Non-invasive transabdominal fetal electroencephalography

Abstract

The examples described herein provide a computer implementation method that includes receiving non-invasive transabdominal fetal electroencephalography (TA-fEEG) signals related to a pregnant subject. The method further includes reducing unwanted noise in the TA-fEEG signals using a first machine learning model. The method further includes reconstructing fetal electroencephalography (fEEG) signals from the TA-fEEG signals using a second machine learning model.

Description

Technical Field

[0004] , , , , ,

[0003] , , ,

[0005] , ,

[0001] Cross-reference with related applications This application claims the benefit of priority based on U.S. Provisional Patent Application (filed on May 5, 2023, application number 63 / 464,341) titled "Non-Invasive Transabdominal Fetal Electroencephalogram Measurement". The entire content of the U.S. Provisional Patent Application is incorporated herein by reference.

Background Art

[0002] Electroencephalogram measurement (EEG) involves measuring the electrical signals of a subject by placing electrodes on the subject. The electrodes measure the signals, and these signals can be presented as an electroencephalogram useful for monitoring and diagnostic purposes. One use of EEG is to measure the brain activity of a subject. EEG is often non-invasive, and the electrodes are placed on the skin of the subject. However, in some cases, the electrodes are surgically implanted into the body of the subject. For example, EEG can be used to monitor a fetus in utero using a direct scalp measurement method, which involves inserting a fetal scalp electrode through the birth canal and attaching it to the scalp of the fetus.

Summary of the Invention

[0003] In one embodiment, a computer-implemented method is provided. The method includes receiving non-invasive transabdominal fetal electroencephalogram (TA-fEEG) signals related to a pregnant subject. The method further includes reducing unwanted noise in the TA-fEEG signals using a first machine learning model. The method further includes reconstructing fetal electroencephalogram (fEEG) signals from the TA-fEEG signals using a second machine learning model.

[0004] In addition to or as an alternative to one or more features described herein, a further embodiment of the method may include that the non-invasive TA-fEEG signals are collected from sensors of a non-invasive sensing device associated with a pregnant subject.

[0005] In addition to or as an alternative to one or more features described herein, further embodiments of the method may include predicting the likelihood of fetal hypoxia in the fetus of a pregnant subject based at least partially on reconstructed fEEG signals.

[0006] In addition to, or as an alternative to, one or more features described herein, further embodiments of the method may include the first machine learning model being a first neural network and the second machine learning model being a second neural network.

[0007] In addition to or as an alternative to one or more features described herein, further embodiments of this method may include training a first machine learning model.

[0008] In addition to or as an alternative to one or more features described herein, further embodiments of the method may include the first machine learning model being trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with realistic volume conductor models of materials and fetal tissue developed for fetal electrocardiograms (ECG).

[0009] In addition to, or as an alternative to, one or more features described herein, further embodiments of this method may include training a second machine learning model.

[0010] In addition to or as an alternative to one or more features described herein, further embodiments of the method may include the second machine learning model being trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with realistic volume conductor models of materials and fetal tissue developed for fetal electrocardiograms (ECG).

[0011] In addition to, or as an alternative to, one or more features described herein, further embodiments of the method may include a first machine learning model being a deep neural network and a second machine learning model being an independent component analysis model.

[0012] In another embodiment, a system for non-invasive transabdominal fetal electroencephalography (TA-fEEG) comprises a non-invasive sensing device having a sensor for detecting TA-fEEG signals from a pregnant subject, and a processing system for communicating with the sensor. The processing system has a memory for storing computer-readable instructions and a processing unit for executing computer-readable instructions. Computer-readable instructions control the processing unit to perform a predetermined operation. The operation includes removing artifacts from the TA-fEEG signal that are due to maternal and fetal cardiac activity and movement, using a first machine learning model. The operation further includes isolating independent sources of activity inherent in the TA-fEEG signal, using a second machine learning model. The operation further includes predicting the likelihood of fetal hypoxia in the fetus of a pregnant subject, at least partially based on at least one of the independent sources of activity inherent in the TA-fEEG signal.

[0013] In addition to, or as an alternative to, one or more features described herein, further embodiments of this system may include the first machine learning model being a first neural network and the second machine learning model being a second neural network.

[0014] In addition to or as an alternative to one or more features described herein, further embodiments of this system may include further training of a first machine learning model.

[0015] In addition to or as an alternative to one or more features described herein, further embodiments of this system may include the first machine learning model being trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with realistic volume conductor models of materials and fetal tissue developed for fetal electrocardiograms (ECG).

[0016] In addition to or as an alternative to one or more features described herein, further embodiments of this system may include training a second machine learning model.

[0017] In addition to or as an alternative to one or more features described herein, further embodiments of this system may include the second machine learning model being trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

[0018] In addition to, or as an alternative to, one or more features described herein, further embodiments of this system may include a first machine learning model being a deep neural network and a second machine learning model being an independent component analysis model.

[0019] In yet another embodiment, a computer implementation method for training a machine learning model is provided. This method includes training a first machine learning model to reduce unwanted noise in non-invasive transabdominal electroencephalography (TA-fEEG) signals. This method further includes training a second machine learning model to reconstruct fetal electroencephalography (fEEG) signals from TA-fEEG signals.

