Methods, systems, and media for training data for classification of a position of a catheter relative to a septum

Abstract

The present disclosure relates to position monitoring of medical devices, and more particularly to techniques for enabling automatic monitoring of the position of a catheter relative to a septum. Aspects of the present disclosure include determining training data to be used to train a machine learning algorithm to classify the position of a catheter relative to a patient's septum. Other aspects of the present disclosure include using a trained machine learning algorithm to classify the position of a catheter relative to a patient's septum.

Description

Technical Field

[0001] This disclosure relates to position monitoring of medical devices, and more specifically to techniques for enabling automated monitoring of the position of a catheter relative to a diaphragm. Background Technology

[0002] A catheter is a medical device that can deliver substances into or out of the body. Specifically, a nasogastric / oral catheter (also known as a nasogastric tube) can deliver nutrients, fluids, and medications directly into a patient's stomach.

[0003] Proper positioning of this type of nasogastric tube is crucial to avoiding complications such as aspiration of the contents into the lungs when the tube is not inserted deep enough, or physical perforation of the stomach wall when the tube is inserted too deeply.

[0004] Nasogastric tubes can be equipped with sensors to detect bioelectrical signals within the body. These sensors can measure electrocardiographic and electromyographic signals emitted by nearby muscles. These signals can help assess whether the nasogastric tube has been correctly positioned.

[0005] Typically, the assessment of nasogastric tube placement is performed by a clinician. The clinician can perform a naso-auricular-xiphoid process measurement to estimate the correct tube length to be inserted. Once the tube is inserted into the esophagus, the clinician can monitor detected electrocardiographic and electromyographic signals to determine if the tube is correctly positioned and fine-tune its location as needed.

[0006] Accurate catheter placement is particularly challenging when no electromyographic (EMG) signal is detected; this often occurs when patients are sedated and unable to breathe spontaneously (usually requiring intubation). Electrocardiographic (ECG) signals are typically much stronger than EMG signals, for example, by about 1000 times, which complicates EMG detection. In such scenarios where EMG signals are difficult to detect, ECG interpretation is necessary. However, inexperienced clinicians may struggle to locate and maintain the correct catheter position because ECG signals are more difficult to visually identify and often lack the precision required for fine-tuning catheter placement. Furthermore, in all cases, position monitoring relies on continuous clinical attention.

[0007] The methods for assessing catheter localization need to be improved. Attached Figure Description

[0008] Figure 1 A schematic diagram of a catheter position monitoring system for patients is shown.

[0009] Figure 2 The diagram illustrates a location monitoring system based on some examples;

[0010] Figure 3 This is a flowchart that details the steps performed by the location monitoring system;

[0011] Figure 4 It is a flowchart that details the steps performed by a location monitoring system when preparing training data for training machine learning algorithms;

[0012] Figure 5 An example of bioelectrical signals detected by a catheter is shown;

[0013] Figure 6 and Figure 7 An example of labeled training data is shown;

[0014] Figure 8 An example of a machine learning architecture for a location monitoring system is shown;

[0015] Figure 9 It is a flowchart that details the steps performed by a location monitoring system when deploying machine learning algorithms;

[0016] Figure 10 It is a flowchart that details the steps performed by the position monitoring system when determining the positioning status of the catheter used for the patient;

[0017] Figure 11 This is the first example of a pattern recognition function used to determine the location status;

[0018] Figure 12 This is a second example of a pattern recognition function used to determine the location status;

[0019] Figure 13 The illustration schematically shows a display device used to show clinicians information about the location of catheters. Detailed Implementation

[0020] The purpose of this disclosure is to determine the correct placement of an esophageal catheter within a patient. According to some embodiments of the invention, this purpose is achieved by methods and systems for training and deploying machine learning algorithms that can be used to position the catheter relative to the patient's diaphragm. Specifically, some embodiments relate to methods and systems for determining training data for the machine learning algorithm, and some embodiments relate to methods and systems for classifying the position of the catheter relative to the patient's diaphragm.

[0021] In a first aspect, a method is provided for determining training data to be used to train a machine learning algorithm to classify the position of a catheter relative to a patient's septum. The method includes the steps of: receiving a set of bioelectrical signals detected by a catheter carrying a plurality of electrodes at corresponding positions along the length of the catheter, such that the electrodes are located at corresponding different distances from the patient's septum, the plurality of electrodes being divided into a plurality of electrode pairs, each signal being detected by one of the electrode pairs, each signal including an electrocardiographic (ECG) component; identifying one or more first bioelectrical signals from the set of bioelectrical signals, the first bioelectrical signals being detected by electrode pairs on the catheter determined to be correctly positioned relative to the septum; and dividing the set of bioelectrical signals into at least two subsets of bioelectrical signals, each bioelectrical signal subset... The set includes one or more bioelectrical signals and corresponds to a corresponding set of electrodes associated with the detection of the one or more bioelectrical signals in the subset of bioelectrical signals, wherein the partitioning is performed such that each set of electrodes is a sequence of electrodes placed continuously along the length of the catheter; each subset of the plurality of subsets is labeled, wherein the labeling includes: labeling a subset that includes at least one first bioelectrical signal from the first bioelectrical signals as a subset of signals detected from correctly positioned electrodes, and labeling a subset that does not include any first bioelectrical signals from the first bioelectrical signals as a subset of signals detected from incorrectly positioned electrodes; and including the bioelectrical signal subsets and their corresponding labels in the training data.

[0022] Typically, the available set of bioelectrical signals is recorded assuming the duct is in the correct position, thus the dataset is imbalanced due to a lack of data at incorrect locations. Furthermore, this type of biological data tends to be high-dimensional and scarce, thus increasing the risk of overfitting.

[0023] The inventors have recognized that a subset of signals from a set of bioelectrical signals recorded when the catheter is in the correct position may, in any case, indicate a mispositioned catheter, since at least some of the signals in the set are typically recorded from electrodes in the wrong position (i.e., above or below the septum). This method divides the complete set of signals into several smaller groups (subsets). Each subset is categorized according to its relative position to a defined “correct position” (i.e., relative to one or more first signals). A subset includes one or more bioelectrical signals. The fewer bioelectrical signals in each subset, the greater the amount of diverse training data. Including more than one bioelectrical signal in a subset, by considering information provided by the neighbors of a particular bioelectrical signal, can advantageously utilize the dependencies between bioelectrical signals in the training data, information generated by the physical arrangement of the electrodes on the catheter.

[0024] Using the techniques described in this paper, it is possible to increase the amount of training data for each possible class while balancing that training data.

[0025] In a second aspect, one or more non-transitory computer-readable media are provided, which store instructions executable by one or more processors. When executed, these instructions cause the one or more processors to perform operations substantially mapped to the steps described according to the first aspect.

[0026] In a third aspect, a system is provided. The system includes one or more processors and one or more non-transitory computer-readable media storing first computer-executable instructions. The non-transitory computer-readable media storing the first computer-executable instructions, when executed by the one or more processors, causes the system to perform actions substantially mapped to operations caused by the instructions according to the second aspect or steps of the method according to the first aspect.

[0027] The characteristics of the steps described according to the first aspect also exist in the instructions stored on the non-transitory computer-readable medium according to the second aspect and the actions performed by the system according to the third aspect. Embodiments of the method can be understood to correspond, where appropriate, to embodiments of the non-transitory computer-readable medium and embodiments of the system.

[0028] In some embodiments, the step of labeling a subset of bioelectrical signals excluding the first bioelectrical signal as a subset of bioelectrical signals detected from an incorrectly positioned electrode includes: determining whether the electrode associated with detecting the one or more bioelectrical signals in the subset is positioned above or below the septum; wherein, when it is determined that the electrode is positioned above the septum, the subset of bioelectrical signals is labeled as a subset of bioelectrical signals detected from an electrode located above the septum, and when it is determined that the electrode is positioned below the septum, the subset of bioelectrical signals is labeled as a subset of bioelectrical signals detected from an electrode located below the septum. This allows for a more specific classification of the positioning, distinguishing between subsets with incorrect positions above the septum and subsets with incorrect positions below the septum.

[0029] In some embodiments, the method further includes: enhancing each bioelectrical signal in the subset, wherein enhancing the bioelectrical signal includes at least one of: stretching the bioelectrical signal over time; compressing the bioelectrical signal over time; or changing the amplitude of the bioelectrical signal. Advantageously, overfitting of the machine learning algorithm can be avoided.

[0030] In some embodiments, at least one bioelectric signal includes an electromyographic (EMG) component, and the method further includes applying a filtering algorithm to each bioelectric signal in the set of bioelectric signals, wherein the filtering algorithm is configured to at least reduce the corresponding EMG component from the corresponding bioelectric signal. For example, a band-stop filter may be applied to reduce the EMG component. In other examples, portions of the bioelectric signal that include the EMG component are removed, and the bioelectric signal is truncated such that the remaining bioelectric signals contain no EMG component or have the reduced EMG component. This reduces the likelihood that the trained machine learning algorithm will over-rely on detecting the EMG component to accurately classify the location of the catheter, since, for example, the EMG component may not be present in a patient who cannot breathe spontaneously. This allows the trained machine learning algorithm to be deployed in a wider range of clinical scenarios.

[0031] In some embodiments, applying the filtering algorithm to the bioelectrical signal includes: identifying a plurality of sub-fractions of the bioelectrical signal, each sub-fraction including data detected during a patient's heartbeat; and calculating an average bioelectrical signal based on the plurality of sub-fractions of the bioelectrical signal. For example, calculating an average can be a simple and relatively low-cost method that can reduce the presence of the EMG component that may only exist in a limited number of sub-fractions.

[0032] In some embodiments, each bioelectrical signal from the set of bioelectrical signals includes data detected during multiple heartbeats of a patient, wherein the method further includes: identifying, within each bioelectrical signal from the set of bioelectrical signals, data detected during an intermediate time period between two consecutive heartbeats in the multiple heartbeats; and deleting the identified data from the bioelectrical signals. This can remove data that does not carry information or carries limited information, based on which location information can be determined, which can improve the training of the machine learning algorithm.

[0033] In some embodiments, the step of identifying one or more first bioelectrical signals detected from the set of bioelectrical signals by an electrode pair on the catheter (the electrode pair being the one closest to the septum among a plurality of electrode pairs) includes: detecting the presence and magnitude of an electromyographic (EMG) component among at least one bioelectrical signal from the set of bioelectrical signals; and selecting at least one bioelectrical signal as the one or more first bioelectrical signals based on the magnitude of the corresponding EMG component. For example, the largest EMG component is selected, or an EMG component with a magnitude above a threshold is selected. This can provide an accurate method for identifying the closest electrode and the first bioelectrical signal. For example, the magnitude of the EMG component can be determined using the root-mean-square value (RMS) or peak amplitude.

[0034] In some embodiments, the step of labeling a subset of first bioelectrical signals that does not include the first bioelectrical signal further includes: labeling the subset of first bioelectrical signals that does not include the first bioelectrical signal using a distance measured between the electrode associated with the subset and the electrode in the electrode pair associated with the correctly positioned subset of bioelectrical signals. By providing a more granular classification (labeling) of the location of the subset associated with the first bioelectrical signal, the utility of the output of the machine learning algorithm can be further enhanced, for example, because clinicians can receive an indication of how far the catheter is positioned from an optimal location. This allows clinicians to position the catheter more accurately.

