Improved method for performing an imaging examination

Abstract

The method for performing an imaging examination on a patient by means of a medical imaging system according to the invention comprises: starting an imaging examination by means of the imaging system; detecting input audio signals by means of at least one microphone, extracting speech acoustics of the patient and system acoustics of the imaging system from the input audio signals by means of a computing unit; classifying the system acoustics by means of the computing unit; weighting the classified system acoustics by means of the computing unit with at least two different, from a storage unit called weighting factors; deriving a modification indication related to the patient of the imaging examination by means of the computing unit by analyzing the perceptibility of the extracted speech acoustics; modifying the speech acoustics and / or the weighted system acoustics in accordance with the modification indication by means of the computing unit; combining the possibly modified speech acoustics and the possibly modified system acoustics into an output audio signal by means of the computing unit; providing the combined output audio signal to a user of the imaging system during the performance of the imaging examination for monitoring the imaging examination by means of a reproduction unit.

Description

Technical Field

[0001] The present invention relates to a method for performing imaging examinations on a patient using a medical imaging system, particularly a CT or MRT or C-arm imaging system, the imaging system, and an associated computer program product. Background Technology

[0002] Traditional imaging systems typically generate a significant amount of system noise during operation. The noise from different components of the imaging system is typically mixed. Noise from instruments not part of the imaging system can also mix with and / or superimpose the system noise.

[0003] Users of imaging systems, such as operators, especially MTRA and / or radiologists, are typically located in a control room during system operation, spatially separated from the examination room containing the imaging system. The control room and examination room are particularly shielded from each other in terms of X-rays and / or electromagnetic fields. In the case of X-rays, this is especially useful for protecting users from potential hazards caused by X-rays. In the case of MRT imaging systems, the magnetic field of the imaging system is typically shielded so as not to impair the functionality of the technical instruments outside the examination room.

[0004] Due to spatial separation, patients undergoing examinations using imaging systems are not physically adjacent to the user of the imaging system, and especially are not within the user's hearing distance. Therefore, according to existing technology, voice communication devices are typically configured to enable voice communication between the patient and the user. For this purpose, the voice communication device typically has a microphone and a speaker for both the patient and the user.

[0005] Especially due to system noise, users often have difficulty understanding patients. It is known to improve audio signals, particularly those including the patient's voice and the system noise as interference, by means of filters and / or frequency-based enhancement. Alternatively or additionally, trained machine learning models can be used to improve the audio signal. Audio signal enhancement is typically used to make the voice stand out relative to the system noise, especially to the point that the latter is no longer perceptible to the user. Summary of the Invention

[0006] The purpose of this invention is to provide a method for performing imaging examinations on a patient using a medical imaging system, particularly a CT or MRT imaging system, the imaging system, and an associated computer program product, said method, imaging system, and computer program product being improved, particularly ensuring improved performance of the imaging examination, preferably by monitoring the improvements in the imaging examination.

[0007] The objective is achieved through the features of the invention. Advantageous designs are described in the following description.

[0008] The method according to the present invention for performing imaging examinations on a patient by means of a medical imaging system, particularly a CT or MRT or C-arm imaging system, comprises the following steps:

[0009] - Initiate an imaging examination using an imaging system.

[0010] - Input audio signals are detected by means of at least one microphone, wherein the at least one microphone is configured and oriented to record the sounds of the patient and the imaging system during the imaging examination and to provide them as input audio signals.

[0011] - The patient's voice and the imaging system's system sounds are extracted from the input audio signal using a computing unit.

[0012] - The system's sound is classified using a computing unit.

[0013] - The system sounds are weighted by a computing unit using at least two different weighting factors retrieved from a storage unit.

[0014] - A modification instruction is derived by analyzing the perceptibility of the extracted speech sounds using a computing unit; this modification instruction is relevant to the patient undergoing imaging examination.

[0015] - The computing unit modifies the speech and / or weighted system sounds according to the modification instructions.

[0016] - The computing unit combines potentially modified speech sounds and potentially modified system sounds into an output audio signal.

[0017] - The combined output audio signal is provided to the user of the imaging system during the imaging inspection by means of the reproduction unit to monitor the imaging inspection.

[0018] An imaging system according to the invention for performing the method according to the invention has

[0019] - Imaging unit,

[0020] - Computing unit,

[0021] - Reproduction unit, and

[0022] - At least one microphone.

[0023] All embodiments of the method according to the present invention are permitted to be referenced.

[0024] The computer program product according to the present invention includes a program code mechanism that, when executed by a computing unit, causes the computing unit to execute the method according to the present invention.

[0025] All embodiments of the method and imaging system according to the invention are permitted to be referenced.

[0026] In the following description, the solutions according to the invention are described not only with reference to the claimed imaging system but also with reference to the claimed method. Features, advantages, or alternative embodiments can be adapted to other separately claimed objects, and vice versa. In other words, embodiments of the imaging system can be improved by the features described or claimed in the context of the corresponding method. In this case, the functional features of the method are realized by the physical units, particularly components, of the imaging system.