[0020] In addition to, or alternatively to, one or more of the features described herein, further embodiments of this method may include that the first machine learning model is trained using electroencephalogram (EEG) measurement data on the actual scalp from premature infants as correct data in combination with a realistic volume conductor model of materials and fetal tissues developed for fetal electrocardiogram (ECG).

[0021] In addition to, or alternatively to, one or more of the features described herein, further embodiments of this method may include that the second machine learning model is trained using electroencephalogram (EEG) measurement data on the actual scalp from premature infants as correct data in combination with a realistic volume conductor model of materials and fetal tissues developed for fetal electrocardiogram (ECG).

[0022] In addition to, or alternatively to, one or more of the features described herein, further embodiments of this method may include that the first machine learning model is a deep neural network and the second machine learning model is an independent component analysis model.

[0023] The above and other features and advantages of the present disclosure will become readily apparent upon consideration of the following detailed description in conjunction with the accompanying drawings.

Brief Description of the Drawings

[0024] The details of the exclusive rights described herein are particularly and distinctly identified and clearly claimed in the claims at the end of the specification. The foregoing and other features and advantages of one or more embodiments described herein will become apparent from the following detailed description taken in conjunction with the accompanying drawings, in which

[0025] [Figure 1A] FIG. 1A shows a block diagram of a system for non-invasive transabdominal fetal electroencephalogram measurement according to one or more embodiments described herein.

[0026] [Figure 1B]FIG. 1B shows a non-invasive sensing device according to one or more embodiments described herein.

[0027] [Figure 1C] FIG. 1C shows the non-invasive sensing device of FIG. 1B being worn by a subject according to one or more embodiments described herein.

[0028] [Figure 2] FIG. 2 shows a block diagram of components of a machine learning learning and inference system according to one or more embodiments described herein.

[0029] [[ID=I7]] [Figure 3] FIG. 3 shows a flowchart of a method for non-invasive transabdominal fetal electroencephalogram measurement according to one or more embodiments described herein.

[0030] [Figure 4] FIG. 4 shows a comparison between neonatal electroencephalogram and reconstructed transabdominal fEEG generated using machine learning according to one or more embodiments described herein.

[0031] [Figure 5] FIG. 5 shows a block diagram of a processing system for implementing one or more embodiments described herein. DETAILED DESCRIPTION OF THE INVENTION

[0032] The figures illustrated herein are exemplary. Without departing from the scope of the embodiments described herein, many modifications may be possible to the drawings or the operations described in the drawings. For example, the operations may be performed in a different order, or operations may be added, deleted, or changed. Also, the term "coupled" and its variations represent that there is a communication path between two elements and do not mean a direct connection without intervening elements / connections between the two elements. All of these modifications are considered to be a part of this specification.

[0033] One or more embodiments described herein provide non-invasive transabdominal fetal electroencephalography (TA-fEEG).

[0034] Decreased oxygenation of the fetal brain in utero (known as fetal hypoxia) can have devastating consequences, including irreversible neurological damage and even death. Fetal assessment aims to identify whether a fetus is at risk of fetal hypoxia and to implement timely interventions (e.g., emergency cesarean section) to prevent harm to the fetus. Conventional approaches for fetal assessment, such as external fetal monitoring (EFM) and biophysical profiling (BPP), utilize ultrasound-based techniques to measure downstream bodily responses to fetal hypoxia. However, these approaches reflect changes that occur after the onset of fetal hypoxia, which may explain why the incidence of conditions associated with fetal hypoxia (e.g., cerebral palsy and fetal death) has not decreased despite the widespread use of EFM and BPP over the past 40 years.

[0035] An alternative and more sensitive approach to detecting early signs of fetal brain hypoxia is to monitor changes in the fetal brain itself, which can be captured as electrical activity. In fact, early studies conducted during labor using transvaginal electrodes attached to the fetal scalp have shown that changes in fetal brain electrical activity induced by hypoxia can precede changes in heart rate by up to 10 minutes. Detecting early signs of fetal brain hypoxia by observing fetal brain activity allows for earlier intervention to reduce the risk of harm to the fetus. Traditionally, the only techniques available for measuring fetal brain electrical activity in utero are fetal magnetoencephalography (fMEG) and direct fetal scalp electrodes (FSE) inserted into the vagina. Unfortunately, these techniques have significant limitations that prevent their widespread adoption.

[0036] fMEG scanners are not widely available due to their high cost. Furthermore, fMEG scanners are not suitable for monitoring the fetus during labor. For example, for an fMEG scanner to operate, the pregnant woman must sit in a saddle-like seat with her abdomen inside the scanner. This design hinders clinicians' access to the mother's abdomen and birth canal, which are critical during childbirth.

[0037] Direct scalp measurement is an invasive technique that involves inserting a fetal sac (FSE) through the birth canal and attaching it to the fetal scalp. This procedure carries risks to both the mother and the fetus, including, for example, abrasions or lesions to the mother's birth canal and uterus, as well as potential damage to the fetal scalp. Furthermore, because the amniotic sac is ruptured to attach the FSE, the use of FSE is limited to delivery, and FSE is unsuitable for monitoring the condition and development of the fetus throughout pregnancy.

