object detection model as described herein. The beam is focused by beamforming, which typically involves signal processing performed, for example, in the ultrasound probe 071 itself or by the processor subsystem 140.

[0051] The ultrasound probe 071 may further include an orientation sensor, e.g., an IMU sensor. The processor subsystem 140 is configured to obtain, via the sensor interface 160, orientation data 124 indicative of the orientation of the ultrasound probe, e.g., having an angle of the sensor 071. This orientation data is used to determine guidance information, as described herein.

[0052] The system 100 further includes a processor subsystem 140 configured to operate the ultrasound probe 071 in an imaging mode to acquire an ultrasound image 124 from the ultrasound probe 071 during operation of the system 100. The processor subsystem 140 is further configured to apply the object detection model 030 to the ultrasound image 124 to determine whether the fetal heart is located within a target region of the ultrasound image addressable by a Doppler mode of the ultrasound probe 071. If the fetal heart is located within the target region, the processor subsystem 140 is further configured to switch the ultrasound probe 071 to a Doppler mode, acquire a Doppler ultrasound signal 124 from the ultrasound probe, and calculate a fetal heart rate from the Doppler ultrasound signal 124.

[0053] The system 100 may further include an output interface 180. The output interface 180 is used for various outputs including a fetal heart rate calculated from the Doppler ultrasound signal, an ultrasound image, a location where the fetal heart is located in the ultrasound image, a signal quality indicator of the Doppler ultrasound signal, and / or guidance information. For example, the output interface may be an output interface to a rendering device, such as a display, a light source, a speaker, a vibration motor, etc., which may be used to generate a perceptible output signal generated based on the output information. For example, as shown, the output interface may be for a display 190. In the Doppler mode, for example, the fetal heart rate 192 is displayed and / or a warning is issued if the fetal heart rate is abnormal and / or a signal quality indicator of the Doppler ultrasound signal is output. In the imaging mode, for example, an ultrasound image, a location of the located fetal body part, and / or guidance information are shown. As another example, the output interface may be constituted by data interface 120, which in these embodiments is an input / output ("IO") interface, through which an output, e.g., fetal heart rate, is stored in data storage 021. This output interface may also provide an output, such as the determined fetal heart rate, for further processing, e.g., the fetal heart rate may be provided to a fetal heart rate monitoring module operated by processor subsystem 140.

[0054] The processor subsystem 140 may further be configured to determine a signal quality indicator of the Doppler ultrasound signal 124. If this signal quality indicator does not meet a predefined quality threshold, the ultrasound probe 071 is returned to an imaging mode via the control signal 124. Instead of or in addition to using the signal quality indicator to control the mode of the ultrasound probe, the processor subsystem 140 may be configured to output the signal quality indicator in a manner perceptible to a user. This may be output via the output interface 180, e.g., on the display 190, and / or via the ultrasound probe's visual indicator 072, e.g., by one or more lights, e.g., using the color, intensity and / or number of lights of the visual indicator 072 used to indicate the signal quality.

[0055] The system 100, the ultrasound sensor 071 and / or the display 190 can be arranged in various ways. For example, the system 100 and the probe 071 form a single device, e.g., the processor subsystem 140, which applies an object detection model to determine the fetal heart rate, is included in the ultrasound sensor 071 or is fixedly connected to the ultrasound sensor 071 or the sensor 071. The system 100 can also be implemented as a fetal monitor (connectable or connected to the probe 071), e.g., with a built-in or externally connectable display 190. The system 100 can also be implemented on the cloud, e.g., data is provided to a mobile device that shows the fetal heart rate on the display 190. The system 100 itself may be located on the mobile device, e.g., by providing its functionality as a mobile application.

[0056] More generally, each system described herein, including but not limited to the system 100 of FIG. 1, can be embodied as or in a single device or apparatus, such as a workstation or server. The device or apparatus may be an embedded apparatus. The device or apparatus may have one or more microprocessors executing appropriate software. For example, the processor subsystem of each system may be embodied by a single central processing unit (CPU), but may also be embodied by a combination or system of such CPUs and / or other types of processing units. Software may be downloaded and / or stored in a corresponding memory, e.g., a volatile memory such as a RAM, or a non-volatile memory such as a flash memory. Alternatively, the processor subsystem of each system may be implemented in the form of programmable logic in the device or apparatus, e.g., as a field programmable gate array (FPGA). In general, each functional unit of each system may be implemented in the form of a circuit. Each system may be implemented in a distributed manner, including different devices or apparatus, such as distributed local servers or cloud-based servers.