[0027] Furthermore, in the following description, the solutions according to the invention are described not only with reference to methods and imaging systems for performing imaging examinations, but also with reference to methods and imaging systems for providing trained machine learning models. Features, advantages, or alternative embodiments can be adapted to other separately claimed objects, and vice versa. In other words, embodiments for providing trained machine learning models can be improved by features described or claimed in the context of performing imaging examinations. In particular, the dataset used in the methods and imaging systems for performing imaging examinations can have the same characteristics and features as the corresponding dataset in the methods and imaging systems for providing trained machine learning models, and the trained machine learning model provided by the corresponding methods and imaging systems can be used in the methods and imaging systems for performing imaging examinations.

[0028] It is important to recognize that system sounds provide the user of the imaging system with indications of the extent to which the imaging examination has progressed, or in particular, as planned. The system sounds, in particular, contain at least in part information that can support the user during the performance of the imaging examination. Monitoring the system sounds is at least one feasible and / or component of monitoring the performance of the imaging examination.

[0029] This invention offers the particular advantage of not partially or completely suppressing system sounds as is known from the prior art, but rather classifying and weighting them according to their type. In addition to speech sounds, the weighted system sounds, based on the weighting, are perceptible to the user by means of the reproduction unit. This advantageously enables user monitoring of the imaging examination, thereby improving both the monitoring of the imaging examination and the imaging examination itself.

[0030] Imaging examinations are particularly methods used to acquire medical images of patients. Depending on the type of imaging system, the images acquired from the patient are 2D, 3D, or 4D image datasets.

[0031] Medical imaging systems can be, in particular, CT imaging systems, MRT imaging systems, or C-arm imaging systems. CT imaging systems typically include a CT gantry as part of an imaging unit, the gantry having a rotatable or movable portion relative to a stationary portion. Purely static CT imaging systems are also known. CT imaging systems particularly use an X-ray emitter as part of the imaging unit for acquiring medical images. MRT imaging systems typically include multiple MRT coils, particularly a main magnetic field coil and gradient coils, as part of the imaging unit. MRT imaging systems typically use MRT coils for acquiring medical images. Alternatively, medical imaging systems can be C-arm imaging systems. C-arm imaging systems particularly include a movable C-arm as part of the imaging unit, where an X-ray emitter is particularly disposed as part of the imaging unit for acquiring medical images.

[0032] In this document, the initiation of an imaging examination should be understood broadly. Initiation specifically refers to the point in time at which one of the activities listed below is first recorded during the imaging examination of the patient. Initiation of an imaging examination can, for example, be performed in conjunction with the patient's preparation for the imaging examination and / or with the manipulation of components of the imaging system, thereby enabling, in particular, interaction with the patient during the preparation and / or execution of the imaging examination. Imaging examinations using an imaging system can, in particular, be initiated manually and / or automatically by the user of the imaging system.

[0033] Patient preparation can include, for example, identifying the patient outside or inside the examination room and / or at least before the patient is supported on the examination bed. Patient preparation can include positioning the examination bed, on which the patient is supported, relative to the imaging system. Positioning the examination bed can particularly include moving the examination bed to, preferably into, the examination area of ​​the imaging system. Alternatively or additionally, patient preparation can include placing an inlet for contrast agent injection.

[0034] The components that manipulate the imaging system may include those that trigger movement of movable parts of the imaging system. These movable parts may be rotating parts of an examination table, a C-arm, a gantry, and / or movable parts of the gantry. Triggering movement of these movable parts can be adapted to orient the imaging system, particularly its movable parts, relative to the patient.

[0035] Alternatively or additionally, the components manipulating the imaging system may include triggering an image acquisition routine by means of one, more, or all of the components of the imaging system, wherein an image of the patient is acquired by means of the image acquisition routine. The image acquisition routine may include instructions for generating X-rays by means of an X-ray radiator; instructions for emitting RF pulses by means of an MRT coil; instructions for outputting patient instructions, such as breathing commands and / or commands to remain still, to the patient by means of a speaker; instructions for detecting the patient's vital parameters by means of sensors; and / or instructions for injecting contrast agent into the patient by means of a contrast agent infusion pump.

[0036] By definition, the time period during an imaging examination begins with the initiation of the examination and ideally ends no earlier than the termination of the image acquisition routine. It is conceivable that the time period during an imaging examination ends with the completion of the image acquisition routine or subsequent work following the imaging examination. Subsequent work following an imaging examination may in particular include image reconstruction and / or removal of the examination table from the imaging system.

[0037] The detection of the input audio signal can be performed throughout the entire imaging examination period, or alternatively, within at least one consecutive sub-period within the imaging examination period. At least one consecutive sub-period can be significantly shorter than the entire imaging examination period. "Continuous" specifically means uninterrupted.

[0038] Advantageously, all steps from detecting the input audio signal to providing a combined output audio signal (including providing the combined output audio signal) are performed in real time, in principle continuously, and / or multiple times successively and / or in parallel. This is explained exemplarily in the following paragraphs.

[0039] The detection of the input audio signal is preferably performed in real time, especially without significant delay. The detection of the input audio signal can, in principle, be performed continuously and / or multiple times sequentially and / or in parallel. This is also particularly suitable for recording sounds and / or providing input audio signals.