[0038] Therefore, while conventional fetal monitoring techniques are suitable for their intended purposes, there is a need for a non-invasive, uninterrupted, and cost-effective approach to accurately measure fetal brain activity throughout each stage of pregnancy and delivery in order to provide healthcare providers with direct insights into the health of the developing fetus.

[0039] The embodiments described above address the shortcomings of the prior art by providing non-invasive transabdominal fetal electroencephalography (fEEG). One or more embodiments described herein use non-invasive TA-fEEG to measure the electrical activity of the fetal brain. Similar to scalp electroencephalography (EEG), an inexpensive and widely used technique in which electrodes are attached to the human scalp to detect voltage changes produced by the brain, TA-fEEG involves attaching electrodes to the mother's abdomen and using them to detect voltage changes produced by the fetal brain. TA-fEEG signals are masked by significant amounts of high-amplitude artifact activity resulting from maternal abdominal muscles, maternal and fetal cardiac activity, fetal movement, sweat-related drift, uterine activity, and combinations and / or simultaneous occurrences thereof. One or more techniques described herein apply artificial intelligence (AI) and machine learning (ML) techniques to reduce or remove unwanted noise in the TA-fEEG signal, and then reconstruct the fetal EEG (fEEG) signal from data collected in the mother's abdomen. In other words, one or more embodiments described herein enable denoising of electrical signals collected by electrodes attached to the abdomen of the mother, and then reconstructing the fEEG signal.

[0040] Referring to Figure 1A, a block diagram of a system 100 for non-invasive transabdominal fetal electroencephalography is provided according to one or more embodiments described herein. The system 100 includes a sensing device 102 that communicates with a processing system 110.

[0041] Sensing device 102 has a non-invasive sensing device 101 having a sensor unit 102 and a connector unit 104. The non-invasive sensing device 101 is configured to be worn by a pregnant woman, for example, around the abdominal region. The sensor unit 102 of the non-invasive sensing device 101 includes one or more sensors 106a, 106b. Although two sensors 106a, 106b are shown, other numbers of sensors may be used in other embodiments. The connector unit 104 connects to the sensor unit 102 so that the pregnant woman can wear the non-invasive sensing device 101 so that it can be positioned around the mother's abdomen. The non-invasive sensing device 101 is adjustable to accommodate different sizes, orientations, configurations, and combinations and / or multiples thereof. One or more of the sensors 106a, 106b may be used to collect data about the pregnant woman and the fetus. One or more of the sensors 106a, 106b may be electrodes configured, for example, to collect electroencephalogram (EEG) data.

[0042] One or more sensors 106a, 106b of the non-invasive sensing device 101 can transmit data, such as electroencephalogram (EEG) data, to a processing system 110 via one or more communication links 108a, 108b. Although two communication links 108a, 108b are shown, other embodiments may use a different number of sensors. For example, according to one embodiment, sensors 106a and 106b can share a communication link. The communication links 108a and 108b may be wired and / or wireless links and may be configured to transmit analog signals and / or digital data.

[0043] The processing system 110 comprises a processing unit 112, a system memory 114, and a machine learning (ML) engine 116. Various components, modules, engines, etc., described with respect to the processing system 110 (e.g., the machine learning engine 116) can be implemented as instructions stored in a computer-readable storage medium, as hardware modules, as special-purpose hardware, such as application-specific hardware, application-specific integrated circuits (ASICs), application-specific dedicated processors (ASSPs), field-programmable gate arrays (FPGAs), embedded controllers, hardwired circuits, etc., or as a combination of these. According to each aspect of this disclosure, the engine described herein may be a combination of hardware and programming. This programming may be instructions executable by a processor stored in tangible memory, and this hardware may include a processing unit 112, for example, the processing unit 521 in Figure 5, for executing those instructions. Thus, the system memory 114, for example, the system memory 523 in Figure 5, can store program instructions that implement the engine described herein when executed by the processing unit 112. Other engines may also be used to include other features and functions described in other examples herein.

[0044] The ML engine 116 can generate and / or use one or more machine learning models, such as a denoising ML model 120, a signal reconstruction ML model 122, and / or combinations thereof and / or multiple thereof. While the denoising ML model 120 and the signal reconstruction ML model 122 are shown as part of the processing system 110, it should be understood that one or more of these models may be stored in a remote processing system (e.g., another processing system, a node in a cloud computing system, and / or similar, including combinations thereof and / or multiple thereof) and may be accessible via a network connection. According to one or more embodiments described herein, the cloud computing system may communicate with the processing system 110 by wired or wireless electronic communication. The cloud computing may complement, support, or replace some or all of the functionality of the elements of the processing system 110. Some or all of the functionality of the elements of the processing system 110 may be implemented as nodes of the cloud computing system. For example, one or more of the denoising ML model 120 and the signal reconstruction ML model 122 may be stored on a node in the cloud computing system and accessible via the Internet. As another example, the machine learning engine 116 can be implemented using a cloud computing system, where the cloud computing system performs training and / or inference as described herein.