[0057] System 100 may alternatively or additionally be a training system for training object detection models. Such a training system need not have sensor interface 160 and / or output interface 180, but may, for example, use data interface 120 to access training data and / or models being trained.

[0058] 2A shows a detailed, but non-limiting example of a method for monitoring fetal heart rate using an ultrasound fetal heart rate monitoring system as described herein. The figure shows a general flow chart illustrating several operations performed to detect the fetal heart and plot the fetal heart rate in real time, for example, on a mobile application. According to the figure, the fetal heart is automatically detected from a 2D ultrasound scan frame in real time, and once the fetal heart is detected, a pulse wave mode can be entered from the 2D image mode to plot the fetal heart rate.

[0059] In a positioning operation POS 210, an ultrasound probe of a fetal heart rate monitoring system may be positioned on the patient's abdomen, for example by a midwife.

[0060] In a fixation operation FIX 220, the ultrasound probe may be repositioned on the abdomen in order to fix the position of the ultrasound probe and / or to obtain a trace. This can be done, for example, by a midwife.

[0061] In the detection operation DET230, the fetal heart is detected and displayed, for example, on a screen connected to a fetal heart rate monitoring system. Single / multiple piezoelectric / CMUT sensors can be used to capture 2D ultrasound data from the initial position. An object detection algorithm can detect fetal anatomical structures (e.g., fetal heart, fetal head, fetal spine, and / or fetal femur) on the ultrasound image acquired from the initial position and guide the user to the next position, for example, by suggesting a position to move the probe. An IMU sensor may be used to provide an algorithm to detect the fetal heart at the angle of the sensor. If the fetal heart is still not detected, the process may return to a fix operation FIX to fix the ultrasound probe. The system may further repeat the detection operation DET. Optionally, as described elsewhere, if the fetal heart is still not detected, information is provided to the user to help place the ultrasound probe in the correct position, for example, in the form of the ultrasound probe type and / or the position of the placed fetal organs, and / or guidance information.

[0062] Once the fetal heart is detected, the probe switches to a Doppler mode and the fetal heart rate can be continuously calculated. In particular, in the monitoring operation MON240, the Doppler mode is enabled and the fetal heart rate monitoring system can calculate the fetal heart rate from the Doppler ultrasound signal. While in the Doppler mode, a signal quality indicator of the Doppler ultrasound signal can be determined as known per se. The signal quality indicator may be output to a user, for example, in the form of one or more lights or another visual indicator on the ultrasound probe itself.

[0063] In an interpretation operation INT250, the tracing and / or fetal heart rate are interpreted, for example by a midwife and / or gynaecologist.

[0064] In the out-of-view operation OOV260, the fetal heart is determined to be out of view, e.g. the signal quality of the Doppler ultrasound signal is insufficient to reliably determine the fetal heart rate, e.g. does not meet a predefined quality threshold. In this case, a fixation operation FIX of the ultrasound probe can be performed again, e.g. the probe can be returned to the imaging mode. Interestingly, the switch to the imaging mode allows the user to help refocus on the fetal heart so that the monitoring of the fetal heart rate continues. The quality threshold is defined such that at this quality threshold the fetal heart is still approximately within the range of the ultrasound image captured in the imaging mode. In this way, refocusing is relatively easy, since the fetal heart is indicated and / or guidance information based on the position of the fetal heart is determined when the signal quality falls below the threshold.

[0065] FIG. 2B shows a detailed, but non-limiting, example of how an ultrasound probe can be switched from imaging mode to Doppler mode.

[0066] In an acquisition operation Aq221, the ultrasound probe operates in an imaging mode to acquire an ultrasound image UI222 from the ultrasound probe. The ultrasound image may be, for example, at least 32x32 pixels, at least 128x128 pixels, or at least 512x512 pixels. The image may be grayscale.

[0067] In a model application operation Appl223, an object detection model OD224 is applied to the ultrasound image UI to determine localization information LOC225 indicating whether the fetal heart is located within a target region of the ultrasound image that can be addressed by the Doppler mode of the ultrasound probe.