[0040] As can be seen from the previous two paragraphs, the playback time of the sound basically corresponds to the recording time of the sound, since the execution of the steps from detecting the input audio signal to providing the combined output audio signal (including providing the combined output audio signal) proceeds without delay, except as deduced from an advantageous embodiment further below.

[0041] Suitable microphones are known in principle. Detection of the input audio signal can be performed using a single microphone or multiple microphones, which can preferably be spatially distributed.

[0042] At least one microphone is configured in principle for recording sound. For this purpose, at least one microphone has, for example, a recording interface, such as a sound sensor. At least one microphone particularly has one or more recording channels. At least one microphone, especially its recording interface, is particularly aligned with the patient and / or imaging system.

[0043] The patient's sounds include, in particular, the patient's voice and optionally, other speech sounds. Additionally, the patient's sounds may include the patient's laute and / or body movements, especially breathing movements, leg movements, and / or arm movements. Other speech sounds include, in particular, the voice of the doctor and / or another person.

[0044] The noise from an imaging system includes, in particular, the noise generated by the imaging system during the imaging procedure and optionally, additional ambient noise. The noise from the imaging system is especially the operating noise during the imaging procedure. Additional ambient noise can be generated by external instruments that are not part of the imaging system but assist in the imaging procedure during the procedure.

[0045] At least one microphone may include a conversion unit for converting recorded sound into an input audio signal. At least one microphone may provide the input audio signal at an output interface, for example, for transmission to a computing unit. The output interface may be wired or wireless.

[0046] At least one microphone can be securely mounted to the imaging system. Multiple microphones can be arranged, for example, across the examination room, partially at the imaging system and partially separate from the imaging system. At least one microphone can include a mixing unit, by means of which the sounds recorded by one or more microphones and / or the input audio signals provided can be combined into an input audio signal that can be transmitted to a computing unit, for example, via an output interface.

[0047] Extracting the patient's speech and the imaging system's system sounds from the input audio signal can include separating and / or filtering the input audio signal. Extraction is performed, in particular, by means of a computational unit, according to an extraction procedure code mechanism. This extraction procedure code mechanism can, in particular, include a first trained machine learning model. Extraction is performed, in particular, by associating the recorded patient's sounds with speech and the recorded imaging system sounds with system sounds. It is conceivable that, during extraction, a portion of those sounds that do not correspond to the speech and system sounds will be filtered out and discarded.

[0048] Classifying system sounds particularly involves associating system sounds with different components of the imaging system and optionally with components of external instruments, which are not part of the imaging system but assist in the performance of imaging examinations. Classification particularly includes dividing system sounds into sound categories. Classification can be performed segmentally and / or channel-wise. "Segmentally" means dividing the system sounds into multiple time segments of arbitrary duration, wherein each of the multiple segments is classified. "Channel-wise" means segmentally mixing and / or superimposing system sounds of different sound categories, wherein the mixed or superimposed system sounds are separated during classification and the separated system sounds are classified separately.

[0049] Each component typically has at least one of its own sound categories. In principle, components can be configured to produce different sounds, each of which can be associated with its own sound category. It is conceivable that, during classification, system sounds may be associated with only one, multiple, or all sound categories of the components of the imaging system. It is conceivable that the associated sound categories may vary segmentally and / or occur partially simultaneously. Typically, at least a portion of the system sounds is not associated with any component of the imaging system, but rather serves as background noise, for example, associated with the environment of the imaging system.

[0050] System sounds particularly include sounds from at least two of the following inexhaustible categories, wherein system sounds are classified into at least two categories during classification:

[0051] - Background noise interference

[0052] - The scanning-specific movements of the examination bed, especially the supporting parts of the examination bed, and / or the movement of the entire examination bed.

[0053] - MRT coils, especially gradient coils

[0054] - Coolant pumps, especially helium pumps,

[0055] - The rotating part of the CT gantry.

[0056] - The movable part of the CT gantry.

[0057] - X-ray radiator

[0058] - Contrast agent infusion pump,

[0059] - Patient instructions,

[0060] - C-arm,

[0061] - Vital parameter sensors, especially patient monitoring systems, and / or

[0062] - External instruments.

[0063] System sounds are classified, especially after the system sounds have been extracted from the input audio signal. Classification of system sounds can be performed using a computational unit based on a classification program code mechanism. This classification program code mechanism can, in particular, include a second trained machine learning model and / or a first trained machine learning model.

[0064] The weighting factor can be time-resolved. The weighting factor can vary between time segments of the same sound category classification. The weighting factor can be time-resolved such that the time segments of the sound category classification have weighting factors that differ from the weighting factors of earlier or later time segments of the same sound category classification. The time segments of the sound can be associated with time-resolved weighting factors based on at least two sound categories. Time-resolved weighting factors are particularly advantageous because the time segments of the same sound category can differ in their importance and / or criticality to surveillance imaging inspection.

[0065] Typically, the weighting factor reflects the necessity of the correspondingly weighted sound perceived by the user of the imaging system. Generally, the more critical and / or important the sound category, especially the corresponding time segment, is to be provided for the surveillance imaging inspection, the higher the weighting factor. The importance or criticality of the sound to the surveillance imaging inspection can be particularly considered in terms of the extent to which the sound in question must be reproduced in accordance with the rhythm of the recorded sound, or whether it is possible to reproduce it earlier or later.