[0045] One or more embodiments described herein can utilize machine learning techniques to perform tasks such as non-invasive transabdominal electroencephalography (TA-fEEG). More specifically, one or more embodiments described herein can achieve the various operations described herein, namely non-invasive TA-fEEG, by incorporating and utilizing rule-based decision-making and artificial intelligence (AI) reasoning. The term “machine learning” broadly refers to the ability of an electronic system to learn from data. A machine learning system, engine, or module may include a trainable machine learning algorithm, which can be trained, for example, in an external cloud environment to learn functional relationships between inputs and outputs, and the resulting model (sometimes referred to as a “trained neural network,” “trained model,” and / or “trained machine learning model”) can be used, for example, for non-invasive TA-fEEG. In one or more embodiments, the machine learning function can be implemented using a trainable artificial neural network (ANN) capable of performing the function. In machine learning and cognitive science, an artificial neural network (ANN) is a group of statistical learning models inspired by the biological neural networks of animals, particularly the brain. Artificial neural networks (ANNs) can be used to estimate or approximate systems and functions that depend on a large number of inputs. Convolutional neural networks (CNNs) are a type of deep feedforward artificial neural network and are particularly useful in tasks including, but not limited to, visual image analysis and natural language processing (NLP). Recurrent neural networks (RNNs) are another class of deep feedforward artificial neural networks and are particularly useful in tasks including, for example, unsegmented concatenated handwriting recognition and speech recognition, but not limited to these. Other types of neural networks are also known and can be used according to one or more embodiments described herein.

[0046] An artificial neural network (ANN) can be embodied as a so-called "neuromorphic" system consisting of interconnected processor elements that function as simulated "neurons" and exchange "messages" with each other in the form of electronic signals. Similar to the so-called "plasticity" of synaptic neurotransmitter connections that transmit messages between biological neurons, the connections in an ANN that transmit electronic messages between simulated neurons are assigned numerical weights corresponding to the strength or weakness of a given connection. These weights can be adjusted and tuned based on experience, making the ANN adaptive to inputs and learnable. For example, an ANN for handwriting recognition is defined by a set of input neurons that can be activated by pixels in an input image. After being weighted and transformed by a function determined by the network designer, the activation values ​​of these input neurons are then passed to other downstream neurons, often called "hidden" neurons. This process is repeated until an output neuron is activated. The activated output neuron determines which character was input. It should be understood that these same techniques can be applied to the non-invasive TA-fEEG described herein.

[0047] The processing system 110 enables the extraction of fEEG signals from signals obtained by non-invasive collection from the maternal abdomen and enables the reconstruction of TA-fEEG using machine learning techniques. According to one or more embodiments described herein, a first machine learning model (e.g., denoising ML model 120), which may be a deep neural network, is used to remove artifacts caused by maternal and fetal cardiac activity and movement, and a second machine learning model (e.g., signal reconstruction ML model 122), which may be independent component analysis (ICA), is used to isolate independent sources of activity inherent in the signal. Signals originating from the fetal brain can be used based on their waveform shape and spectrogram. This approach enables the measurement of fetal neural activity as a more sensitive and earlier detection of fetal hypoxia pre- or during labor compared to conventional approaches for detecting fetal hypoxia. Furthermore, this approach is non-invasive.

[0048] Figures 1B and 1C show another example of a non-invasive sensing device 101 according to one or more embodiments described herein. In this example, the non-invasive sensing device 101 includes a lumbar electrode 130 and an abdominal electrode 132. As shown in Figure 1C, the non-invasive sensing device 101 may be fitted around a subject 140. According to one or more embodiments described herein, the non-invasive sensing device 101 may be fitted to the subject 140 in an adjustable manner, for example, using hook-and-loop fasteners (e.g., VELCRO® 134). Furthermore, as shown in Figure 1C, the lumbar electrode 130 and / or abdominal electrode 132 can be connected to a processing system 110 via an analog-to-digital converter (ADC) 142. However, in other embodiments, the ADC 142 may be omitted. Also, a cardiac electrode 136 may be connected to the processing system 110, for example, via the ADC 142. The processing system can receive and / or process fEEG and electrocardiogram (ECG) signals from the lumbar electrode 130, the abdominal electrode 132, and / or the cardiac electrode 136.

[0049] Here, with reference to Figure 2, the system for training and using machine learning models will be described in more detail. In particular, Figure 2 shows a block diagram of the components of the machine learning training and inference system 200 according to one or more embodiments described herein. The system 200 performs training 202 and inference 204. During training 202, the training engine 216 trains a model (e.g., trained model 218) to perform a task such as performing a non-invasive TA-fEEG. It should be understood that the trained model 218 can represent one or more trained machine learning models, such as a denoising ML model 120 and / or a signal reconstruction ML model 122. In some cases, the training engine 216 trains multiple machine learning models, such as the denoising ML model 120 and the signal reconstruction ML model 122. According to one or more embodiments, the denoising ML model 120 can be trained to remove artifacts caused by maternal and fetal cardiac activity and movement. According to one or more embodiments, the signal reconstruction ML model 122 can be trained to isolate independent activity sources inherent in the TA-fEEG signal. According to one or more embodiments described herein, the denoising ML model 120 may be a first neural network, and the signal reconstruction ML model 122 may be a second neural network. Inference 204 is the process of implementing the trained model 218 to perform a task, for example, performing a non-invasive TA-fEEG, in the context of a larger system (e.g., system 226). All or part of the system 200 shown in Figure 2 may be implemented, for example, by all or part of the processing system 110 in Figure 1A.