[0068] As object detection models are known per se, the object detection model OD can output localization information LOC, for example in the form of class probabilities corresponding to one or more types of objects to be detected, in combination with the location of the recognized object, for example the position and optionally the size of the recognized object, the location being represented for example as one or more coordinates in an ultrasound image, a bounding box, etc. Various object detection models are known per se and can be used.

[0069] In one embodiment, the object detection model comprises a neural network, in particular a deep neural network. In particular, the inventors of the present application have achieved good results using the YOLO object detection model by J. Redmon et al., "You Only Look Once: Unified, Real-Time Object Detection" (available at https: / / arxiv.org / abs / 1506.02640 and incorporated herein by reference). YOLO is a region proposal network that applies a single neural network to the entire image to detect a particular object. The network divides the image into regions and predicts a bounding box and probability for each region. YOLO divides the input image into an S×S grid. If the center of an object falls within a grid cell, that grid cell is responsible for detecting the object. Each grid cell predicts B bounding boxes and a confidence score for these boxes. The confidence score reflects whether the box really contains the object of interest and also reflects the accuracy of the prediction. Each bounding box has five predictions x, y, w, h and a confidence. The (x,y) coordinates represent the center of the box relative to the grid cell boundary. The width (w) and height (h) are predicted for the entire image. The confidence level of the YOLO output is a regression trained to output the intersection of union (IoU) between the output bounding box and the ground truth bounding box. Other models based on the same principle can also be used. As a concrete example, the YOLOv3-tiny model can be applied to a 416x416 image.

[0070] Specifically, the object detection model OD may be trained and applied on 2D fetal heart four-chamber images. The positive training examples may include axial views of the fetal heart, e.g., collected and curated from a cine loop of 2D fetal scans. Planes that do not include the fetal heart are used as negative training examples. A training dataset may be created using these images by labeling them as fetal heart images, annotating them with a bounding box of the fetal heart, or labeling them as non-fetal heart images. These images may be pre-processed to remove text annotations, if necessary. Annotation may be performed, for example, using the DarkLabel tool. Training may be performed using the Darknet deep learning framework. The inventors of the present application have found that training for 4000 iterations with a batch size of 64 images produces good results.

[0071] For example, a full training-testing process may include starting with data, annotating, cleaning the annotated data, training data, e.g., based on patient ID, splitting into validation and test data, training multiple models on the training data by varying hyperparameters, validating the models on the validation data, finalizing the model with the highest validation accuracy, testing the final model on unknown data, and calculating evaluation metrics.

[0072] Based on the localization information LOC, it can be determined whether the fetal heart is located within a target area of ​​the ultrasound image that can be handled by the Doppler mode of the ultrasound probe. For example, this is the case when the fetal heart is recognized by the model with at least a given threshold confidence and the location where the fetal heart is recognized is within the target area. The target area can correspond, for example, to the ultrasound image UI itself or to a sub-area thereof. If the fetal heart is located within the target area, the ultrasound probe can be switched to a Doppler mode Sw227 and start monitoring the fetal heart rate, as discussed elsewhere, and optionally use the localization information LOC to focus the Doppler ultrasound beam on the located position. The acquisition operation Aq and subsequent steps may be repeated as long as the fetal heart is not found to be within the target area. Furthermore, the localization information LOC may be used in various ways, in particular the location may be output and / or used to determine guidance information to help the user guide the ultrasound probe to the fetal heart, as discussed in more detail elsewhere.

[0073] FIG. 3 shows a detailed, but non-limiting, example of locating the fetal heart. In this example, an object detection model is used to locate the fetal heart in an ultrasound image 310 by outputting a bounding box 320 within which the fetal heart is detected. The ultrasound image 300 and the bounding box 320 indicating the location of the fetal heart in the ultrasound image 310 are shown in this example in a user interface 300 shown on a display. As shown, the object detection model can optionally further output a confidence value of the detection, which is also shown on the display, for example, as a label (0.985 in this example), as a color or line type of the bounding box, etc. If the fetal heart is not detected, for example, the confidence value does not exceed a threshold, then the bounding box is typically not shown. The object detection model is configured to detect additional anatomical structures of the fetus, for example, one or more of the fetal head, fetal spine, and fetal femur. If detected, the localization for these additional anatomical structures is also shown on the ultrasound image 310.