[0066] Weighting the classified system sounds can include invoking weighting factors. Alternatively or additionally, the weighting factors can be invoked before weighting. In this document, invoking weighting factors from a storage unit should be understood broadly and can include executing program code in which the weighting factors are hard-coded and temporarily stored in working memory, which serves as the storage unit. It is conceivable that a user adjusts the weighting factors using an input mechanism before or during imaging examination and transmits the adjusted weighting factors to the storage unit for weighting the classified system sounds. The storage unit can be part of a computing unit and / or an imaging system.

[0067] Weighting factors are typically associated with a sound category. The weighting factor, in arbitrary units, describes the relative change in the perceptibility of a system sound with respect to that sound category. A weighting factor greater than 1 increases the perceptibility of that system sound. A weighting factor equal to 1 keeps the perceptibility of that system sound unchanged, thus keeping the system sound associated with that sound category substantially unchanged. A weighting factor less than 1 decreases the perceptibility of that system sound. A weighting factor equal to zero eliminates the perceptibility of that system sound. At least one of the weighting factors can be 1. Weighting factors not equal to 1 alter perceptibility, particularly by means of corresponding phase and / or frequency and / or amplitude adjustments.

[0068] Weighting of classified system sounds typically occurs after the system sounds have been classified. This weighting can be performed using a computational unit based on a weighting program code mechanism.

[0069] The process of obtaining modification instructions specifically includes analyzing the perceptibility of speech sounds, particularly their perceptibility compared to weighted system sounds. The perceptibility of speech sounds can be parameterized by perceptibility values ​​describing the perceptibility of the speech sounds. The process of obtaining modification instructions is particularly performed under the condition that both speech sounds and system sounds are perceptible to a human receiver, particularly a user, when a combined output audio signal is provided, wherein preferably, speech sounds have higher priority than system sounds.

[0070] The modification instructions specifically state that the speech and / or system sounds are modified prior to combination by means of phase adjustment and / or frequency adjustment and / or amplitude adjustment and / or playback timing adjustment. The modifications should be made in such a way that the basic characteristics of the sounds are maintained, but in the event of another sound occurring close in time, the sound interferes less with the perception of that other sound, and / or the sound itself is better perceived.

[0071] Perceptibility values ​​can typically be categorized into objective, immutable values ​​and subjective, modifiable values. The objective, immutable values ​​relate to the patient's volume and / or average human hearing threshold, while the subjective, modifiable values ​​ensure that the patient's voice is recognized by the user and / or that the user understands the patient. Compared to weighting factors, perceptibility values ​​are typically not purely objectively determined but rather take into account the individualized characteristics of the patient's speech.

[0072] Retrieving a modification instruction can include calling a particularly objective perceptibility value. Alternatively or additionally, the particularly objective perceptibility value can be called before retrieving the modification instruction. In this document, calling a particularly objective perceptibility value from a storage unit should be understood broadly and can include executing program code in which the particularly objective perceptibility value is hard-coded and temporarily stored in working memory, which serves as the storage unit. It is conceivable that a user adjusts a particularly objective or subjective perceptibility value using an input mechanism before or during imaging examination and transmits the adjusted perceptibility value to the storage unit to retrieve the modification instruction. The storage unit can be part of a computing unit and / or an imaging system.

[0073] The modification instruction is typically obtained after the speech sound is extracted. The modification instruction can be obtained via a computational unit using a mechanism of code, which is particularly related to objective and subjective perceptibility values. The mechanism of code can, in particular, include a third trained machine learning model and / or a first trained machine learning model and / or a second trained machine learning model.

[0074] Modifying the speech sound and / or the weighted system sound particularly involves executing a desired modification instruction by means of a computing unit. Based on the desired modification instruction, the modification may involve modifying only the speech sound, only the weighted system sound, or both the speech sound and the weighted system sound. The modification can be performed, in particular, by means of corresponding phase and / or frequency and / or amplitude adjustments. The modification is preferably performed under the condition that the intelligibility of the speech sound relative to the weighted system sound is ensured.

[0075] Modifications are made, particularly after obtaining modification instructions and after weighting the system sound. Modifications can be performed using a computational unit based on a modification program code mechanism. The modification program code mechanism can, in particular, include a fourth trained machine learning model and / or a first trained machine learning model and / or a second trained machine learning model and / or a third trained machine learning model.

[0076] The combination can include mixing and / or adding potentially modified speech sounds and potentially modified system sounds. Potentially modified speech sounds are those modified in previous steps according to modification instructions, or in other cases, speech sounds extracted from the patient. Potentially modified system sounds are those modified in previous steps according to modification instructions, or in other cases, weighted system sounds from the imaging system.

[0077] Combining is performed, especially after modification. Combining can be performed using a computing unit according to the combination program code structure. The computing unit can be configured to transmit the combined output audio signal to the playback unit and / or storage unit, and / or provide the playback unit with the combined output audio signal.