[0050] Training 202 begins with training data 212, which may be structured or unstructured data. According to one or more embodiments described herein, training data 212 for training a denoising ML model 120 and / or a signal reconstruction ML model 122 includes a combination of actual scalp electroencephalogram (EEG) data of premature infants that can be used as ground truth data and a realistic volume conductor model of materials and fetal tissues developed for fetal ECGs (e.g., fat, muscle, amniotic sac, vernix caseosa, fetal head and brain and / or combinations thereof and / or multiples). The realistic volume conductor model of materials and fetal tissues, combined with EEG data from premature infants, simulates a fEEG signal on the mother's abdomen. During training 202, the fEEG signal may be augmented by randomly scaling its amplitude and added to an augmented artifact signal (e.g., distorted and scaled low-frequency noise, muscle activity, fetal and maternal ECGs, fetal movements and / or combinations thereof and / or multiples thereof, similarly).

[0051] The learning engine 216 receives training data 212 and a model form 214. The model form 214 represents an untrained base model. The model form 214 may have pre-set weights and biases, which may be adjusted during training. It should be understood that the model form 214 may be selected from a number of different model forms depending on the task to be performed. For example, if training 202 trains a model to perform image classification, the model form 214 may be a CNN model form. Training 202 may be supervised learning, semi-supervised learning, unsupervised learning, reinforcement learning, and / or similar, as well as combinations and / or multiple thereof. For example, supervised learning can be used to train a machine learning model to classify objects of interest in images. To do this, the training data 212 includes labeled images, including images of objects of interest with relevant labels (ground truth) and other images with relevant labels that do not contain objects of interest. In this example, the learning engine 216 receives training images from the training data 212 as input, makes predictions to classify the images, and compares the predictions to known labels. The learning engine 216 then adjusts the model weights and / or biases based on the comparison results, for example, using backpropagation. Training 202 may be run multiple times (referred to as "epochs") until an appropriate model (e.g., a denoising ML model 120 and / or a signal reconstruction ML model 122) is trained.

[0052] Training 202 may include both training a trained model 218 (for example, one or more of the denoising ML model 120 and / or signal reconstruction ML models 122) and validating the trained model 218. For example, the denoising ML model 120 and / or signal reconstruction ML model 122 can be trained on augmented data, validated using augmented data, and tested on real data collected from pregnant subjects. According to one or more embodiments described herein, training and / or validating one or more of the denoising ML model 120 and / or signal reconstruction ML models 122 can be performed using synthetic data generated based on real data collected from pregnant subjects.

[0053] Once trained, the trained model 218 can be used to perform inference 204 to perform tasks such as reconstructing fEEG and removing "unwanted" signals. The inference engine 220 applies the trained model 218 to new data 222 (e.g., real-world data that is not part of the training data). For example, in the case of TA-fEEG, the new data 222 may be TA-fEEG electroencephalogram data acquired in a non-invasive manner with respect to the subject as described herein, and the new data 222 is not part of the training data 212. Thus, the new data 222 represents data not presented to the trained model 218. The inference engine 220 makes a prediction 224 (e.g., using the new data 222, at least in part based on the reconstructed fEEG signals, to predict the likelihood of fetal hypoxia in the fetus of a pregnant subject) and passes the prediction 224 to system 226 (e.g., processing system 110 in Figure 1A). Based on the prediction 224, system 226 can take action, perform actions, perform analysis, etc., including combinations and / or a combination of these. In some embodiments, system 226 can add to and / or modify new data 222 based on the prediction 224.

[0054] In one or more embodiments, the predictions 224 generated by the inference engine 220 are periodically monitored and validated to ensure that the inference engine 220 is operating as expected. Based on the validation, additional training 202 may be performed using the trained model 218 as a starting point. The additional training 202 may include all or part of the original training data 212 and / or all or part of the new training data 212. In one or more embodiments, training 202 includes updating the trained model 218 to account for expected changes in the input data.

[0055] Figure 3 shows a flowchart of Method 300 for non-invasive transabdominal fetal electroencephalography according to one or more embodiments described herein. Any suitable system or apparatus, such as System 100 in Figure 1A, Processing System 110 in Figure 1A, Machine Learning Learning and Inference System 200 in Figure 2, Processing System 500 in Figure 5, and / or combinations thereof and / or multiple thereof, can perform Method 300. Method 300 is described below with reference to Figure 1A, but is not limited thereto.

[0056] In block 302, the processing system 110 receives non-invasive TA-fEEG signals related to the pregnant subject from the non-invasive sensing device 101. In block 304, the machine learning engine 116 reduces (or removes) unwanted noise in the TA-fEEG signals using the denoising ML model 120. In block 306, the machine learning engine 116 reconstructs the fEEG signals from the TA-fEEG signals using the signal reconstruction ML model 122.

[0057] According to one or more embodiments described herein, non-invasive TA-fEEG signals are collected from sensors of a non-invasive sensing device associated with a pregnant subject.