[0074] 4A-4C show detailed, but non-limiting, examples of determining guidance information for guiding an ultrasound probe to the fetal heart.

[0075] For example, as discussed with respect to FIG. 2B, the guidance information is determined from the position of one or more fetal body parts recognized by the object detection model in the ultrasound image by the ultrasound sensor while the ultrasound sensor is in the imaging mode. The guidance information is further based on the orientation of the ultrasound probe determined by the IMU sensor or other orientation sensor. Interestingly, based on the orientation of the IMU sensor and using the geometry of the abdomen, the position of the IMU sensor on this abdomen can be determined, for example, if the IMU sensor is oriented vertically, the IMU sensor is placed in the center of the abdomen, if tilted to the left, the IMU sensor is placed on the left side of the abdomen, etc. Other methods of determining the position of the ultrasound sensor on the abdomen, for example, based on the ultrasound image, are also possible and can be used to determine the guidance information as well.

[0076] 4A-4C show fetal heart rate monitoring systems 410, 420, 430 with displays on which guidance information is displayed in the form of suggested positions 411, 412, 413 to which the ultrasound probe should be moved. In this example, the suggested positions are determined at a granularity of a suggested cell in a grid representing the abdomen. In this particular example, a 3x3 grid is used. The positions are highlighted on a visualization of the maternal abdomen where this location is magnified. This visualization can be a standard image or diagram of the maternal abdomen. Thus, the guidance can be used without the need to store or display the ultrasound image used to determine this guidance, which is beneficial as mentioned elsewhere. The highlighting of the positions may indicate the progress of the guidance. For example, on the fetal monitor shown in FIGS. 4A-4C, the color of the highlighted positions 411-431 in the examples of FIGS. 4A and 4B indicates that the fetal heart has not yet been detected, whereas the color of the highlighting in FIG. 4C indicates that the fetal heart has been detected. In visualizations 440, 450, 460, alternative highlighting of locations in the form of border styles of proposed grid cells to move the probe to is shown. As shown, the borders of the cells can have a different color and / or a thicker border than other cells.

[0077] The visualizations 440-460 also show that the estimated position of the fetus in the abdomen is visualized. This can be based on a standard image or illustration of the fetus. For example, the entire fetus can be visualized, as illustrated in FIG. 460, or a portion of the fetus can be shown, as illustrated in FIG. 440, 450, which corresponds to a known position of the fetal anatomy, for example based on the position to which the ultrasound probe has been moved and / or previously located fetal anatomy. An exemplary guidance procedure is illustrated. Using this guidance procedure, a user, e.g., a care provider, can position the probe towards the fetal heart much faster (reducing the need for manual palpation due to automatic object detection) while following familiar existing protocols.

[0078] The guidance may begin by suggesting an initial location for placing the ultrasound probe. As shown in Figure 4A, this initial location is, for example, the left central part of the abdomen. This location is often used in current protocols, as it typically allows for detection of the fetal spine, allowing for easier extrapolation of the remaining anatomical structures.

[0079] If the fetal spine is not detected, alternative initial positions may be suggested, for example, right center, top center, bottom center, etc.

[0080] If the fetal spine is detected but the fetal head is not yet detected, guidance information is determined to guide the ultrasound probe to a predicted location of the fetal head, for example, the predicted location of the fetal head is determined under the assumption that the fetus is positioned head down, resulting in the proposed position 421 of FIG.

[0081] If the fetal head is not detected, an alternative likely position is determined and suggested, for example based on the assumption that the fetus is positioned head-up.

[0082] When the fetal spine and fetal head are detected, guidance information can be determined to guide the ultrasound probe to the expected location of the fetal heart. If the fetal spine and fetal head are located, the expected location of the fetal heart can be accurately determined based on normal fetal geometry. This is illustrated by proposal 431 in FIG. 4C, where it is proposed that the fetal heart be placed in the center cell of the grid.

[0083] If the fetal heart is not detected, an alternative location for the fetal heart can be determined or the process can be restarted, for example, from detection of the spine.