[0078] Providing a combined output audio signal can include receiving and / or recalling the combined output audio signal from a computing unit and / or a storage unit. Providing the output audio signal to a user of the imaging system can include reproducing the output audio signal to the user by means of a reproduction unit.

[0079] The playback unit may include a speaker and / or headphones. The playback unit may also be part of a voice communication device, which the user can use during imaging examinations.

[0080] Users can advantageously monitor imaging examinations based on the provided output audio signals. Monitoring of imaging examinations includes, in particular, continuing the imaging examination as initially started, adjusting the imaging examination (e.g., by outputting commands to the patient via a speaker and / or instructing adjustments to the image acquisition routine), prematurely ending the imaging examination, and / or triggering alarm signals.

[0081] One embodiment proposes that the requested modification instruction includes shifting the playback time of at least one segment of the speech sound. Preferably, the playback time can be modified according to the modification instruction to reduce the temporal overlap with the system sound segment. This embodiment is particularly advantageous because it reduces the superposition of system sound and speech sound. This embodiment is also advantageous in cases where forward or backward masking is perceptible to the user. Forward masking occurs, in particular, when the sound acts as a masking sound, especially in the period up to 1 second (e.g., 50ms to 500ms) after the masking sound ends, making it more difficult or hindering the perception of subsequent, smaller sounds. Backward masking occurs, in particular, when the sound acts as a masking sound, for example, in the period up to 0.5 seconds (e.g., 5ms to 20ms) before the masking sound appears, making it more difficult or hindering the perception of smaller sounds that appear shortly before it. The shift in playback time can be a maximum of 20 seconds, especially less than 2 seconds, or especially less than 1 second in cases of high importance, but greater than 0 seconds. The shift in playback time specifically means that a time offset is calculated relative to the recording time of the sound and the playback time derived therefrom, and this time offset is additionally added to the actual playback time by means of a calculation unit until at least one segment of the speech sound is provided after the shift ends.

[0082] An improvement to the above implementation proposes that the modification instruction be obtained based on the time-resolved weighting factor of the classified system sounds. This implementation is particularly advantageous in order to avoid the shift that occurs even though the corresponding segments of the system sounds can be superimposed on the speech sounds due to their relatively small weighting factors.

[0083] One implementation proposes that the modification according to the modification instructions includes normalization of the volume of the speech sound with respect to a reference value. The reference value can be the average volume level of the speech sounds of other patients, wherein the reference value is retrieved from a storage unit, for example, before normalization. This implementation is particularly advantageous because it ensures that even patients speaking relatively softly can be perceived by the user of the imaging system.

[0084] Typically, trained machine learning models mimic the cognitive functions that connect the human brain to other human brains. Machine learning models are particularly capable of adapting to new situations and recognizing and extrapolating patterns through training on training data. Another term for "trained machine learning model" is "trained function." The parameters of a machine learning model can typically be tuned through training. Supervised, semi-supervised, and unsupervised training methods, reinforcement learning, and / or active learning can be used in particular. Representation learning (an alternative term is "feature learning") can also be used. The parameters of a machine learning model can be iteratively tuned through multiple training steps. A specific cost function can be minimized during training. Backpropagation algorithms can be used during the training of neural networks. Machine learning models can include neural networks, support vector machines, decision trees, and / or Bayesian networks, and / or machine learning models can be based on k-means clustering, Q-learning, genetic algorithms, and / or association rules. Neural networks can be deep neural networks, convolutional neural networks, or convolutional deep neural networks. Furthermore, neural networks can be adversarial networks, deep adversarial networks, and / or generative adversarial networks.

[0085] Convolutional neural networks (CNNs) can process input images very efficiently because convolutions can extract different image features based on different kernels, allowing important image features to be found during training by adjusting the weights of the convolution kernels. Furthermore, through weight sharing (Gewichtteilen) within the convolution kernels, fewer parameters need to be trained, preventing overfitting during training and enabling faster training or more layers in the network, thereby improving network performance. A similar approach applies to sound.

[0086] According to one aspect, a machine learning model can include one or more Residual Networks (ResNets). A ResNet is particularly an artificial neural network with at least one skipped or hopped connection that skips at least one layer of the artificial neural network. A ResNet can particularly be a convolutional neural network with one or more skipped connections, each skipping one or more convolutional layers. According to some examples, a ResNet can be represented as an m-layer ResNet, where m is the number of layers in the corresponding architecture, and according to some examples, values ​​of 34, 50, 101, or 152 can be used. According to some examples, such an m-layer ResNet can include (m-2) / 2 skipped connections. The skipped connection can be viewed as a bypass that directly passes the output of the previous layer to the next layer across one or more skipped layers. Instead of directly adjusting the desired mapping, the skipped layers must then "balance" the residual mapping, which adjusts the directly passed output.

[0087] Adjusting the residual mapping is computationally simpler to optimize compared to direct mapping. Furthermore, the problem of vanishing / exploding gradients during optimization when training machine learning models is addressed by skipping layers if such a problem occurs, by regularizing the output of the directly passed layers. The use of ResNet offers the advantage of enabling the training of significantly deeper networks. This results in more accurate predictions of system sounds within the proposed range of systems providing, for example, classification—thus contributing to improved associations.