[0058] According to one or more embodiments described herein, Method 300 may include predicting the likelihood of fetal hypoxia in a pregnant subject's fetus based at least in part on reconstructed fEEG signals.

[0059] According to one or more embodiments described herein, the first machine learning model is a first neural network, and the second machine learning model is a second neural network.

[0060] According to one or more embodiments described herein, Method 300 may include the step of training the denoising ML model 120 as described herein. For example, the denoising ML model 120 can be trained using actual scalp electroencephalogram data from premature infants as ground truth data, in combination with realistic volume conductor models of materials and fetal tissue developed for ECG.

[0061] According to one or more embodiments described herein, Method 300 may include the step of training a signal reconstruction ML model 122 as described herein. For example, the signal reconstruction ML model 122 can be trained using actual scalp electroencephalogram data from premature infants as ground truth data, in combination with realistic volume conductor models of materials and fetal tissues developed for ECG.

[0062] According to one or more embodiments described herein, the denoising ML model 120 is a deep neural network, and the signal reconstruction ML model 122 is an independent component analysis model.

[0063] Additional processes may be included, and it should be understood that the processes illustrated in Figure 3 are illustrative, and that other processes may be added, or existing processes may be deleted, modified, or rearranged without departing from the scope of this disclosure.

[0064] Figure 4 shows a comparison between neonatal electroencephalogram (EEG) 402 and reconstructed transabdominal fetal electroencephalogram (fEEG) generated using machine learning according to one or more embodiments described herein. Neonatal EEG 402 is generated, for example, from electrodes attached to the scalp of a premature infant 403. TA-fEEG 404 is generated from electrodes attached to an intrauterine fetus 405. In this example, the premature infant 403 and the intrauterine fetus 405 are of the same gestational age, i.e., the premature infant 403 and the intrauterine fetus 405 are substantially the same gestational age. The signals from the electrodes attached to the intrauterine fetus 405 are processed using a denoising ML model 120 and a signal reconstruction ML model 122 according to one or more embodiments described herein to generate a reconstructed TA-fEEG 404. The denoising ML model 120 and the signal reconstruction ML model 122 were able to reconstruct the fetal EEG data and remove unwanted signals. Non-invasive TA-fEEG404 can be compared to direct scalp electrode electroencephalography (e.g., neonatal EEG402) of gestationally matched neonates. That is, reconstructed TA-fEEG404 can be compared to neonatal EEG402, and neonatal EEG402 can be generated, for example, using gestationally matched fetal magnetic electroencephalography (fMEG) data. Both reconstructed TA-fEEG404 and neonatal EEG402 show signal patterns characteristic of fetal age, where neural activity exhibits discontinuous patterns. Auditory stimuli with event-related potentials (e.g., specific deviations in the brain's electrical potentials that occur in response to a stimulus) can be observed by non-invasive acquisition and compared to equivalent measurements in fMEG.

[0065] According to one or more embodiments described herein, a possible and conceivable use case for TA-fEEG is the prediction of fetal hypoxia. Fetal hypoxia is difficult to detect. Electronic fetal monitoring relies on the downstream effects of hypoxia on fetal cardiac activity, which limits its accuracy and timeliness. For example, conventional approaches to predicting fetal hypoxia are as follows: oxygenated blood flows into the placenta, oxygen exchange occurs in the intervillous space, the fetal nervous system (parasympathetic and sympathetic nervous systems) responds to the oxygen state, and electronic fetal monitoring shows a change in heart rate. This process is slow and error-prone. One or more embodiments described herein improve the timeliness and accuracy of conventional methods for detecting fetal hypoxia by applying a trained machine learning model to the TA-fEEG signal to target the upstream effects of neural activity earlier for more accurate detection of fetal hypoxia. The use of the denoising ML model 120 and the signal reconstruction ML model 122 enables reliable measurement of fetal EEG signals and leads to earlier detection of fetal hypoxia compared to conventional approaches (e.g., using downstream approaches related to cardiac activity). The TA-fEEG approach described herein offers many possible clinical implications.

[0066] In the context of labor, fetal hypoxia and / or injury are highly likely to occur. One or more embodiments described herein provide detectors for earlier detection of fetal hypoxia, improved monitoring sensitivity, earlier intervention, reduced incidence of cerebral palsy at birth, reduced rates of unnecessary cesarean sections, improved maternal and infant outcomes as a result of innovative labor monitoring, and combinations and / or more thereof. In the context of prepartum, high-risk fetuses can be targeted to provide additional clinical information. One or more embodiments described herein provide enabling targeting of fetuses with brain abnormalities or congenital abnormalities such as spina bifida, providing parents with a neurological outlook and expectations at birth, providing auditory stimulation as a primary hearing test for fetuses at risk of congenital hearing loss, and / or combinations and / or more thereof. It should be understood that other intrapartum and prepartum applications of one or more embodiments described herein are also possible and are not limited to the examples provided.