[0035] In some embodiments, the step of labeling a subset that does not include the first bioelectrical signal further includes: labeling the subset that does not include the first bioelectrical signal using a plurality of intermediate electrodes between the electrode associated with the subset and the electrode pair on the catheter, the electrode pair being one of the plurality of electrode pairs positioned closest to the septum. By providing a more granular classification (labeling) of the location of the subset associated with the first bioelectrical signal, the machine learning algorithm can indicate the number of electrodes between a given electrode or a given electrode pair and the electrode pair positioned closest to the septum. For example, this can further infer distances based on the arrangement of electrodes on the catheter. This allows clinicians to receive more useful positioning information and more accurately locate the catheter.

[0036] In some embodiments, the partitioning of the bioelectrical signal set is performed such that at least two subsets of bioelectrical signals partially overlap, such that an electrode associated with detecting one or more bioelectrical signals in the first subset is also associated with detecting one or more bioelectrical signals in the second subset. This can generate more subsets of bioelectrical signals from a given set, thereby increasing the amount of available training data. Furthermore, the overlap increases the granularity of the training data by taking into account a larger number of electrode locations relative to the septum, which can improve the accuracy of the machine learning algorithm.

[0037] In some embodiments, the bioelectrical signal set is divided into at least two bioelectrical signal subsets, such that each subset is associated with a predetermined number of electrodes. This ensures that each subset is of the same size. For example, having subsets of the same size can reduce bias in the training dataset. In the example, each subset is associated with a pair of electrodes. In the example, each subset is associated with a combination of 3, 4, or 5 electrodes.

[0038] In some embodiments, the set of bioelectrical signals is divided into at least two subsets of bioelectrical signals such that the number of subsets of signals identified as associated with correctly positioned electrodes and marked as detected from correctly positioned electrodes is in a predetermined ratio to the number of subsets of signals identified as associated with electrodes positioned above or below the septum and marked as detected from electrodes located above or below the septum. This allows the training data to include a target ratio of correctly positioned markers to incorrectly positioned markers (i.e., markers above or below the septum), which can improve the accuracy of the machine learning algorithm because the training data is balanced according to the requirements of the implementation of the machine learning algorithm. In an embodiment, the predetermined ratio is 1:1. Advantageously, the number of correctly positioned markers is equal to or substantially similar to the number of incorrectly positioned markers, thereby providing a perfectly balanced training dataset.

[0039] In some embodiments, the bioelectric signal includes voltage timing data.

[0040] In some embodiments, receiving the set of bioelectrical signals includes receiving bioelectrical signals from the catheter during patient use. Advantageously, real-time data can be used to continuously expand the training data.

[0041] In some embodiments, receiving the set of bioelectrical signals includes receiving pre-recorded bioelectrical signals. That is, the bioelectrical signals may be derived from historical clinical data.

[0042] In some embodiments, the machine learning algorithm is a neural network.

[0043] In some embodiments, the plurality of electrodes are equidistantly spaced along the length of the catheter. The machine learning algorithm can become more accurate because the regular positional relationship between the electrodes increases the flexibility to subset the bioelectrical signal set. Furthermore, equidistantly spaced electrodes simplify the assessment of the catheter's location compared to electrodes arbitrarily distributed along its length, for example.

[0044] In some embodiments, when the first bioelectrical signal is identified as being associated with an electrode (which is the farthest or closest of the plurality of electrodes relative to the length of the catheter), the set of bioelectrical signals is not used as training data. Advantageously, for example, bioelectrical signals derived from catheters that are positioned too low or too high as a whole can be discarded, thereby improving the effectiveness of the training data.

[0045] In some embodiments, each electrode pair is a pair of adjacent electrodes. That is, the electrodes are adjacent to each other in the electrode sequence on the catheter and are the nearest neighbors.

[0046] In some such embodiments, the one or more first bioelectrical signals include a single first bioelectrical signal detected by the electrode pair positioned closest to the septum among the plurality of electrode pairs on the catheter.

[0047] In a fourth aspect, a method is provided for classifying the position of a catheter relative to a diaphragm, wherein the diaphragm is the patient's diaphragm. The method includes the following steps: (a) receiving a first set of bioelectrical signals detected by a catheter carrying a plurality of electrodes at corresponding positions along the length of the catheter, such that the electrodes are located at corresponding different distances from the patient's septum, the plurality of electrodes being divided into a plurality of electrode pairs, each signal being detected by one of the electrode pairs, each signal including an electrocardiographic (ECG) component; (b) dividing the first set of bioelectrical signals into at least two first bioelectrical signal subsets, each first bioelectrical signal subset including one or more bioelectrical signals and corresponding to a corresponding set of electrodes associated with the detection of the one or more bioelectrical signals in the bioelectrical signal subset, wherein the division is performed such that each set of electrodes is a sequence of electrodes placed continuously along the length of the catheter; (c) inputting the first bioelectrical signal subsets into a machine learning algorithm trained to classify each subset into a plurality of classes, the classes including one or more classes for mislocated electrodes and a class for correctly located electrodes; and (d) receiving a plurality of input classes from the machine learning algorithm. (e) inputting the plurality of input classifications into a pattern recognition function configured to classify the location of the catheter; and (f) using output classification from the pattern recognition function to classify the location of the catheter.

[0048] The method described according to the fourth aspect allows for the processing of bioelectrical signals received from the catheter and the generation of an output classification to categorize the catheter's location. In some examples, such an output classification can be a binary classification (e.g., "correctly placed" or "incorrectly placed") and can allow clinicians to quickly assess whether the catheter needs repositioning or whether it can remain in its current position. In other examples, a more granular output classification can provide additional levels of information about the catheter's location.

[0049] When receiving multiple input classifications from the machine learning algorithm and feeding these multiple input classifications into a pattern recognition function to determine the output classification, the classifications of multiple instances from the received bioelectrical signals can be aggregated to assign a collective classification.

[0050] By aggregating individual categories into collective categories, the decision-making process can become more transparent and interpretable. Stakeholders can investigate the contribution of each instance to the overall classification, which fosters trust and understanding, especially in domains requiring interpretation, such as healthcare. For example, if an error occurs, tracing back through the aggregation process to identify and correct the source of the error at the instance level may be more direct than attempting to diagnose a single piece of machine learning algorithm.

[0051] Additionally, this method benefits from modularity, as the pattern recognition function can be updated without retraining the entire machine learning algorithm. This can improve the uptime of systems employing this method, as it reduces the time spent on offline maintenance or updates.

[0052] Furthermore, training a model to predict aggregate categories may require a large amount of labeled data at the aggregate level, which can be costly or time-consuming to collect. Aggregation methods can leverage more readily available instance-level data or human experts, reducing the need for large amounts of labeled data. Additionally, if the aggregate category is rare or imbalanced in the dataset, the machine learning algorithm may struggle because it lacks sufficient examples to accurately predict such edge cases. In contrast, the appropriate output classification for rare or imbalanced aggregate categories can be directly identified and considered by pattern recognition algorithms.

[0053] In a fifth aspect, one or more non-transitory computer-readable media are provided, which store instructions executable by one or more processors. When executed, these instructions cause the one or more processors to perform operations substantially mapped to the steps described according to the fourth aspect.

[0054] In a sixth aspect, a system is provided. The system includes one or more processors and one or more non-transitory computer-readable media storing first computer-executable instructions. The non-transitory computer-readable media storing the first computer-executable instructions, when executed by the one or more processors, cause the system to perform actions substantially mapped to the operations caused by the instructions according to the fifth aspect or the steps of the method according to the fourth aspect.

[0055] The characteristics of the steps described according to the fourth aspect also exist in the instructions stored on the non-transitory computer-readable medium according to the fifth aspect and the actions performed by the system according to the sixth aspect. Embodiments of the method can be understood to correspond, where appropriate, to embodiments of the non-transitory computer-readable medium and embodiments of the system.

[0056] In some embodiments, the pattern recognition function is configured to output an output confidence value associated with the output classification, the output confidence value indicating the probability of a correct output classification. In addition to the output classification, clinicians can also use the output confidence value to evaluate actions taken with the catheter, such as holding the catheter in its current position or repositioning it.

[0057] In some embodiments, the pattern recognition function is configured to output multiple output classifications, each associated with a corresponding output confidence value. Thus, clinicians can receive the most likely output classification, the second most likely classification, and so on. For example, this can improve the scope of information provided to clinicians, enabling them to make more accurate assessments and decisions regarding catheter placement.

[0058] In some embodiments, the pattern recognition function is configured to compare the plurality of input classifications with a predetermined list, the predetermined list including multiple candidate combinations of input classifications, wherein each candidate combination is associated with a corresponding candidate output classification, and wherein the method further includes determining an output classification based on the predetermined list. For example, having such a predetermined list enables direct processing, error diagnosis and analysis, and interpretation of decision-making processes effectively performed through the pattern recognition function. For example, updating the predetermined list (e.g., by consulting a clinical expert) can be a rapid way to correct, for example, potential classification errors, compared to retraining a machine learning algorithm.

[0059] In some embodiments, the method further includes: calculating a corresponding variance of the input classification for each candidate combination of input classifications in the predetermined list, and identifying the candidate combination of input classifications that produces the smallest variance; wherein the output classification is set as the candidate output classification of the identified candidate combination. For example, calculating the variance of the input classification based on the candidate combination can be a simple process, and the process can be clearly explained to understand the origin of the output classification.

[0060] In some embodiments, calculating the corresponding difference for each candidate combination of the input classification includes: calculating the difference for each input classification based on the corresponding corresponding candidate input classification for each candidate combination of the input classification; and aggregating the calculated differences to determine the corresponding difference for the input classification. In this way, an instance-by-instance comparison of the input classification and candidate combinations can be performed, for example, by directly summing to determine the corresponding difference. This can also be readily interpreted to understand the origin of the output classification, for example.

[0061] In some embodiments, the method further includes: calculating an output confidence value for the output classification based on the minimum difference.

[0062] In some embodiments, the predetermined list includes all possible permutations of the input category as candidate combinations of the input category, wherein the method further includes: identifying the candidate combinations of the input category that are the same as the input category from the list, and wherein the output category is set as the candidate output category of the identified candidate combinations. The predetermined list can be considered as a lookup table.

[0063] In some embodiments, the first bioelectrical signal set is a first set of a plurality of bioelectrical signal sets, each set corresponding to a bioelectrical signal detected during a corresponding time period; and the method further includes: performing steps (b) to (d) for each bioelectrical signal set to receive a plurality of input classifications associated with each bioelectrical signal set from the machine learning algorithm; forming an aggregation of input classifications, the aggregation of input classifications including the plurality of input classifications received for each bioelectrical signal set; and inputting the aggregation of input classifications to the pattern recognition function, and receiving the output classification as an output from the pattern recognition function based on the aggregation of input classifications. This can effectively achieve time-averaged output classification, which can reduce sensitivity to, for example, instantaneous or short-term classification changes, such as those that might be caused by a bioelectrical signal set in the plurality of bioelectrical signals having a different classification set than the remaining bioelectrical signals in the plurality of bioelectrical signals. This could be a more useful output for clinicians as it can prevent unnecessary changes in catheter position that might be caused by instantaneous classification changes.