[0088] Generative adversarial models (GA models) consist of a generator function and a discriminator function. The generator function creates synthetic data, and the discriminator function distinguishes between synthetic and real data. By training the generator and / or discriminator functions, the generator function is configured to create synthetic data that is incorrectly classified as real by the discriminator function, while the discriminator function is configured to distinguish between real data and synthetic data generated by the generator function. In a game-theoretic sense, GA models can be interpreted as zero-sum games. The training of the generator and / or discriminator functions is specifically based on minimizing the cost function.

[0089] By using a GA (Generalized Approach) model, synthetic data can be generated based on a set of training data, and this synthetic data has the same characteristics as the training dataset. Training a GA model can be based on unannotated data (unsupervised learning), making the cost of training the GA model relatively low.

[0090] Recurrent machine learning models are typically machine learning models whose output depends not only on the input values ​​and parameters adjusted through the training process, but also on the hidden state vector, which is based on previous input values ​​used in the recurrent machine learning model. Recurrent machine learning models can, in particular, include additional stored states or additional structures that incorporate time delays or feedback loops.

[0091] The structure upon which recurrent machine learning models are based can be, in particular, a neural network called a recurrent neural network. This recurrent neural network can be described as an artificial neural network in which connections between nodes, arranged sequentially over time, form a directed graph. A recurrent neural network can be interpreted specifically as a directed acyclic graph. A recurrent neural network can be, in particular, a finite-spiking recurrent neural network or an infinite-spiking recurrent neural network (where a finite-spiking network can be expanded and replaced by a strictly feedforward neural network, and an infinite-spiking network cannot be expanded and replaced by a strictly feedforward neural network).

[0092] Training recurrent neural networks can be based, in particular, on algorithms such as BPTT (Backpropagation through time), RTRL (real-time recurrent learning), and / or genetic algorithms. By using recurrent machine learning models, it is possible to use input data including sequences of variable length. This specifically means that the method can be used not only with a fixed number of input datasets (and for each different number of input datasets used as input, training must be performed differently), but also with any number of input datasets. This means that, regardless of the number of input datasets in different sequences, the entire set of training data can be used for training, and the training data is not reduced to training data corresponding to a specific number of input datasets that follow each other.

[0093] Transformer networks are neural network architectures that typically include an encoder, a decoder, or both. In some cases, the encoder and / or decoder consists of multiple corresponding encoding and decoding layers. An attention mechanism is preferably present within each encoding and decoding layer. Sometimes called self-attention, this mechanism associates data points (such as words or pixels) within a set of data points with other data points within that set. For example, a self-attention mechanism can be implemented in a model to examine segments of sound within a recorded audio signal and determine the relative importance of other segments of sound within that audio signal.

[0094] The encoder can be configured to transform the input into a numerical representation. The numerical representation can include a vector for each input token, such as for each sound. The encoder can be configured to implement an attention mechanism such that each vector of a token is influenced by other tokens in the input. The encoder can be configured such that the representation parses the desired output of the Transformer network. The decoder can be configured to transform the input into a sequence of output tokens. The decoder can be configured to implement a masked self-attention mechanism such that each vector of a token is influenced only by other tokens on one side of the sequence. Furthermore, the decoder can be autoregressive, meaning that intermediate results, such as previously predicted sequences of tokens, are fed back. In some examples, the decoder's input is based on or corresponds to the encoder's output. Additionally, the Transformer network can include a classification module that maps the encoder or decoder's output to a set of learned outputs, particularly the outputs of one, more, or all of the following steps:

[0095] - Input audio signals are detected using at least one microphone, wherein at least one microphone is set up and oriented to record sounds from the patient and the imaging system during the imaging examination and to provide them as input audio signals.

[0096] - The patient's voice and the imaging system's system sounds are extracted from the input audio signal using a computing unit.

[0097] - The system's sound is classified using a computing unit.

[0098] - The system sounds are weighted by a computing unit using at least two different weighting factors retrieved from a storage unit.

[0099] - A modification instruction is derived by analyzing the perceptibility of the extracted speech sounds using a computing unit; this modification instruction is relevant to the patient undergoing imaging examination.

[0100] - The computing unit modifies the speech and / or weighted system sounds according to the modification instructions.

[0101] - The computing unit combines potentially modified speech sounds and potentially modified system sounds into an output audio signal.

[0102] In some examples, Transformer model training can be performed in two phases: a pre-training phase and a fine-tuning phase. In the pre-training phase, the Transformer model can be trained on a large corpus to learn the semantics upon which the problem is based. This pre-trained Transformer model can be used with different languages. For the specific applications described herein, fine-tuning can involve further training the Transformer network using medical text and / or medical ontologies such as RADLEX and / or SNOMED with expert-annotated meanings. In some examples, this allows the Transformer model to learn typical relationships and synonyms of medical expressions. The advantages of Transformer networks include their ability to efficiently handle long-range dependencies in the input data due to their attention mechanism. Furthermore, the encoder used in Transformer networks can process data in parallel, saving computational resources during inference. Additionally, the decoder of a Transformer network can iteratively generate sequences of output tokens with high confidence due to autoregression.