[0067] Regarding cerebral palsy (CP), it is a common motor disorder in childhood, accompanied by lifelong symptoms including mobility impairment, developmental delay, chronic pain, seizure disorders, and combinations and / or multiples of these. Children with cerebral palsy incur approximately 26 times higher medical costs, and the estimated lifetime national cost for all children born with cerebral palsy in 2000 was $11.5 billion. Many CP cases are congenital, occurring prenatally or at birth due to unrecognized fetal distress. Electronic fetal monitoring using fetal heart rate patterns is used in over 90% of deliveries, but despite its widespread use, the incidence of cerebral palsy has not decreased; rather, since the adoption of this technology, cesarean section rates and associated maternal risks and medical costs have steadily increased. This highlights the current costs of inadequate fetal monitoring systems (e.g., economic costs, health costs, and long-term risks) and the need for more sensitive and accurate monitoring for fetal hypoxia. Fetal electroencephalography (EEG) has been shown to be more accurate and timely, enabling earlier and more precise intervention for fetal distress, aiming to reduce the incidence of cerebral palsy and the rate of unnecessary cesarean sections.

[0068] It is understood that one or more embodiments described herein can be implemented in combination with any other type of computing environment currently known or to be developed later. For example, Figure 5 shows a block diagram of a processing system 500 for implementing the technology described herein. According to one or more embodiments described herein, the processing system 500 is an example of a cloud computing node in a cloud computing system. In the embodiments, the processing system 500 has one or more central processing units ("processors" or "processing resources" or "processing devices") 521a, 521b, 521c, etc. (collectively or generally referred to as processors 521 and / or processing devices 521). In embodiments of this disclosure, each processor 521 may include a reduced instruction set computer (RISC) microprocessor. The processors 521 are coupled via a system bus 533 to system memory (e.g., random access memory (RAM) 524) and various other components. Read-only memory (ROM) 522 is coupled to the system bus 533 and may include a basic input / output system (BIOS), which controls certain basic functions of the processing system 500. The system memory 523 may include ROM 522, RAM 524, and / or any other suitable memory devices, and may include combinations and / or multiple forms thereof.

[0069] Furthermore, the diagram shows that an input / output (I / O) adapter 527 and a network adapter 526 are coupled to the system bus 533. The I / O adapter 527 may be a small computer system interface (SCSI) adapter that communicates with a hard disk 523 and / or a storage device 525, or other similar components. The I / O adapter 527, the hard disk 523, and the storage device 525 are collectively referred to herein as mass storage 534. An operating system 540 for running on the processing system 500 may be stored in the mass storage 534. The network adapter 526 interconnects the system bus 533 with an external network 536, enabling the processing system 500 to communicate with other such systems.

[0070] A display 535 (for example, a display monitor) is connected to the system bus 533 by a display adapter 532, which may include a graphics adapter and video controller for improving the performance of graphics-intensive applications. In one aspect of this disclosure, adapters 526, 527, and / or 532 may be connected to one or more I / O buses connected to the system bus 533 via an intermediate bus bridge (not shown). Suitable I / O buses for connecting peripherals such as hard disk controllers, network adapters, and graphics adapters typically include common protocols such as Peripheral Component Interconnect (PCI). Additional input / output devices are shown to be connected to the system bus 533 via a user interface adapter 528 and a display adapter 532. A keyboard 529, a mouse 530, and a speaker 531 may be interconnected to the system bus 533 via the user interface adapter 528, which may include, for example, a SuperI / O chip that integrates multiple device adapters into a single integrated circuit.

[0071] In some aspects of this disclosure, the processing system 500 includes a graphics processing unit 537. The graphics processing unit 537 is a dedicated electronic circuit designed to manipulate and modify memory to accelerate the generation of images in a frame buffer intended for output to a display device. Generally, the graphics processing unit 537 is highly efficient at manipulating computer graphics and performing image processing, and also has a highly parallel structure, which makes it more effective than a general-purpose CPU for algorithms where large data blocks are processed in parallel.

[0072] Accordingly, as configured herein, the processing system 500 comprises processing capabilities in the form of a processor 521, storage capabilities including system memory (e.g., RAM 524) and mass storage 534, input means such as a keyboard 529 and a mouse 530, and output capabilities including a speaker 531 and a display 535. In some embodiments of this disclosure, a portion of the system memory (e.g., RAM 524) and the mass storage 534 together store an operating system 540 for coordinating the functions of the various components shown in the processing system 500.

[0073] In this specification, various embodiments are described with reference to the relevant drawings. Alternative embodiments can be devised without departing from the claims. Various connections and positional relationships between elements (e.g., above, below, adjacent, etc.) are described in the following description and drawings. Unless otherwise specified, these connections and / or positional relationships may be direct or indirect, and the embodiments described herein are not intended to be limited in this respect. Thus, connections between entities may refer to either direct or indirect connections, and positional relationships between entities may be direct or indirect positional relationships. Furthermore, the various tasks and process steps described herein may be incorporated into more comprehensive procedures or processes that have additional steps or functions not described in detail herein.

[0074] The following definitions and abbreviations shall be used for interpretation of the claims and specification. As used herein, the terms “equipped,” “containing,” “included,” “having,” “having,” “incorporating,” “incorporating,” or other variations thereof are intended to cover non-exclusive inclusion. For example, a composition, mixture, process, method, article or apparatus containing a list of elements is not necessarily limited to those elements alone, but may include other elements not expressly enumerated, or other elements inherent in such composition, mixture, process, method, article or apparatus.