[0064] In some embodiments, the pattern recognition function computes an intermediate output classification for each plurality of input classifications in the aggregation of input classifications, and computes the output classification based on the majority of the intermediate output classifications. This can be a simple method for computed output classifications that is still time-averaged and resistant to transient changes in output classifications, but does not necessarily require more complex computational steps.

[0065] In some embodiments, the method further includes: calculating a corresponding intermediate confidence value for each intermediate output classification; and calculating an overall confidence value for the output classification based on the intermediate confidence values. For example, similar to the overall confidence value calculated for a single set of bioelectrical signals described above, this overall confidence value can provide clinicians with an indication of the reliability of the output classification and the probability of error.

[0066] In some embodiments, the first set of bioelectrical signals is a first set of a plurality of bioelectrical signal sets, each set corresponding to a bioelectrical signal detected during a corresponding time period; and the method further includes: performing steps b) to d) for each set of bioelectrical signals to receive from the machine learning algorithm a plurality of input classifications associated with each set of bioelectrical signals; forming an aggregation of input classifications, the aggregation of input classifications including the plurality of input classifications received for each set of bioelectrical signals; and calculating an average set of input classifications based on the aggregation of input classifications; and inputting the average set of classifications to the pattern recognition function, and receiving the output classification as an output from the pattern recognition function based on the average classification. This can allow the use of a pattern recognition function configured for use with a single set of input classifications, for example, by effectively performing time averaging or otherwise aggregating input classifications into an average set of input classifications. This can reduce the sensitivity of the output classification to transient or short-term changes in the input classification, for example, this can help clinicians determine whether the catheter has been properly placed.

[0067] In some embodiments, the pattern recognition function is configured to output one of the following: a first output classification corresponding to the catheter being positioned too low relative to the patient's septum; a second output classification corresponding to the catheter being correctly positioned relative to the patient's septum; and a third output classification corresponding to the catheter being positioned too high relative to the patient's septum. In an example, for instance, the pattern recognition function is configured to output the distance to the correct positioning of the catheter. In an example, for instance, the pattern recognition function is configured to output multiple electrodes based on the correct positioning of the catheter. For example, the pattern recognition function may output one or more of the above output classifications.

[0068] The above embodiments allow for the insertion of a catheter into a patient and continuous, automated monitoring of its position. This makes it easier to locate and / or maintain the correct position of the catheter. Automated monitoring of the catheter avoids the need for additional procedures (e.g., X-rays) to confirm its location, which are often time-consuming and expensive. For example, reducing the risk of mispositioning the catheter can reduce the risk of feeding tube puncture of the duodenum or the risk of aspiration, which can lead to pneumonia.

[0069] These embodiments also benefit devices that operate in conjunction with the catheter. For example, in clinical settings employing ventilation techniques such as neuroally-adjusted ventilatory assistance (NAVA), a reliable, continuous source of bioelectrical signals from the catheter is required. Accurate catheter positioning, as described in these embodiments, allows this reliable bioelectrical signal source to be provided to the ventilation system. Furthermore, false triggering due to signal leakage caused by mispositioning of the catheter can lead to false alarms, inaccurate monitoring of respiratory rate and / or tidal volume, and false triggering of the ventilation system, resulting in patient asynchrony with the ventilation system. Accurate catheter positioning resolves these problems.

[0070] The invention will now be described more fully below with reference to the accompanying drawings, in which embodiments of the invention are illustrated.

[0071] Figure 1 An overview of a patient 50 with an esophageal catheter 30 is schematically shown, which extends along the patient's esophagus 18 into the patient's stomach 15 and records bioelectrical signals 25, such as electrocardiogram (ECG) signals from the heart 22 and electromyography (EMG) signals from the diaphragm 20.

[0072] The catheter 30 includes multiple electrodes, labeled e1 to e9, at corresponding locations along its length. In this example, nine electrodes are present; however, it should be understood that different numbers of electrodes may be present in other examples. Figure 1 In the example, electrode e 1-9 A linear array of electrodes is arranged at equal intervals, such that each electrode is spaced substantially the same distance from its two adjacent electrodes. Each electrode is located at a different distance from the septum 20. When the electrodes are oriented towards the patient's esophagus, for example when the electrodes are located... Figure 1 When the electrode is above the indicator axis A, it can be considered that the electrode is located above the diaphragm. Similarly, when the electrode is facing the patient's stomach, for example when the electrode is located above the diaphragm... Figure 1 When the electrode is below indicator axis B, it can be considered that the electrode is below the diaphragm. That is, above and below should be understood as the degree to which the catheter is inserted into the stomach relative to the diaphragm. Similarly, when the electrode is between indicator axes A and B, it can be considered that the electrode is aligned with the diaphragm or located near the diaphragm.

[0073] Using paired electrodes e n and e x≠n and adjacent paired electrodes (e.g., e) n and e n-1 or e n and e n+1This is used to perform bipolar measurements and generate bioelectric signals. The following discussion considers bioelectric signals generated by nearest neighbor pairs, but more generally, non-near neighbor pairs can also be used or alternatively.

[0074] Therefore, these nine electrodes generate a set of bioelectrical signals, which in this example is based on eight-channel voltage timing data of eight adjacent pairs. The set of bioelectrical signals can be understood below as multiple signals, where each of these signals corresponds to a corresponding pair of electrodes of catheter 30. Figure 5 An example of a set of bioelectrical signals detected by the electrodes of catheter 30 is shown. Figure 5 In the examples, the bioelectrical signals were recorded as voltage time-series data with a time resolution of 2 kHz and an amplitude resolution of 20 bits. Examples of the ECG components of the signals are labeled 25a, while examples of the EMG components are labeled 25b.

[0075] The bioelectrical signal 25 typically includes contributions from both ECG and EMG signals. The heart 22 emits an ECG signal whenever the patient's heart beats. Given the proximity of the heart 22 to the diaphragm 20, the ECG can be a reliable source of information from which the location of the diaphragm can be inferred. The ECG signal includes the PQRST complex, which describes the pattern of the heart's bioelectrical activity during the cardiac cycle. The initial peak (P wave) represents the propagation of a bioelectrical impulse initiated by the sinoatrial node located in the right atrium, passing through both atria of the heart and causing them to contract. The Q, R, and S waves together depict the subsequent ventricular contraction triggered by the P wave. The terminal peak (T wave) represents ventricular recovery, preparing for the next cardiac cycle. An EMG signal is emitted during diaphragmatic contraction (e.g., during respiratory effort); that is, when the patient breathes spontaneously, the diaphragm contracts and emits an EMG signal.

[0076] The position monitoring system 100 receives signals detected by electrodes, thereby monitoring the position of the catheter 30 within the patient 50. Later, based on... Figure 2 The position monitoring system 100 will be described in more detail. However, generally, the position monitoring system is responsible for determining whether the catheter 30 is correctly or incorrectly positioned and notifying the clinician. In this example, the position monitoring system 100 outputs positioning information describing whether the catheter 30 is correctly or incorrectly positioned to the display unit 200. This will be discussed later according to... Figure 13 The output positioning information and the display of such information are described in more detail. The substance delivery device 210 administers fluids, nutrients, and medications to the patient via the catheter 30. For example, the substance delivery device 210 may wait for the catheter 30 to be correctly positioned before delivering fluids, nutrients, or medications. For example, the correct positioning of the catheter can be confirmed by the position monitoring system 100 and / or by a clinician. It should be understood that... Figure 1Other components, not shown, can similarly utilize positioning information from the position monitoring system 100. For example, if the catheter is determined to be in the wrong position, the ventilation control system can temporarily disable ventilation to reduce the risk of erroneous breathing, which could be caused, for example, by noise in the received signal due to the incorrect position.

[0077] Now go to Figure 2 The location monitoring system 100 will be described in more detail below. The location monitoring system 100 includes a processor 140 that interacts with and executes data and program routines stored in a memory 110. In this example, the memory 110 includes a data storage device 111 for storing clinical data 112 and training data 114, and stores program routines 115 that allow execution of a clinical data to training data process 116, a machine learning algorithm training process 118, a deployed machine learning algorithm process 120, and a location determination process 122.

[0078] As an overview, the position monitoring system 100 receives clinical data 112 including bioelectrical signals detected by the catheter, processes the clinical data 112 to create training data 114 for training a machine learning algorithm, trains the machine learning algorithm based on the training data 114, and deploys the machine learning algorithm 120. Once deployed, the trained machine learning algorithm 120 can be used to monitor the position of the catheter 30 when it is inserted into the patient 50, based on bioelectrical signals 25 detected by the catheter 30. This overview is provided by Figure 3 The flowcharts in the figure are depicted, and data 112, 114 and processes 116, 118, 120, 122 will be described in more detail later with reference to the figures.

[0079] More generally, the location monitoring system 100 may include any suitable arrangement or configuration of one or more computing devices that allows for the storage of clinical data 112 and training data 114, as well as the execution of a clinical data to training data process 116, a machine learning algorithm training process 118, a deployed machine learning algorithm process 120, and a location determination process 122. For example, the storage device may include remotely accessible cloud storage or locally accessible solid-state drive or hard disk storage, or storage on, for example, a local networked server. For example, instruction programs for the aforementioned processes may also be stored on the same storage device, or on their respective respective storage devices. Similarly, for example, the execution of the aforementioned processes may be performed on a local processor (e.g., the local processor of a personal computer located in a clinical environment), on a local networked server, or on a cloud-based server. Each process may be executed on a correspondingly different processor or on the same processor. The location monitoring system 100 may include circuitry configured (using one or more non-transitory computer-readable media) to implement the functions described herein. For example, processors suitable for executing instructions include general-purpose microprocessors and special-purpose microprocessors, as well as a single processor or one of multiple processors or cores in any type of computer. The processor may be supplemented by or incorporated into an ASIC (Application-Specific Integrated Circuit). Those skilled in the art will understand that the exemplary embodiments described above can be implemented by any suitable software, hardware, or hardware configuration or combination thereof. Exemplary hardware platforms for implementing the exemplary embodiments may include, for example, Intel x86-based platforms with compatible operating systems, Windows OS, Mac platforms and MAC OS, and mobile devices with operating systems such as iOS and Android. In yet another example, an exemplary embodiment of the described method may be embodied as a program containing lines of code stored on a non-transitory computer-readable storage medium that can be executed on a processor or microprocessor at compile time.

[0080] Figure 2The location monitoring system 100 includes a catheter data input module 130. The catheter data input 130 receives bioelectrical signal data from catheter 30, from which EMG and / or ECG signals 25 of the patient 50 have been detected. The catheter data input 130 can be arranged such that catheter 30 interfaces with a personal computing device, such as that of the location monitoring system 100, and a program running on the personal computing device of the location monitoring system 100 processes the data received from catheter 30. In other examples, an intermediate peripheral device (e.g., an oscilloscope) receives data from catheter 30, and an onboard processor (e.g., a field-programmable gate array (FPGA)) can perform processing on the received data. For example, catheter data input 130 can perform data normalization or data cleaning on the data received from the catheter. The catheter data input 130 can then provide the received data to processor 140 for storage, such as as clinical data 112, or for use by the processes of the location monitoring system 100. The use of the data received from catheter 30 will be described in more detail later.