[0103] The computing unit may in particular include a code mechanism having at least one trained machine learning model, the code mechanism being configured for extraction, classification, retrieval, and / or combination. The computing unit may in particular include extraction code mechanism, classification code mechanism, weighted code mechanism, retrieval code mechanism, modification code mechanism, and / or combination code mechanism.

[0104] In this document, the term "computer program product" should be understood broadly. A computer program product can be a computer program or include a computer program. In particular, a computer program product has a program code structure that maps to the method steps according to the invention. This allows for the defined and repeatable execution of the method according to the invention and the ability to control the forwarding of the method according to the invention. The computer program product is preferably configured such that a computing unit can execute the method steps according to the invention by means of the computer program product. The program code structure can be loaded in the memory of the computing unit and typically executed by means of the processor of the computing unit through access to the memory. When the computer program product, especially the program code structure, is executed in the computing unit, all embodiments of the described method according to the invention can typically be executed. The computer program product is stored, for example, on a physical, computer-readable medium and / or digitally stored as data packets in a computer network. The computer program product can be a physical, computer-readable medium and / or a data packet in a computer network. Therefore, the invention can also be based on a physical, computer-readable medium and / or data packets in a computer network. Physical, computer-readable media can typically be directly connected to the computing unit, for example, by placing the physical, computer-readable media into a DVD drive or plugging it into a USB port, thereby enabling the computing unit to access the physical, computer-readable media in a particularly readable manner. Data packets are preferably retrieved from a computer network. The computer network can have the computing unit or be indirectly connected to the computing unit via a wide area network (WAN) or (wireless) local area network (WLAN or LAN) connection. Computer program products can, for example, be digitally stored on a cloud server at a storage location on the computer network, and transmitted to the computing unit via the Internet via a WAN and / or via a WLAN or LAN, particularly by invoking a download link pointing to the storage location of the computer program product.

[0105] Features, advantages, or alternative implementations mentioned in the description of the device can also be applied to the method, and vice versa. In other words, embodiments of the method can be improved by means of the features of the device, and vice versa. The device according to the invention is particularly suitable for use in the method.

[0106] In this invention, nouns and pronouns referring to people are generally not limited to a specific gender. Attached Figure Description

[0107] The invention is described and illustrated in detail below with reference to the embodiments shown in the accompanying drawings. In principle, the same structures and units are retained in the following description of the drawings and named using the same reference numerals as when the corresponding structures or units first appear.

[0108] The attached diagram shows:

[0109] Figure 1 The method according to the invention is shown.

[0110] Figure 2 A first embodiment of the method according to the invention is shown, and

[0111] Figure 3 An imaging system according to the present invention is shown. Detailed Implementation

[0112] Figure 1 A flowchart is shown of a method according to the invention for performing imaging examinations on a patient by means of a medical imaging system, particularly a CT or MRT or C-arm imaging system, having steps S100 to S108.

[0113] Method step S100 indicates that an imaging examination using an imaging system is initiated.

[0114] Method step S101 indicates that an input audio signal is detected by means of at least one microphone, wherein the at least one microphone is configured and oriented to record the sounds of the patient and the imaging system during the imaging examination and to provide them as the input audio signal.

[0115] Method step S102 means that the patient's voice and the imaging system's system sound are extracted from the input audio signal by means of a computing unit.

[0116] Method step S103 indicates that the system sounds are classified using a computing unit. The system sounds include sounds from at least two of the following sound categories, wherein the system sounds are divided into at least two sound categories during classification:

[0117] - Background noise interference

[0118] - The scanning-specific movements of the examination bed, especially the supporting parts of the examination bed, and / or the movement of the entire examination bed.

[0119] - MRT coils, especially gradient coils

[0120] - Coolant pumps, especially helium pumps,

[0121] - The rotating part of the CT gantry.

[0122] - The movable part of the CT gantry.

[0123] - X-ray radiator

[0124] - Contrast agent infusion pump,

[0125] - Patient instructions,

[0126] - C-arm,

[0127] - Vital parameter sensors, especially patient monitoring systems, and / or

[0128] - External devices.

[0129] Method step S104 indicates that the classified system sound is weighted by means of a computing unit with at least two different weighting factors called from a storage unit.

[0130] Method step S105 indicates that a modification instruction is obtained by means of a computing unit by analyzing the perceptibility of the extracted speech sounds, the modification instruction being related to the patient undergoing the imaging examination.

[0131] Method step S106 means that the speech sound and / or weighted system sound are modified by means of the computing unit according to the modification instruction.

[0132] Method step S107 means that the possibly modified speech sound and the possibly modified system sound are combined into an output audio signal by means of a computing unit.

[0133] Method step S108 indicates that the combined output audio signal is provided to the user of the imaging system during the imaging inspection by means of the reproduction unit to monitor the imaging inspection.

[0134] Figure 2 A flowchart illustrating a first embodiment of the method according to the present invention is shown.