[0075] Furthermore, the term “exemplary” is used herein to mean “serving as an example, case, or illustration.” Any embodiment or design described herein as “exemplary” should not necessarily be construed as being preferable or advantageous to other embodiments or designs. The terms “at least one” and “one or more” may be understood to include any integer one or more, i.e., 1, 2, 3, 4, etc. The term “multiple” may be understood to include any integer two or more, i.e., 2, 3, 4, 5, etc. The term “connection” may include both indirect and direct “connections.”

[0076] The terms “approximately,” “substantially,” “about,” and their variations are intended to include the degree of error associated with the measurement of a particular quantity based on the equipment available at the time of filing. For example, “approximately” may include a range of ±8%, ±5%, or ±2% of a given value.

[0077] The flowcharts and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of the systems, methods, and computer program products according to the various embodiments described herein. In this regard, each block in a flowchart or block diagram may represent a module, segment, or part of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions described in a block may be executed in an order different from the order shown in the diagram. For example, two blocks shown consecutively may actually be executed substantially simultaneously, depending on the functions involved, or they may be executed in reverse order. It should also be noted that each block in a block diagram and / or flowchart, as well as combinations of blocks in a block diagram and / or flowchart, can be implemented by a purpose-specific hardware-based system that performs a given function or operation, or a combination of purpose-specific hardware and computer instructions.

[0078] The descriptions of various embodiments are presented for illustrative purposes only and are not intended to be exhaustive or limitful to the disclosed embodiments. Those skilled in the art will see many modifications and variations that do not deviate from the scope of the embodiments described. The terminology used herein has been selected to best describe the principles, practical applications, or technical improvements of the embodiments, or to enable those skilled in the art to understand the embodiments described herein.

Claims

1. A step of receiving non-invasive transabdominal fetal electroencephalography (TA-fEEG) signals related to a pregnant subject, A step of reducing unwanted noise in the TA-fEEG signal using a first machine learning model, The second machine learning model is used to reconstruct the fetal electroencephalogram (fEEG) signal from the TA-fEEG signal. Computer implementation methods, including those mentioned above.

2. The computer implementation method according to claim 1, wherein the non-invasive TA-fEEG signal is collected from a sensor of a non-invasive sensing device associated with the pregnant subject.

3. The computer implementation method according to claim 1, further comprising the step of predicting the likelihood of fetal hypoxia in the fetus of the pregnant subject based at least in part on the reconstructed fEEG signal.

4. The computer implementation method according to claim 1, wherein the first machine learning model is a first neural network, and the second machine learning model is a second neural network.

5. The computer implementation method according to claim 1, further comprising the step of training the first machine learning model.

6. The computer implementation method according to claim 5, wherein the first machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

7. The computer implementation method according to claim 1, further comprising the step of training the second machine learning model.

8. The computer implementation method according to claim 7, wherein the second machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

9. The computer implementation method according to claim 1, wherein the first machine learning model is a deep neural network, and the second machine learning model is an independent component analysis model.

10. A non-invasive sensing device having a sensor that detects TA-fEEG signals in pregnant subjects; A processing system having a memory for storing computer-readable instructions and a processing unit for executing the computer-readable instructions, and communicating with the sensor; Equipped with, Instructions readable by the aforementioned computer control the processing unit. A first machine learning model is used to remove artifacts from the TA-fEEG signal caused by maternal and fetal cardiac activity and movement. A second machine learning model is used to isolate independent activity sources inherent in the TA-fEEG signal. A step of predicting the likelihood of fetal hypoxia in the fetus of the pregnant subject, based at least partially on at least one of the independent activity sources inherent in the TA-fEEG signal, Perform an action that includes A system for performing non-invasive transabdominal fetal electroencephalography (TA-fEEG).

11. The system according to claim 10, wherein the first machine learning model is a first neural network, and the second machine learning model is a second neural network.

12. The system according to claim 10, wherein the operation further includes the step of training the first machine learning model.

13. The system according to claim 12, wherein the first machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

14. The system according to claim 10, wherein the operation further includes the step of training the second machine learning model.

15. The system according to claim 14, wherein the second machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

16. The system according to claim 10, wherein the first machine learning model is a deep neural network, and the second machine learning model is an independent component analysis model.

17. A step to train a first machine learning model to reduce unwanted noise in non-invasive transabdominal fetal electroencephalography (TA-fEEG) signals. A step of training a second machine learning model to reconstruct fetal electroencephalography (fEEG) signals from the TA-fEEG signals, A computer implementation method for training machine learning models, including [specific examples of machine learning models].

18. The computer implementation method according to claim 17, wherein the first machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

19. The computer implementation method according to claim 17, wherein the second machine learning model is trained using actual electroencephalogram (EEG) data from the scalp of premature infants as ground truth data, in combination with a realistic volume conductor model of materials and fetal tissue developed for fetal electrocardiograms (ECG).

20. The computer implementation method according to claim 17, wherein the first machine learning model is a deep neural network, and the second machine learning model is an independent component analysis model.

Images