[0081] Figure 2 The location monitoring system 100 includes a display data output module 124. The function of the display data output module 124 will be described in more detail later, but it typically receives information to output to the display unit 200. Therefore, the display data output module 124 may include graphics processing capabilities that can, for example, render graphics related to the output of processes from the location monitoring system (e.g., location determination process 122). Clinicians can use the information output by the display data output module 124 and the display unit 200.

[0082] An example of the overall process flow performed by the location monitoring system 100 will now be described. Steps S1 to S6 of the overall process flow are as follows: Figure 3 As shown.

[0083] At item S1, clinical data 112 is obtained, and clinical data 112 includes data obtained from a catheter (e.g., Figure 1 The set of bioelectrical signals detected by the electrodes of the catheter 30 described in the text.

[0084] Typically, the purpose of item S1 is to acquire data that can be used to train machine learning algorithms, as described below. Clinical data 112 can be data recently collected from patients or previously collected historical data. In some embodiments, data is collected only from the adult population, thereby defining adults as the target population for using machine learning algorithms. In other embodiments, for example, data can be collected only from the pediatric population, thereby defining children as the target population for using machine learning algorithms. In a further embodiment, the clinical data includes a mixture of adults and children, allowing a single model to be trained and applied independently of the patient's age. In yet another embodiment, for example, data recorded during use on animals can be used as training data for use on human patients, or for training and use on a veterinary target population. In a further embodiment, clinical data 112 is collected during use of catheter 30. For example, clinical data collected during use can be added to an existing set of historical clinical data 112, or the clinical data can be used to form a new set of clinical data 112 based on, for example, specific training needs. For example, clinical data can be obtained from federated computing systems, where data is shared among nodes in a network in a standardized or coordinated manner that protects patient privacy.

[0085] At item S2, clinical data is processed to generate training data, which corresponds to... Figure 2 The clinical data to training data process 116 is shown. Both clinical data and training data comprise a set of bioelectrical signals detected by electrodes of the catheter. As used herein, clinical data refers to bioelectrical signals prior to processing to form training data. The inventors have observed that the clinical dataset 112 acquired at item S1 (e.g., a historical dataset from the intensive care unit) is typically recorded when the catheter is in the correct placement position. Therefore, from a machine learning perspective, the clinical dataset may be imbalanced due to a lack of data recorded when the catheter is in the incorrect placement position. Generally, the purpose of item S2 is to improve the training data available for the machine learning algorithm trained at item S3. The steps taken at item S2 to achieve this are as follows: Figure 4 The above is presented, and will be described in more detail later with reference to the accompanying drawing.

[0086] At term S3, training data is used to train the machine learning algorithm, corresponding to Figure 2 The machine learning algorithm training process 118 is described in section S3. Typically, the purpose of item S3 is to train the machine learning algorithm so that, given a bioelectrical signal detected by the electrodes of catheter 30, the machine learning algorithm can determine whether the catheter is misplaced or correctly placed relative to the septum, or classify it accordingly. The output of item S3 is the trained machine learning algorithm.

[0087] At item S4, the trained machine learning algorithm is deployed, corresponding to Figure 2 The machine learning algorithm deployment process 120 is described in the text. Typically, the purpose of item S4 is to use the bioelectrical signals detected by the electrodes of catheter 30 and received by the position monitoring system 100 at the catheter data input module 130 as input to the trained machine learning algorithm, and to output position information through the machine learning algorithm.

[0088] At item S5, the location information output by a machine learning algorithm is used to determine the positioning status of the catheter, corresponding to Figure 2 The location determination process described in section 122. Typically, the purpose of item S5 is to determine how the output of the machine learning algorithm should be used to determine the location status. The steps taken at item S5 are as follows: Figure 10 As shown in the accompanying drawing, and will be described in more detail later with reference to the drawing.

[0089] At step S6, the location status is output to the user. Typically, the purpose of step S6 is to output the location status to the clinician, and it corresponds to the operation performed by processor 140. Figure 2 The display data output module 140 and display unit 200 perform the following processes. For example, the positioning status can indicate whether the catheter is correctly or incorrectly positioned. In other examples, the positioning status can indicate that the catheter is incorrectly positioned (due to being too high or too low). In a further example, the positioning status can indicate that the catheter is incorrectly positioned and, for example, indicate the estimated distance from the correct positioning.

[0090] Steps S1 through S6 do not need to be performed in strict order. For example, steps S1 and S2 may occur multiple times before step S3. For example, steps S1, S2, and S3 may occur multiple times before step S4. Steps S1 through S6 do need to be performed sequentially. For example, the first step (e.g., step S1) may occur at an earlier time, while the second step (e.g., step S2) may occur at a significantly later time (e.g., hours, weeks, or months later).

[0091] Processing clinical data

[0092] The process of processing clinical data to prepare training data will now be described, corresponding to... Figure 2 Item S3 in the middle. Figure 4 A flowchart detailing the steps taken to process clinical data and prepare training data based on it is shown. The flowchart includes items S1 and S3 to help illustrate the processes preceding and following item S2.

[0093] At item S1, as previously described, clinical data is acquired, including a set of bioelectrical signals detected by the electrode pairs on the catheter. As previously mentioned, the inventors have recognized that such data often presents too much correct catheter placement and too little incorrect catheter placement. The steps of item S2, described below, can help address this problem.

[0094] In step S202, a data preparation process is performed on the bioelectrical signals. Generally, this may include some form of data normalization and / or data cleaning, for example, to ensure the consistency of the bioelectrical signal format under consideration, or to optimize the data for use as training data, for example, by reducing the size of the data, or by removing data portions that contribute relatively little to the training process.

[0095] In this example of the process, the bioelectrical signal is segmented into individual heartbeats. This can be performed based on, for example, the identification of the PQRST complex or its constituent components. For instance, the R wave can be detected and used to identify a specific heartbeat. Thus, the bioelectrical signal can contain bioelectrical activity associated with a single heartbeat detected by the electrode pair. In other examples, the bioelectrical signal comprises multiple heartbeats, and either segmentation does not occur or is not segmented into individual heartbeats.

[0096] In this example of the procedure, data from a time window surrounding the R wave (e.g., within a ±300 ms window) is retained, while data detected in the intermediate period between two consecutive heartbeats in a multi-beat sequence is removed. This ensures capture of the entire PQRST complex within the ECG signal while removing the signal between heartbeats, which contain relatively little (if any) usable data related to electrode location. This can occur in conjunction with segmentation, where individual heartbeats are isolated and irrelevant data outside the PQRST complex is removed, or not in conjunction with segmentation, where multiple heartbeats are recorded but irrelevant data between the PQRST complexes is removed.

[0097] In another example of process 202, a filtering algorithm is applied to reduce the presence of the EMG component within the bioelectrical signal. The filtering algorithm may include calculating a moving average of the heart rate, where the EMG component exists only on a subset of that heart rate. Thus, the moving average reduces the influence of the EMG component. Other filtering algorithms can be used to reduce the EMG component, such as applying a band-stop filter to the EMG spectrum (frequency around 60–100 Hz), or, for example, completely excluding heart rate beats where the EMG component is detected. Removing the EMG component ensures that the machine learning algorithm learns how to predict catheter location based solely on the ECG component, rather than relying on the EMG component. Therefore, the machine learning algorithm can correctly determine the location of a catheter inserted into a patient who is taking sedatives and is therefore unable to breathe spontaneously (typically requiring intubation).

[0098] In another example of process 202, data augmentation is applied to at least some of the bioelectrical signals. For example, augmenting the bioelectrical signals includes at least one of the following: stretching the bioelectrical signals over time; compressing the bioelectrical signals over time; or altering the amplitude of the bioelectrical signals. Advantageously, the number of data-augmented samples can be increased synthetically. Synthetic data augmentation can be crucial for machine learning classification, especially for biological data, which is often high-dimensional and scarce. Data augmentation can also be useful in preventing overfitting and making machine learning algorithms more robust.

[0099] At item S203, a first bioelectrical signal is identified from the set of bioelectrical signals that is associated with an electrode pair correctly positioned relative to the septum. As used herein, in the context of a first bioelectrical signal, “first” is used as a marker and does not imply any type of location information about the bioelectrical signal. Generally, multiple first bioelectrical signals can be identified, each of which is associated with a corresponding electrode pair that is considered correctly positioned relative to the septum. For example, the electrode pair closest to the septum can be identified, but other electrode pairs can be considered to be positioned close enough to the septum that they also represent correctly positioned electrodes. For clarity, the following discussion considers a single first bioelectrical signal.

[0100] In this example, identifying a first bioelectrical signal can be performed by detecting the presence of an EMG component within the bioelectrical signal. As an illustrative example, an EMG component can be detected within the bioelectrical signals of channels 2-4, but not within channels 1, 5, 7, or 8. The magnitudes of the detected EMG components are compared, and the largest EMG component is identified. For example, as an illustrative example, channel 5 has the largest EMG component. The magnitude of the EMG component can be an amplitude measurement, such as an interpeak measurement or root mean square amplitude. The associated channel is identified as the electrode pair closest to the septum. When channel 5 is identified as having the largest EMG component, electrodes e5 and e6 are identified as the electrode pair closest to the septum. In this example, the method is extended to identify multiple first bioelectrical signals. For example, a bioelectrical signal with an EMG component above a threshold magnitude can be considered as detected by a correctly positioned electrode. The threshold magnitude can be relative to an absolute measurement of the EMG component or relative to the magnitude of the EMG component of other bioelectrical signals. For example, an EMG component within 10% of the maximum detected EMG component can be considered as associated with a correctly positioned electrode. The threshold size can be determined by the precise arrangement of multiple electrodes on the catheter or by the characteristics of the electrodes. In other examples, the identification process for one or more first bioelectrical signals may include examining which of the received bioelectrical signals is associated with a marker or tag that identifies it as a bioelectrical signal detected by an electrode on the catheter determined to be correctly positioned relative to the septum.

[0101] This identification process using the EMG component occurs, for example, before the aforementioned filtering algorithm in step S201, enabling the identification of the first bioelectrical signal, after which the EMG component is removed. In some examples, further filtering of the bioelectrical signal can be performed to improve the visibility of the EMG component, thereby identifying one or more first bioelectrical signals. For example, a bandpass filter can be applied at the characteristic frequency of the EMG component, and a bandstop filter can be applied at the characteristic frequency of the ECG component, thereby reducing the ECG component and relatively enhancing the EMG component.

[0102] In an alternative embodiment, a bioelectrical signal from the set of bioelectrical signals may (e.g., in historical clinical data) be labeled as the signal detected by the electrode pair closest to the septum. In this embodiment, further analysis may not be required to determine the first bioelectrical signal, other than identifying which bioelectrical signal in the historical clinical data includes the label.