[0135] Method step S105' indicates that a modification instruction is obtained by means of a computing unit through analysis of the perceptibility of the extracted speech sounds, the modification instruction being relevant to the patient undergoing the imaging examination. The modification instruction is obtained based on a time-resolved weighting factor of the categorized system sounds. Furthermore, obtaining the modification instruction includes analyzing the perceptibility of the speech sounds compared to the weighted system sounds based on the human receiver, particularly the user of the imaging system. The modification instruction includes shifting the playback time of at least one segment of the speech sound. This shift is at most 20 seconds, particularly less than 2 seconds or less than 1 second, but greater than 0 seconds.

[0136] Method step S106' indicates that the speech sound and / or weighted system sound are modified according to modification instructions by means of a calculation unit, wherein the speech sound and / or system sound are modified before combination by means of phase adjustment and / or frequency adjustment and / or amplitude adjustment and / or playback timing adjustment. The playback timing is modified according to the modification instructions to reduce temporal overlap with segments of the system sound. The modification according to the modification instructions includes normalization of the speech sound's volume with respect to a reference value, wherein the reference value is the average volume level of the speech sounds of other patients, and wherein the reference value is invoked before normalization.

[0137] Figure 3 A schematic block diagram of an imaging system 10 according to the invention for performing the method according to the invention is shown.

[0138] The imaging system 10 includes an imaging unit 11, a computing unit 12, a playback unit 13 for the user U, and at least one microphone 14. The imaging system 10 is particularly capable of being a CT, MRT, or C-arm imaging system. Figure 3 A CT imaging system is illustrated exemplarily. Furthermore, the imaging system 10 has an examination area 15 in which the patient P is located during the imaging examination. At least one microphone 14 is present. Figure 3 The center is aligned with the examination area 15 of the patient P. The reproduction unit 13 is wirelessly connected to the computing unit 12.

[0139] Although the details of the invention have been described and illustrated in detail through preferred embodiments, the invention is not limited to the disclosed examples and other variations can be derived by those skilled in the art without departing from the scope of protection of the invention.

Claims

1. A method for performing an imaging examination on a patient using a medical imaging system, particularly a CT or MRT or C-arm imaging system, the method comprising the following steps: - Initiate the imaging examination using the imaging system. - Input audio signals are detected by means of at least one microphone, wherein the at least one microphone is configured and oriented to record the sounds of the patient and the imaging system during the imaging examination and to provide them as input audio signals. - The patient's voice and the imaging system's system sound are extracted from the input audio signal using a computing unit. - The system sounds are classified using the computing unit. - The system sounds are weighted by the computing unit using at least two different weighting factors retrieved from the storage unit. - The computational unit determines a modification instruction by analyzing the perceptibility of the extracted speech sounds, the modification instruction being relevant to the patient undergoing the imaging examination. - The computing unit modifies the speech sound and / or the weighted system sound according to the modification instruction. - The computing unit combines the possibly modified speech sounds and the possibly modified system sounds into an output audio signal. - The combined output audio signal is provided to the user of the imaging system during the imaging inspection by means of the reproduction unit to monitor the imaging inspection.

2. The method according to claim 1, The system sound includes sound from at least two of the following sound categories, and the system sound is classified into the at least two sound categories during classification: - Background noise interference - The scanning-specific movement of the examination bed, particularly the supporting portion of the examination bed, and / or the movement of the entire examination bed. - MRT coils, especially gradient coils - Coolant pumps, especially helium pumps, - The rotating part of the CT gantry. - The movable part of the CT gantry. - X-ray radiator - Contrast agent infusion pump, - Patient instructions, - C-arm, - Vital parameter sensors, especially patient monitoring systems, and / or - External devices.

3. The method according to claim 2, The time-resolved weighting factor is associated with the time segment of the sound according to the at least two sound categories.

4. The method according to any one of the preceding claims, The modification instruction includes shifting the playback time of at least one segment of the voice sound.

5. The method according to claim 4, The shift is a maximum of 20 seconds, especially less than 2 seconds or less than 1 second, but greater than 0 seconds.

6. The method according to any one of claims 4 or 5, The playback time point is modified according to the modification instruction to reduce the overlap in time with the segments of the system sound.

7. The method according to claims 3 and 6, The modification indication is obtained by calculating the time-resolved weighting factor of the classified system sound.

8. The method according to any one of the preceding claims, Determining the modification instruction includes analyzing the perceptibility of the speech sound compared to the weighted system sound based on the human receiver.

9. The method according to claim 8, The user of the imaging system is the receiver.

10. The method according to any one of the preceding claims, The modification according to the modification instruction includes the normalization of the volume of the voice sound with respect to the reference value.

11. The method according to claim 10, The reference value is the average volume level of the speech sounds of other patients, and the reference value is invoked before the normalization.

12. The method according to any one of the preceding claims, The speech and / or system sound are modified before the combination by means of phase adjustment and / or frequency adjustment and / or amplitude adjustment and / or playback timing adjustment.

13. An imaging system for performing the method according to any one of the preceding claims, the imaging system having - Imaging unit, - The computing unit, - The reproduction unit, - The at least one microphone.

14. The imaging system according to claim 13, The computing unit includes a program code mechanism having at least one trained machine learning model configured for the extraction, the classification, the retrieval, and / or the combination.

15. A computer program product comprising a program code mechanism that, when executed by a computing unit, causes the computing unit to perform the method according to any one of claims 1 to 12.

Images