[0084] If the fetal heart is detected, the system can switch to Doppler mode to monitor the fetal heart as described. This can occur while the guidance is still searching for the fetal spine or fetal head. In this case, it is possible to switch directly to Doppler mode, but if desired, it is also possible to continue the guidance process first to also locate the fetal spine and fetal head and determine the complete position and orientation of the fetus. This information can be output, for example, to a user, or can be used automatically to predict the angle at which the fetal head will progress during labor and / or birth, for example, using a trained machine learning model.

[0085] To determine the guidance information, several approaches are possible. One possibility is to use an explicit geometric model of the fetal anatomy. This model is relatively simple, for example the fetus is modeled as a straight line representing the spine, terminating in a sphere representing the fetal head. For the determined positions of the line and the sphere the position of the fetal heart is determined. An alternative is to use a machine learning capable guidance model trained to output guidance information, for example outputting cells in the grid of Figs. 4A-4C. For example, this guidance model has respective models that are used to determine the position of the fetal spine, fetal head and fetal heart, respectively, in a step subsequent to the fetal heart localization. Another option is to apply a trained feature extractor to extract one or more features from the ultrasound image UI and compare these extracted features to a trained fetal image dataset for which guidance information is available. For example, in this case a general feature extractor can be used.

[0086] FIG. 5 illustrates a block diagram of a computer-implemented method 500 for ultrasound fetal heart rate monitoring. Method 500 corresponds to operations of system 100 of FIG. 1. However, this is not a limitation and method 500 may be performed using another system, apparatus, or device. Method 500 may include acquiring 510 sensor data from an ultrasound probe placed on the maternal abdomen in an operation titled "OBTAIN SENSOR DATA." Method 500 may include accessing 520 model data representing a trained object detection model in an operation titled "ACCESS OBJECT DETECTION MODEL." The object detection model is configured to locate the fetal heart in an ultrasound image. Method 500 may include operating 530 an ultrasound probe in an imaging mode to acquire an ultrasound image from the ultrasound probe in an operation titled "ULTRASOUND IMAGING." The method 500 may include applying an object detection model to determine 540 whether the fetal heart is located in a position of the ultrasound image that can be addressed by the Doppler mode of the ultrasound probe in an operation titled "DETECT HEART". The method 500 may include determining 550 whether the fetal heart is located within a target region in an operation titled "HEART DETECTED?". The method 500 may include switching 560 the ultrasound probe to Doppler mode if the fetal heart is detected in an operation titled "SWITCH TO DOPPLER". The method 500 may further include acquiring 570 a Doppler ultrasound signal from the ultrasound probe in an operation titled "OBTAIN DOPPLER SIGNAL". Method 500 may further include step 580 of calculating a fetal heart rate from the Doppler ultrasound signal in an operation entitled "MONITOR FHR." It will also be appreciated that, in general, the operations of method 500 of FIG. 5 may be performed in any suitable order, e.g., sequentially, simultaneously, or combinations thereof, subject to, e.g., input / output relationships necessitating a particular order, where applicable. The method may be combined with further steps, e.g., the object detection model and / or guidance model described herein may be trained before being applied. Separate computer-implemented methods of training the object detection model and / or guidance model are also contemplated.

[0087] The method may be implemented on a computer as a computer-implemented method, as dedicated hardware, or as a combination of both. As also shown in Fig. 6, instructions for a computer, e.g., executable code, are stored on a computer-readable medium 600, e.g., in the form of a series of machine-readable physical marks 610 and / or as a series of elements, e.g., with different electrical, magnetic, or optical properties or values. The medium 600 may be transitory or non-transitory. Examples of computer computer-readable media include memory devices, optical storage devices, integrated circuits, servers, online software, etc. Fig. 6 shows an optical disk 600. Alternatively, the computer-readable medium 600 has data 610 representing a trained object detection model and / or guidance model for use with the techniques described herein.

[0088] Examples, embodiments, or optional features, whether non-limiting or not, should not be understood as limiting the claimed invention.