[0103] At step S204, the set of bioelectrical signals is divided into at least two subsets, each of which includes signals associated with consecutively placed electrodes. For example, channels 1-3, corresponding to electrode pairs e1 and e2, e2 and e3, and e3 and e4, can form subsets comprising signals associated with consecutively placed electrodes. Similarly, channels 2-4, 3-5, 4-6, 5-7, and 6-8 can be grouped together. Accordingly, subsets can overlap, such that channels are represented by multiple subsets. Allowing overlap increases the number of subsets that can be derived from the set of bioelectrical signals. This improves the number of electrode locations represented by each subset, which allows for more accurate estimation of catheter locations using trained machine learning algorithms, as dependencies between adjacent (e.g., channel 2 is adjacent to channels 1 and 3) bioelectrical signals can be learned and utilized by the machine learning algorithm. When subsets include signals from multiple channels, the classification problem is effectively transformed into a multi-channel temporal classification problem. The data augmentation process discussed above can generally be applied to each bioelectrical signal within a subset.

[0104] In examples of detecting bioelectrical signals using non-adjacent electrode pairs (e.g., e1 and e3, e2 and e4, e3 and e5, etc.), a subset of signals associated with a set of consecutively placed electrodes can still be prepared when partitioning the set of bioelectrical signals. For example, a subset could include bioelectrical signals detected by electrode pairs e1 and e3, as well as those detected from electrode pairs e2 and e4. Therefore, the set of electrodes associated with this subset is e1, e2, e3, and e4, a sequence of electrodes placed consecutively along the length of the duct. A second subset could include bioelectrical signals detected from electrode pairs e5 and e7, as well as those detected from electrode pairs e6 and e8. Therefore, the set of electrodes associated with this second subset is e5, e6, e7, and e8, also a sequence of electrodes placed consecutively along the length of the duct. In both cases of detecting bioelectrical signals using adjacent or non-adjacent electrode pairs, these subsets can overlap, resulting in electrodes or electrode pairs being found in multiple subsets.

[0105] At item S205, each subset of multiple subsets is labeled according to whether the associated electrode is correctly positioned relative to the diaphragm or incorrectly positioned relative to the diaphragm.

[0106] In one example of the labeling process, a subset including the first bioelectrical signal is labeled as detected from correctly positioned electrodes. That is, the electrode closest to the septum is considered correctly positioned. A subset excluding the first bioelectrical signal is labeled as detected from incorrectly positioned electrodes. This provides a simple way to label all training data, as only the closest electrode pairs need to be identified. Figure 6 This is illustrated schematically. In Figure 6In the example, the septum is closest to electrodes e4 and e5, which correspond to channel 4. Therefore, channel 4 is identified as the first bioelectrical signal, for example, based on having the largest EGM component. Any subset including channel 4 is marked as detected from the correctly positioned electrode; that is, in this case, subsets 2, 3, and 4 are marked "C". Subsets 1, 5, and 6 that do not include the first bioelectrical signal are marked as misplaced and indicated by the mark "NC".

[0107] In another example of the labeling process, subsets are labeled based on whether they do not include the first bioelectrical signal and are associated with electrodes positioned below or above the septum. Figure 7 Use and Figure 6 The same data illustrated here illustrates this point. Similarly, channel 4 is associated with the electrode closest to the septum, and subsets 2, 3, and 4 are identified as correctly positioned. Because it is associated with the electrode above the septum, subset 1, including channels 1-3, is labeled "H". Because it is associated with the electrode below the septum, subsets 5 and 6, including channels 5-7 and 6-8 respectively, are labeled "L". The physical arrangement of the electrodes can be used to determine whether the electrodes are above or below the septum. For example, based on the catheter and electrode design, electrode 1 is oriented proximally relative to electrodes 4 and 5, so it is known that electrode 1 is above the septum if electrodes 4 and 5 are correctly positioned. Similarly, electrode 9 is oriented distally towards the catheter and is inferred to be below the septum. Therefore, in this example, the configuration of the catheter's electrodes is used to determine a labeling strategy, such as determining whether an electrode or electrode pair is above or below the septum.

[0108] In some embodiments, when the first bioelectrical signal is identified as being associated with an electrode (which is the farthest or closest of the plurality of electrodes relative to the length of the catheter), the set of bioelectrical signals is not used as training data. In other words, using Figure 5 In the example, if the first bioelectrical signal corresponds to channel 1 or 8, the entire set of signals (channels 1 to 8) can be ignored when determining the training data. Advantageously, the set of bioelectrical signals recorded from catheters positioned entirely below or above the septum can be ignored, which in turn can improve the training data, since signals from catheters positioned in this manner may not represent correctly positioned electrodes. Specifically, if EMG components are used to determine the first signal, one of channels 1 or 8 is typically identified as having the largest EMG component (because that channel is closest to the septum). However, since the entire catheter is positioned above (in the example of channel 8 with the largest EMG component) or below (in the example of channel 1 with the largest EMG component), none of these channels can truly represent a correctly positioned electrode pair, and thus such a dataset can be ignored.

[0109] Importantly, the set of bioelectrical signals representing correctly positioned catheters has been divided into subsets, some of which are marked as mispositioned because all electrodes associated with these subsets are located near the septum externally (either above or below). This increases the number of bioelectrical signals that can be considered mispositioned relative to the original clinical dataset. This improves the effectiveness of the data used to train machine learning algorithms.

[0110] It should be understood that the labeling strategy can be modified based on, for example, the physical arrangement of the electrodes (e.g., the spacing between the electrodes), the number of electrodes and the regularity of the electrode spacing, the size of the subset, or the degree of overlap of the subset.

[0111] In some examples, knowledge of the physical arrangement of catheters and electrodes can be used as additional markers during the labeling process to indicate whether a marker is correctly or incorrectly positioned. For example, if a subset of bioelectrical signals does not include a first bioelectrical signal, the subset can be labeled using distances measured between the electrodes associated with that subset and the electrode pair associated with the first signal. For instance, this distance could be between the geometric midpoint of the electrode pair associated with the first signal and the geometric midpoint of the nearest electrode pair that does not include the first bioelectrical signal, thus providing an indication of the distance between the electrodes associated with the subset that does not include the first bioelectrical signal and their correct positioning. The geometric midpoint can be considered as a midpoint along the length of the catheter between the respective electrodes. Other measurements are also possible, such as the distance between the geometric midpoints of all associated electrodes and the geometric midpoint of the correctly positioned electrode pair, or the distance between the nearest electrode among the associated electrodes and the nearest electrode among the correctly positioned electrode pair.

[0112] At item S206, the subset and its corresponding labels are stored as training data. Many procedures can be performed during the storage phase.

[0113] In some examples, downsampling can be performed on bioelectrical signal data. For instance, when the original sampling rate is 2 kHz, the signal can be downsampled to 400 Hz. A rate of 400 Hz has been found to be a safe upper limit because frequencies below 400 Hz have been shown to capture critical information about the PQRST complex within ECG signals. This reduction can decrease the size of the input data, thereby increasing training speed without reducing, or within tolerable limits, the accuracy of the trained machine learning algorithm.

[0114] In some examples, data balancing techniques can be performed. Data balancing of the labeled data helps match larger classes with smaller classes. For example, after labeling, there might be more subsets of bioelectrical signals labeled "H / placed too high" compared to "L / placed too low" or "placed correctly." Therefore, "H / placed too high" can be reduced to match the size of "L / placed too low" or "placed correctly," for example, by drawing random samples from the larger classes. Similarly, for example, all classes can be downsampled to match the size of the smallest class. By ensuring that each class has enough examples (represented by the corresponding labels) for machine learning to effectively learn how to identify, producing a flat distribution of class populations can help improve the accuracy of trained machine learning algorithms.

[0115] In some examples, dataset reduction can be performed based on the self-similarity of bioelectrical signals, or by intermittently sampling bioelectrical signals without calculating their self-similarity. It has been observed that consecutive heartbeats within the same set of bioelectrical signals are generally highly similar, while temporally separated heartbeats exhibit greater variation. For example, the change in heartbeat at the start of a 10-minute recording compared to the end of a 10-minute recording is typically greater than that of a consecutive preceding or subsequent heartbeat. For example, discarding similar consecutive heartbeats while retaining temporally spaced heartbeats can reduce the training data size without loss of accuracy or with tolerable loss of accuracy, as the reduced training data still captures a diverse set of heartbeats. In some embodiments, for example, autocorrelation techniques are employed to identify particularly different heartbeats within the bioelectrical signals to improve heartbeat diversity. The reduced training data size can improve processing speed, for example, by reducing the training time required before a trained machine learning algorithm can be deployed.

[0116] Use training data to train machine learning algorithms.

[0117] An example of training a machine learning algorithm using the training data prepared at item S2 will now be described, corresponding to item S3. Generally, the machine learning algorithm to be trained can be an algorithm suitable for solving time series classification problems. The general task of time series classification problems is to predict the label of the entire time series based on temporal patterns within the time series. More specifically, the location monitoring system 100 uses a machine learning algorithm capable of outputting location labels, such as correctly located, incorrectly located, too high, or too low, based on the bioelectrical signals input to the catheter 30.

[0118] In the examples, decision trees, k-nearest neighbors, and support vector machines can be suitable for solving time-series classification problems. In other examples, the machine learning algorithm to be trained is a deep learning model (e.g., a convolutional neural network), which is particularly well-suited for solving time-series classification problems because it can extract relevant features from the raw ECG signal without manual feature engineering and can capture long-term dependencies in the time sequence. Specifically, variants of convolutional neural networks can be used, such as the known RandOm Convolution Kernel Transform (ROCKET) and Mini-ROCKET. Algorithms utilizing the remaining network architecture can also be used, for example.

[0119] In some examples, transfer learning methods can be employed to leverage pre-trained neural networks that have already been trained on large datasets to perform conceptually similar tasks. For instance, an existing time-series signal classifier can be retrained using the training data provided at item S2. This can, for example, reduce the overall computational cost of training machine learning algorithms.

[0120] Figure 8 The architecture of a machine learning algorithm for training on training data is illustrated schematically, based on an example. The architecture includes a feature extractor 302, followed by a linear classifier 304 that outputs a classification 306.

[0121] Training can be performed in a software environment. For example, one such software environment is a Jupyter notebook.

[0122] exist Figure 8 In this example, training data 114, including a set of bioelectrical signals and associated labels, is initially passed to feature extractor 302. In this example, this is a ROCKET or MiniROCKET convolutional model that uses a predetermined number of convolutional kernels to extract features. In other examples, other feature extraction models may be used.

[0123] In the example using the ROCKET model, during the generation phase, convolutional kernels are generated based on random sampling of kernel attributes (including kernel length, weights, bias, dilation, and padding) according to a predefined distribution. These kernels are then convolved with the training data to extract features from each kernel. In the example using the MiniROCKET model, during the generation phase, instead of random sampling of kernel parameters, the dilation, length, and padding parameters are fixed, the weights are restricted to two possible values, and the bias is derived directly from the convolution output.

[0124] In the example using the ROCKET model, during the transformation phase, each generated kernel is applied to each time series of input training data, resulting in a corresponding feature map. Two aggregated output features are then computed based on this feature map: Positive Proportion Value (PPV) and a maximum value or global max pooling. The PPV metric can be considered to capture the frequency with which the time series training data exhibits a positive response to the kernel. The maximum value can be considered to capture the highest response of the time series to the corresponding kernel, thus indicating the region with the greatest similarity. Therefore, for k kernels, 2k features are generated. In the example using the MiniROCKET model, during the transformation phase, the maximum value feature is not relied upon, and only the PPV value is computed.

[0125] In some examples, a machine learning framework employing stochastic gradient descent (such as PyTorch) can be used to train the feature extractor 302, and in others, batch gradient descent or mini-batch gradient descent can be used. In the examples, the following hyperparameter configurations can be used to perform training:

[0126] ●Loss Function: CrossEntropyLoss

[0127] ●Optimizer: Adam Algorithm

[0128] ●Scheduler: ReduceLROnPlateau

[0129] ● Batch size: 256

[0130] ●Learning rate: 0.0001

[0131] ●Number of tolerances: 5

[0132] Hyperparameter optimization can be performed using a Tree-structured Parzen Estimator (TPE) (a well-established method for automatic hyperparameter tuning) to further optimize the training process. The execution of a TPE allows hyperparameter values ​​to be updated beyond their initial settings. For example, while the tolerance count might initially be set to 5, it could be updated to 3 after a TPE.

[0133] In the example, a cache-based feature extraction method can be used to allow Mini-ROCKET to perform better on a training dataset of size 114. Specifically, the training data 114 can be divided into several smaller blocks, and then Mini-ROCKET can iterate over these blocks to extract features that will be used subsequently during training.

[0134] In this example, 9996 features corresponding to 9996 kernels are extracted from feature extractor 302. However, it should be understood that the number of kernels and therefore the number of features extracted are hyperparameters that may differ in other examples.

[0135] exist Figure 8 In the example, the features extracted by feature extractor 302 are then fed into linear classifier 304, which in this example includes nodes X1, X2, X3…X n The dense layer is followed by a softmax activation function. A linear classifier can be trained using, for example, ridge regression or logistic regression to capture the extracted features. The output of the softmax activation function is a label or classification of 306: "0": too high; "1": correct; or "2": too low.

[0136] It should be understood that the machine learning algorithm described above is only one example, and other architectures are possible. The benefits of the training data prepared at the previously described term S2 are not limited to use with the specific example of the machine learning algorithm described above.

[0137] Deploying machine learning algorithms

[0138] At item S4, deploy the machine learning algorithm that has already been trained at item S3. Figure 9 A flowchart detailing the steps taken when deploying and using a deployed, trained machine learning algorithm is shown. The flowchart includes items S3 and S5 to help illustrate the processes preceding and following item S4.

[0139] At item S3, as previously described, the training data prepared at item S2 is used to train the machine learning algorithm. However, more generally, any machine learning algorithm trained to classify each subset of bioelectrical signals into multiple categories (where one or more categories are for mislocated electrodes and at least one category is for correctly located electrodes) can be used to process the output of the machine learning algorithm described below. For example, the machine learning algorithm may have already been trained on a separate, different system and imported into the current location monitoring system 100, and does not necessarily have been trained according to item S3.

[0140] At item S400, the trained machine learning algorithm prepared at item S2 is deployed. Generally, this means the machine learning algorithm is operable to receive new data, rather than training data, including unseen data without known ground truth labels, and the machine learning algorithm will generate predictions for it. This data can be considered real-time data, as it typically comes from catheters used by patients. In some examples, once the training phase of item S2 is complete and the machine learning algorithm is deployed, any parameters, weights, or other variables of the algorithm optimized during training are fixed. In other examples, for example, even once deployed, further training and optimization steps can be performed to continuously improve or otherwise update the machine learning algorithm. Deployment can mean that the algorithm is available in a local computing unit (e.g., a desktop computer with direct access to the algorithm) or on a server (e.g., a cloud server or local server with remote access to the algorithm via a network). It should be understood that the exact manner in which the computation of the trained machine learning algorithm is processed is not important to the present invention and can vary between embodiments.

[0141] At item S401, catheter data, including bioelectrical signals detected from the patient, is received. Figure 1 and 2 In this example, these bioelectrical signals are detected from catheter 30 and arrive via catheter data input module 130. The bioelectrical signals correspond to a time period, which can vary between embodiments. For example, a bioelectrical signal may correspond to a single heartbeat detected by the catheter. In other examples, the bioelectrical signal may correspond to several consecutive heartbeats detected by the catheter.

[0142] Similar to the processing of historical clinical data described in item S202, the catheter data input module 130 can perform a data preparation process on the bioelectrical signals. This data preparation process may include some form of data normalization and / or data cleaning, for example, to ensure the degree of consistency of the bioelectrical signal format under consideration. In some examples, for instance, the processing may prepare the data in a manner best suited for use with the deployed machine learning algorithm, such as preparing the data in a specific file format, or removing specific components of data that are not needed for classification purposes and would otherwise degrade the performance of the deployed machine learning algorithm. For example, the machine learning algorithm may have been trained based on a single heartbeat, while the received bioelectrical signals correspond to several consecutive heartbeats. For example, the received bioelectrical signals may be divided into single heartbeats before being input into the deployed machine learning model.

[0143] At item S403, the set of bioelectrical signals is divided into subsets of bioelectrical signals, where each subset includes signals associated with sequentially placed electrodes. This process can be substantially similar to the process described for item S204.

[0144] At step S405, a subset of the bioelectrical signals is input into a deployed training machine learning algorithm, and at step S407, the deployed training machine learning algorithm generates a prediction label for each subset of bioelectrical signals. This prediction label is a form of location information, as the machine learning algorithm, given a subset of bioelectrical signals, is prepared to predict whether the electrode associated with each subset is correctly located. As previously mentioned, the location information can be a binary classification, such as "correctly located" or "incorrectly located," or it can be more granular, such as "too high," "correctly located," or "incorrectly located." Depending on the configuration of the machine learning algorithm, alternative predictions can be generated, such as indicating the number of electrodes between pairs associated with a subset and correctly located pairs, or an estimated distance between pairs associated with a subset and correctly located pairs.

[0145] The location information output by the machine learning algorithm and received at item S407 forms multiple input classifications that will be used at item S5, which will be based on... Figure 10 Further description.

[0146] Process the output of machine learning algorithms to determine the location state.

[0147] At item S5, location information output by a machine learning algorithm is used to determine the positioning status. The machine learning algorithm generates predictive classifications based on bioelectrical signals detected by the catheter, and these predictive classifications represent the location information. Figure 10 A flowchart detailing the steps taken to determine a location state based on location information output by a machine learning algorithm is shown. The flowchart includes items S4 and S6 to illustrate the processes preceding and following item S5.

[0148] At item S4, as described above, the deployed machine learning algorithm receives bioelectrical signal data detected by catheters within the patient's body and generates a predictive classification for a subset of the bioelectrical data. Therefore, the output of item S4 is multiple classifications, which can then be used as inputs to items S501 through S505, and are referred to hereinafter as input classifications.

[0149] At item S501, multiple inputs are categorized and input into the forming process. Figure 2 The pattern recognition function is part of the mid-position determination process 122.

[0150] At item S503, the pattern recognition function determines the catheter's positioning status, which includes an output classification that categorizes the catheter's position. The following will be based on... Figure 11 and Figure 12Examples of pattern recognition functions are described in more detail below. Generally, a pattern recognition function aggregates instance-level classifications (e.g., in the form of input classifications) into fewer, higher-level classifications (e.g., a single classification of the overall location of a duct), where each instance-level classification corresponds to a subset of bioelectrical signals classified by a machine learning algorithm. Generally, the classifications produced by the pattern recognition process may match or differ from the latent categories generated by the machine learning algorithm.

[0151] At item S505, the location status and constant output classification are received from the pattern recognition function, and at item S6, the location status is output to the user.

[0152] Example of pattern recognition function

[0153] Figure 11 A first example of pattern recognition functionality is illustrated, in which an output classification 126 is generated based on an input classification 121 generated by a machine learning algorithm 120 using a lookup table 123a.

[0154] As shown in the figure, in the example, the input classification 121 generated by the machine learning algorithm 120 may include the electrode e corresponding to the classification. n A list of associated categories “C”. In this example, input category 121 has classes including “0”, “1”, and “2”, where “0” indicates the associated electrode is too high, “1” indicates the associated electrode is correctly positioned, and “2” indicates the associated electrode is too low. In this example, electrodes are arranged in pairs for clarity, but as mentioned earlier, other groupings are possible. In some examples, input category 121 includes a list of categories but does not explicitly store the corresponding associated electrodes, and the corresponding associated electrodes may be implicitly encoded in the category location. For example, it may be known that the first entry in input category 121 always corresponds to electrodes e1 and e2; in this case, it may not be necessary to explicitly record the electrode identifiers.

[0155] In this example, lookup table 123a lists every permutation of the possible input categories, hereinafter referred to as candidate combinations of input categories. For example, in this example, there are 7 categories, each with 3 subcategories, resulting in 2187 distinct combinations, each represented by a corresponding candidate combination in lookup table 123a. Each candidate combination is stored in association with a corresponding candidate output category, stored in [the relevant database]. Figure 11 In the example, the "Output" column. For example, candidate output categories can be defined by expert clinicians or determined through, for example, analysis of historical clinical data. In other examples, algorithmic methods will be used to determine candidate output categories.

[0156] Given any input category 121, the pattern recognition function searches lookup table 123a and finds matching candidate combinations and associated candidate output categories. The candidate output categories can then be set as the output category 126 of the pattern recognition function.

[0157] In this example, lookup table 123a includes every permutation of the input categories, but in other examples it may include a subset, such as the majority of every permutation of the input categories, but still excludes combinations that can be pre-rejected because they do not produce a meaningful output category. Instead, the pattern recognition function may, for example, display “incorrect” categories and produce a location-related category only when the input category matches a candidate combination in lookup table 123a. For example, this could help diagnose catheter malfunctions in contrast to incorrect localization. For example, it should be understood that the lookup table may include enough permutations of the input categories for clinicians to use, and edge case combinations (e.g., combinations that do not correspond to a reasonable description of catheter location even considering the inherent errors in machine learning algorithm classification) may be removed from the lookup table to save storage space.

[0158] In some examples, additionally, lookup table 123a may include a confidence value associated with each candidate combination, stored in Figure 11 In the example, the confidence value of the matching candidate combination can be output as an output confidence value 127 in association with the output classification 126. For example, the confidence value represents the likelihood that the output classification is correct and can be considered as the probability corresponding to the correct output classification. Therefore, clinicians can understand the likelihood that the catheter will be positioned as indicated by the output classification. Figure 11 In the examples, the confidence values ​​range from 0 to 100, but it should be understood that any suitable representation is possible. Confidence values ​​can be predefined manually by clinicians, for example, when evaluating candidate combinations. In other examples, confidence values ​​can be determined algorithmically, for example, by penalizing combinations that do not follow the expected pattern.

[0159] The precise contents of lookup table 123a can vary depending on factors such as the patient's height, catheter design, and more generally, the mapping from the classification of the constituent electrode pairs to the overall classification of catheter location. For example, if a first catheter design is being used in a child, a first lookup table including suitable candidate combinations and candidate output classifications can be used, while for a second catheter design being used in an adult, a second lookup table including more suitable candidate combinations compared to the first lookup table can be used. For example, the catheter design can specify the total number of electrodes and their relative spacing, while the patient's height or age can determine the correct catheter composition based on, for example, electrode signals.

[0160] Figure 12 The illustration schematically depicts a pattern recognition function according to a second example, in which the pattern recognition function includes a predetermined list 123b of reference candidate combinations for classification, the predetermined list being similar to... Figure 11 The lookup table is used, but the variance is determined using an algorithmic approach based on a pre-defined list of reference candidate combinations.