[0089] It should be noted that the above-described embodiments are illustrative rather than limiting of the invention, and that those skilled in the art can design many alternative embodiments without departing from the scope of the appended claims. Any reference signs placed between parentheses in the claims shall not be construed as limiting the claims. Use of the verb "comprise" and its conjugations does not exclude the presence of elements or steps other than those stated in the claims. The absence of a plurality of elements does not exclude a plurality of the elements. The phrase "at least one of" after a list or set of elements indicates the selection of all or any part of the elements from that list or set. For example, the phrase "at least one of A, B, and C" is to be understood as including A only, B only, C only, both A and B, both A and C, both B and C, or all of A, B, and C. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In a device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. 1. An ultrasound fetal heart rate monitoring system, comprising: a sensor interface for acquiring sensor data from an ultrasound probe placed on the maternal abdomen; a data interface for accessing model data representing a trained object detection model, the object detection model configured to localize a fetal heart in an ultrasound image; and Processor Subsystem the processor subsystem comprising: operating the ultrasound probe in an imaging mode to acquire ultrasound images from the ultrasound probe; applying the object detection model to the ultrasound image to determine whether a fetal heart is located within a target region of the ultrasound image that is addressable by a Doppler mode of the ultrasound probe; and If the fetal heart is located within the target region, switch the ultrasound probe to the Doppler mode, acquire a Doppler ultrasound signal from the ultrasound probe, and calculate a fetal heart rate from the Doppler ultrasound signal. It is configured as follows: the processor subsystem is further configured to determine guidance information for guiding the ultrasound probe to the fetal heart from the positions of one or more fetal body parts located by the object detection model, and output the determined guidance information to a display. Ultrasound fetal heart rate monitoring system.

2. The system of claim 1 , wherein the system is configured to operate the ultrasound probe in the imaging mode or the Doppler mode, but not in both modes simultaneously.

3. 3. The system of claim 1, wherein the processor subsystem is further configured to determine a signal quality index of the Doppler ultrasound signal and return the ultrasound probe to the imaging mode if the signal quality index does not meet a predetermined quality threshold.

4. 4. The system of claim 3, wherein the processor subsystem is configured to output the signal quality indicator in a manner perceptible to a user, preferably by outputting the signal quality indicator to a visual indicator on the ultrasound probe.

5. The system of claim 1 , wherein the ultrasonic probe comprises one or more piezoelectric transducers and / or one or more capacitive mechanical ultrasonic transducers (CMUTs).

6. The system of claim 1 , wherein the processor subsystem is further configured to focus the Doppler ultrasound beam on the target region.

7. The system of claim 1 , further comprising an output interface to a display, the processor subsystem configured to indicate the location of the fetal heart in the ultrasound image on the display.

8. 8. The system of claim 7, wherein the object detection model is further configured to localize one or more of the fetal head, the fetal spine, and the fetal femur, and the processor subsystem is configured to indicate the localization on the display.

9. 2. The system of claim 1, wherein the object detection model is configured to localize one or more of the fetal head, the fetal spine, and the fetal femur, and the processor subsystem is configured to determine the guidance information using the localization.

10. The processor subsystem: If the fetal spine is detected and the fetal head is not detected, determining guidance information for guiding the ultrasound probe to an expected location of the fetal head; and / or If the fetal spine and the fetal head are detected, determining guidance information for guiding the ultrasound probe to an expected location of the fetal heart. The system of claim 9 , configured to:

11. The system of claim 1 , wherein the ultrasound image is not shown on the display.

12. 2. The system of claim 1, wherein the processor subsystem is further configured to acquire orientation data indicative of an orientation of the ultrasound probe via the sensor interface and to use the orientation data to determine a position of the fetal heart and / or the guidance information.

13. acquiring sensor data from an ultrasound probe placed on the maternal abdomen; accessing model data representing a trained object detection model, the object detection model configured to localize a fetal heart in an ultrasound image; operating the ultrasound probe in an imaging mode to acquire ultrasound images from the ultrasound probe; applying the object detection model to the ultrasound image to determine whether a fetal heart is located within a target region of the ultrasound image that is addressable by a Doppler mode of the ultrasound probe; and if the fetal heart is located within the target area, switching the ultrasound probe to the Doppler mode, acquiring Doppler ultrasound signals from the ultrasound probe, and determining the fetal heart rate from the Doppler ultrasound signals. wherein the method comprises: determining guidance information for guiding the ultrasound probe to the fetal heart from the positions of one or more fetal body parts located by the object detection model, and outputting the determined guidance information to a display; The computer-implemented ultrasound fetal heart rate monitoring method further comprises:

14. 14. A transitory or non-transitory computer readable medium having data representing instructions that, when executed by a processor system, cause the processor system to perform the computer-implemented ultrasound fetal heart rate monitoring method of claim 13.

Images