[0161] The pattern recognition function includes a predetermined list 123b of reference candidate combinations. Each reference candidate combination is a set of reference candidate classifications, where each reference candidate classification corresponds to a subset of electrodes of the catheter, which is analogous to the set of input classifications 121. Therefore, the reference candidate combinations can represent expected combinations of input classifications resulting from the use of the catheter. Thus, for example, a reference candidate combination can represent a combination of input classifications resulting from the correct placement of the catheter and input classifications resulting from the catheter being positioned too low or too high relative to the septum. Therefore, for example, each reference candidate combination includes candidate output classifications based on whether it represents that the catheter is correctly positioned, positioned too low, or positioned too high.

[0162] The pattern recognition function compares the input classification set of 121 with each reference candidate combination and calculates the difference between the input classification and each reference candidate combination. This process can be viewed as calculating the similarity between the input classification and each reference candidate combination separately.

[0163] In the example shown, this is performed by considering the difference between the reference candidate categories stored in the "Ref-1" row and the corresponding input category 121 stored in the "Input" row to calculate the channel-wise difference. The channel-wise difference between categories is stored in the "Var" row, and in this example, the magnitude of the difference is taken such that it is stored as a positive value. This channel-wise difference means that the difference calculation is performed for each input category and its corresponding reference candidate category. That is, the input categories of electrodes e1-e2 are compared with their reference candidate categories, and the difference between e1 and e2 is calculated; the input categories of electrodes e2-e3 are compared with their reference categories, and the difference between e2 and e3 is calculated, and so on. This applies to all reference categories for a given combination of reference candidate categories. Then, in this example, the channel-wise differences are aggregated by summing the difference values ​​and stored in the "SumVar" column to represent the difference of the input categories as a whole based on the combination of reference candidate categories. In other examples, for instance, weighted aggregation can be used, where some differences are weighted more heavily than others in the aggregation. For the second reference candidate combination "Ref-2", the third reference candidate combination ( Figure 12 (not shown in the image) Repeat this process.

[0164] Then, the output classification of the pattern recognition function is set as a candidate output classification, which corresponds to the reference candidate combination with the lowest difference from the input classification. Figure 12 In the example, the difference in input classification is 4 according to the first reference candidate combination "Ref-1"; and 1 according to "Ref-2". Therefore, setting the output classification of the pattern recognition function to the candidate output classification of "Ref-2" (i.e., 1) corresponds to the correct localization of the duct.

[0165] As described above, by calculating and aggregating the channel-wise differences in the classification, the differences in the input classification can be calculated based on the reference candidate combinations. Other methods are also possible. For example, initially only a subset of channel-wise differences can be aggregated or even calculated. In the example above, all electrode pairs contribute equally to the final aggregated difference, but in some examples, when determining the overall location of the catheter, the farthest electrode ( Figure 1 The positions of e7-e9 in the example can be considered as being closer to the nearest electrode ( Figure 1In some examples, e1-e3 are more important, for example, to reduce the risk of the catheter tip coming into contact with the duodenum. Therefore, the method for calculating the differences can be to weight the farthest electrode rather than the nearest electrode. In other examples, the weight of the central electrode, located between the nearest and farthest electrodes, may be less than that of the nearest and farthest electrodes, because the inventors observed that classifying the electrode closest to the septum (e.g., which is typically the central electrode) can be more difficult.

[0166] In the example above, the difference is a simple subtraction between two categories. However, in other examples, more complex functions could be used, such as non-linear functions that impose a stronger penalty on larger differences compared to smaller differences. The method used can depend on the categories used; in the example above, there are three categories: “0”, “1”, and “2”. However, in other examples, more categories could be used, or continuous numbers such as distance metrics could be used, thus allowing for different difference calculation methods.

[0167] Similar to Figure 11 An exemplary pattern recognition function can output a confidence value, which, in some examples, can be based on a calculated difference. For example, the confidence value can be calculated via an equation (e.g., Confidence = 100 - (SumVar × SF), where SumVar is stored...). Figure 11 The calculated difference in the “SumVar” column of the predefined list 123b, SF is a scaling factor that determines the effect of SumVar on the confidence value. For example, SF can be 10; if SumVar is always positive and does not exceed 10, the confidence value will be a value between 0 and 100. Those skilled in the art should understand that other measures of confidence are possible and do not necessarily need to be positive values ​​between 0 and 100. In other examples, an alternative function may be used, such as Confidence = 100 - SumVar 2 Its main function is to impose a disproportionate confidence value penalty on large aggregation differences.

[0168] In an example of a pattern recognition function, the function can be configured to output several categories, each with a corresponding associated confidence value. For example, the pattern recognition function can output a first output category associated with a first output confidence value, and a second output category associated with a second output confidence value. For instance, the output of the pattern recognition function could be "1: 90%; 0: 10%", indicating that the probability of output category "1" is 90%, and the probability of output category "0" is 10%. This approach can be easily extended to three or more output categories, each with an accompanying confidence value.

[0169] For example, in the case of a lookup table, the pattern recognition function can output an output classification associated with candidate combinations that match the input classification. However, it can further consider candidate combinations within individual channel-by-channel variations and whether the output classification changes as a result. For instance, the pattern recognition function can examine whether, when channels e4-e5 change from classification 1 to 0, the new matching candidate combinations will have a different candidate output classification than the current output's candidate output classification. This can represent a scenario where the catheter is located at the boundary between two output classifications, and both can be output with corresponding confidence levels. Therefore, clinicians can obtain a more accurate or subtle impression of catheter location.

[0170] In the example, the pattern recognition function learns from previous output classifications, which can be considered a storage capacity. The pattern recognition function can be constrained or otherwise biased towards output classifications based on predetermined relationships between them. For example, after a "too low" classification, the pattern recognition function might be constrained or biased towards outputting "too low" or "correctly positioned" in subsequent time periods, rather than "too high." This prevents unstable jumps from "too low" to "too high," which could be unpredictable but likely due to misclassification by the machine learning algorithm. Therefore, for example, the output classifications are smoother and easier for clinicians to follow. The degree of constraint could be related to the rate at which the pattern recognition function generates new output classifications and the rate at which catheter position is expected to change.

[0171] exist Figure 11 and Figure 12 In the example, the pattern recognition function outputs one of the categories "0", "1", and "2". This corresponds to an example of a category output by the machine learning algorithm trained at item S3. In other examples, the machine learning algorithm may output more fine-grained categories, such as "0", "1", "2", "3", "4", and "5", corresponding to, for example, too low, slightly too low, correctly positioned, slightly too high, and too high. Alternatively, for example, the machine learning algorithm may output the predicted distance to the correct location (in millimeters), in which case the input category could be, for example, a number between 0 and 50. However, the pattern recognition function can process these categories into an output of, for example, one of "0", "1", and "2". In this way, the pattern recognition function reduces the number of instance-level categories when generating the output category. In other examples, the situation may be reversed: the pattern recognition function may receive an input category from a machine learning algorithm with categories "0", "1", and "2" and generate a more fine-grained output category based on, for example, "0", "1", "2", "3", "4", and "5".

[0172] In the example, the pattern recognition function is configured to identify anomalous behavior. For example, if a particular subset of bioelectrical signals / electrode pairs and therefore the corresponding input classification is generally different from the rest of the set, that corresponding input classification can be identified and excluded from the calculation of the output classification. For example, an input classification that differs from the nearest matching reference candidate combination is marked as an outlier. For example, channels e2-e3 can be marked as outlier channels because their input classification differs from the nearest reference candidate combination. For example, if this occurs consecutively a predetermined threshold number of times, the channel can be marked as faulty and ignored when determining the output classification in the future. In other examples, the position system 100, more generally, or in some examples specifically as the catheter data input module 130, is configured to detect noisy signals from the catheter before the signal is passed to a deployed machine learning algorithm, where the noise level of the signal is expected to cause misclassification by, for example, the machine learning algorithm. For example, signal processing or catheter recalibration can be performed in such scenarios to smooth the signal and thus reduce noise.

[0173] For example, item S5 may include a step that allows for time averaging, which can make the output classification more stable and less prone to short-term switching between classifications.

[0174] For example, the multiple input classifications received from the machine learning model at item S501 may correspond to bioelectrical signals from a single heartbeat detected by a catheter. The catheter data input module 130 can collect multiple bioelectrical signals from different corresponding times, thus corresponding to multiple heartbeats. These multiple bioelectrical signals from different corresponding times can be provided to a machine learning algorithm to generate a set of multiple input classifications, where each set of multiple input classifications corresponds to a bioelectrical signal from a corresponding time period. The set of multiple input classifications can then be input to a pattern recognition function. For example, the pattern recognition function can compute a potential output classification for each of the multiple input classifications and determine the overall output classification corresponding to the majority of these potential output classifications. Thus, the pattern recognition function is configured to process input classifications corresponding to bioelectrical signals corresponding to different corresponding time periods.

[0175] This method can also be used when multiple heartbeats are captured from the bioelectrical signals received from catheter 30, instead of just a single heartbeat as described above. For example, a first set of bioelectrical signals received from the catheter can capture the first five heartbeats, and a machine learning algorithm can generate a first set of input classifications based on these first five heartbeats. A second set of bioelectrical signals can be received from the catheter capturing the subsequent five heartbeats and fed to a machine learning algorithm, which generates a second set of input classifications based on these subsequent five heartbeats. A pattern recognition function can combine the second set of input classifications with the first set of input classifications to determine the output classification, as described above. In this example, the first set of bioelectrical signals capturing the first five heartbeats corresponds to a first time period, and the second set of bioelectrical signals capturing the subsequent five heartbeats corresponds to a second time period, which follows the first time period. In other examples, these time periods can partially overlap to allow for the calculation of a moving average output classification.

[0176] Similarly, confidence values ​​can be based on the degree of variation between corresponding time periods. For example, if there is a large variation between the input classification in the first time period and the second (subsequent) time period, this may decrease the confidence value in the output classification. Smaller variations may increase the confidence value in the output classification because, for example, it may indicate stable behavior or location.

[0177] In other examples, the input classification undergoes an averaging process before being passed to the pattern recognition function. For example, a direct numerical average can be calculated, or the majority classification for each channel can be calculated.

[0178] Monitoring mode

[0179] In some examples, the position monitoring system 100 can disable position monitoring via the position determination process 122 while the catheter is being inserted by a clinician, because it can be assumed that the catheter will be mispositioned for most of the time during the process. For example, the position determination process 122 can be automatically resumed when a trigger signal is detected from an electrode indicating proximity to the septum. The trigger signal can be a bioelectrical signal, which a machine learning algorithm determines originates from, for example, a correctly positioned electrode pair with high confidence. For example, the position determination process 122 can be manually resumed when the clinician determines that the catheter is close enough to the septum to initiate the position determination process 122.

[0180] In some examples, the position monitoring system 100 can enter proximity mode and steady-state mode, influencing the position determination process 122. Proximity mode can shorten the previously described time averaging period, allowing the user to receive more frequent position updates. For example, more frequent position updates may be useful when the catheter is initially moved into place, to avoid, for example, over-insertion of the catheter and collisions with the patient's body. Once the catheter is determined to be correctly positioned, the position monitoring system can enter steady-state mode, which can increase the time averaging, for example, to reduce short-term categorical variations. In some examples, steady-state mode includes a clinician-absent mode. For example, this can be used when the clinician does not actively reposition the catheter, thus assuming that any change in the catheter's positional state is due to some other reason, such as patient movement or accidental catheter repositioning. Therefore, the data display output module 124 can act differently in clinician-absent mode compared to a mode where clinician-initiated catheter repositioning is expected. For example, given that catheter movement is not expected, the position monitoring system 100 may be more sensitive to catheter movement or movement of the catheter above a threshold size, for example, in clinician-absent mode.

[0181] Output location status to user

[0182] At item S6, the positioning status of conduit 30 is output to the user. Figure 13 An example of a graphical user interface 400 is schematically shown, which is displayed on display unit 200 and is typically used to display information related to catheter location to clinicians. Display unit 200 is controlled by display data output module 124, which may include a graphics processing unit and typically communicates with processor 140 to determine the information displayed by display unit 200.

[0183] In this example, the graphical user interface 400 includes a bioelectrical signal section 402, a channel-by-channel position section 404 displaying position status information, and an overall catheter position section 406. The bioelectrical signal section 402 displays bioelectrical signal data received from the catheter 30, and this data can be interpreted by a clinician independently of the position status information. For example, the channel-by-channel position section 404 displays classifications generated by a machine learning model, indicating, for example, which channels and associated electrodes are correctly positioned, positioned too low, or positioned too high, channel-by-channel. The overall catheter position section 406 displays output classifications from a pattern recognition function, such as indicating to the clinician whether the catheter as a whole is correctly positioned, too low, or too high. The information displayed by the channel-by-channel position section 404 and the overall catheter position section 406 can be symbolically represented, for example. In other examples, text or numbers corresponding to the classifications can be displayed. In this example, additionally or alternatively, based on the output classifications, the display can incorporate visual effects such as color, brightness, flickering, or luminescence to improve clinician visibility. For example, the bioelectrical signal section 402 can be enhanced with these visual effects.

[0184] In some embodiments, auditory effects may be used in conjunction with or replace visual effects. For example, an alarm may be triggered when the location is incorrect. This alarm may further indicate how the location is incorrect; that is, for example, a first alarm sound may indicate that the sound is too high, while a second alarm sound may indicate that the sound is too low. For example, the tone or volume may vary depending on the location state. For example, the auditory effects may be managed by the display data output module 124.

[0185] As described above, the display data output module may further include a component that notifies the clinician based on location status updates depending on the clinician's presence. For example, when in a steady-state, clinician-absent mode, the display data output module may notify the clinician via electronic communication devices or add sound or visual effects, enabling notification to the clinician without actively monitoring the patient.

[0186] For example, display unit 200 can be an LCD, LED, OLED, CRT, or any other type of display panel capable of displaying visual content, and can be part of a television or computer monitor. In the example, display unit 200 can be, for example, an augmented reality, mixed reality, or virtual reality headset worn by a clinician.

[0187] Other embodiments

[0188] The above embodiments should be understood as illustrative examples of the present invention. It should be understood that any feature described with respect to any embodiment can be used alone, in combination with other described features, and in combination with one or more features of any other embodiment or any combination of any other embodiment. Furthermore, equivalents and modifications not described above may be employed without departing from the scope of the invention as defined by the appended claims.

Claims

1. A method for determining training data to be used for training a machine learning algorithm to classify a position of a catheter relative to a septum of a patient, the method comprising the steps of: receiving a set of bioelectric signals detected by a catheter carrying a plurality of electrodes at respective positions along a length of the catheter, such that the electrodes are located at respective different distances from a septum of a patient, the plurality of electrodes being divided into a plurality of electrode pairs, each signal being detected by one of the plurality of electrode pairs, each signal including an electrocardiogram (ECG) component; from the set of bioelectric signals, identifying one or more first bioelectric signals, each first bioelectric signal being detected by an electrode pair on the catheter determined to be correctly positioned relative to the septum; dividing the set of bioelectric signals into at least two subsets of bioelectric signals, each subset of bioelectric signals including one or more bioelectric signals and corresponding to a respective group of electrodes associated with the detection of the one or more bioelectric signals in the subset of bioelectric signals, wherein the dividing is performed such that each group of electrodes is a sequence of electrodes placed consecutively along the length of the catheter; labeling each of the plurality of subsets, wherein the labeling includes: labeling a subset including at least one of the first bioelectric signals as a subset of signals detected from correctly positioned electrodes, and labeling a subset not including any of the first bioelectric signals as a subset of signals detected from incorrectly positioned electrodes; and including the subsets of bioelectric signals and their respective labels in the training data.

2. The method of claim 1, wherein the step of labeling a subset of bioelectric signals not including the first bioelectric signals as a subset of bioelectric signals detected from incorrectly positioned electrodes includes: determining whether the electrodes associated with the detection of the one or more bioelectric signals in the subset are positioned above or below the septum, wherein upon determining that the electrodes are positioned above the septum, the subset of bioelectric signals is labeled as a subset of bioelectric signals detected from electrodes positioned above the septum, and upon determining that the electrodes are positioned below the septum, the subset of bioelectric signals is labeled as a subset of bioelectric signals detected from electrodes positioned below the septum.

3. The method of any one of claims 1 or 2, further comprising: enhancing each bioelectric signal in the subset, wherein enhancing a bioelectric signal includes at least one of: stretching the bioelectric signal over time; compressing the bioelectric signal over time; or changing an amplitude of the bioelectric signal.

4. The method of claim 1, wherein at least one bioelectric signal includes an electromyogram (EMG) component, and the method further comprises: applying a filtering algorithm to each bioelectric signal in the set of bioelectric signals, wherein the filtering algorithm is configured to at least reduce a respective electromyogram (EMG) component from the respective bioelectric signal.

5. The method of claim 4, wherein applying the filtering algorithm to a bioelectric signal includes: identifying a plurality of sub-portions of the bioelectrical signal, each sub-portion comprising data detected during a heartbeat of the patient; and computing an average bioelectrical signal from the plurality of sub-portions of the bioelectrical signal.

6. The method of claim 1, wherein each bioelectrical signal from the set of bioelectrical signals comprises data detected during a plurality of heartbeats of the patient, wherein the method further comprises: identifying, in each bioelectrical signal from the set of bioelectrical signals, data detected within an intermediate time period between two consecutive heartbeats of the plurality of heartbeats; and removing the identified data from the bioelectrical signal.

7. The method of claim 1, wherein the step of identifying the one or more first bioelectrical signals from the set of bioelectrical signals comprises: detecting, in at least one bioelectrical signal from the set of bioelectrical signals, the presence and magnitude of an electromyography (EMG) component; and, selecting at least one bioelectrical signal as the one or more first bioelectrical signals based on the magnitude of the respective EMG component.

8. The method of claim 1, wherein the step of marking does not include a subset of any of the first bioelectrical signals further comprises: using the distance between the electrodes associated with the subset and the electrodes associated with the properly positioned subset of bioelectrical signals to flag subsets that do not include any of the first bioelectrical signals.

9. The method of claim 1, wherein, performing the partitioning of the set of bioelectrical signals such that at least two subsets of bioelectrical signals partially overlap, such that an electrode associated with detecting one or more bioelectrical signals in a first subset is also associated with detecting one or more bioelectrical signals in a second subset.

10. The method of claim 2, wherein, partitioning the set of bioelectrical signals into the at least two subsets of bioelectrical signals such that the number of subsets determined to be associated with properly positioned electrodes and flagged as subsets of signals detected from properly positioned electrodes is in a predetermined proportion to the number of subsets determined to be associated with electrodes positioned above or below the diaphragm and flagged as subsets of signals detected from electrodes positioned above the diaphragm or below the diaphragm.

11. The method of claim 1, wherein when a first bioelectrical signal is identified as being associated with an electrode that is the farthest or closest electrode relative to the length of the catheter among the plurality of electrodes, the set of bioelectrical signals is not included in the training data.

12. The method of claim 1, wherein each electrode pair is a pair of adjacent electrodes.

13. The method of claim 12, wherein the one or more first bioelectrical signals comprise a single first bioelectrical signal detected by a pair of electrodes on the catheter, the pair of electrodes being the pair of electrodes among the plurality of pairs of electrodes positioned closest to the diaphragm.

14. One or more non-transitory computer-readable media storing instructions executable by one or more processors, wherein the instructions, when executed, cause the one or more processors to perform operations comprising: receiving a set of bioelectric signals detected by a catheter, the catheter carrying a plurality of electrodes at respective locations along a length of the catheter, thereby positioning the electrodes at respective different distances from a septum of a patient, the plurality of electrodes being divided into a plurality of electrode pairs, each signal being detected by one of the plurality of electrode pairs, each signal including an electrocardiogram (ECG) component; from the set of bioelectric signals, identifying one or more first bioelectric signals detected by a pair of electrodes on the catheter, each first bioelectric signal being determined to be correctly positioned relative to the septum; dividing the set of bioelectric signals into at least two subsets of bioelectric signals, each subset of bioelectric signals including one or more bioelectric signals and corresponding to a respective group of electrodes associated with the detection of the one or more bioelectric signals in the subset of bioelectric signals, wherein the dividing is performed such that each group of electrodes is a sequence of electrodes placed contiguously along the length of the catheter; labeling each of the plurality of subsets, wherein the labeling includes: labeling a subset including at least one of the first bioelectric signals as a subset of signals detected from correctly positioned electrodes, and labeling a subset not including any of the first bioelectric signals as a subset of signals detected from incorrectly positioned electrodes; and including the subsets and their respective labels in training data for training a machine learning algorithm to classify a position of a catheter relative to a septum.

15. A system, the system comprising: one or more processors; and one or more non-transitory computer-readable media storing first computer-executable instructions that, when executed by the one or more processors, cause the system to perform acts comprising: receiving a set of bioelectric signals detected by a catheter, the catheter carrying a plurality of electrodes at respective locations along a length of the catheter, thereby positioning the electrodes at respective different distances from a septum of a patient, the plurality of electrodes being divided into a plurality of electrode pairs, each signal being detected by one of the plurality of electrode pairs, each signal including an electrocardiogram (ECG) component; from the set of bioelectric signals, identifying one or more first bioelectric signals, each first bioelectric signal being detected by a pair of electrodes on the catheter determined to be correctly positioned relative to the septum; dividing the set of bioelectric signals into at least two subsets of bioelectric signals, each subset of bioelectric signals including one or more bioelectric signals and corresponding to a respective group of electrodes associated with the detection of the one or more bioelectric signals in the subset of bioelectric signals, wherein the dividing is performed such that each group of electrodes is a sequence of electrodes placed contiguously along the length of the catheter; labeling each of the plurality of subsets, wherein the labeling includes: labeling a subset including at least one of the first bioelectric signals as a subset of signals detected from correctly positioned electrodes, and labeling a subset not including any of the first bioelectric signals as a subset of signals detected from incorrectly positioned electrodes; labeling a subset of the first bioelectrical signals that does not include any of the first bioelectrical signals as a subset of signals detected from a mispositioned electrode; and including the subset and its corresponding label in training data for training a machine learning algorithm to classify a position of the catheter relative to the septum.

Images