Method and system for SSEP with monitorable baseline waveform determination

Abstract

A method implemented by a somatosensory evoked potential (SSEP) system is disclosed. The method includes obtaining at least one SSEP recording from a subject and determining, based on the at least one SSEP recording, whether a baseline potential is monitorable. The method further includes obtaining an ongoing SSEP recording from the subject, comparing the ongoing SSEP potential with the monitorable baseline potential, and issuing an alert if the ongoing SSEP potential deviates from the monitorable baseline potential according to defined criteria. This may allow a medical professional to determine whether to take corrective action, such as repositioning the subject, to prevent or mitigate iatrogenic damage to the subject's nervous system. The SSEP system can be substantially automated, thereby relying less on the discretion of the medical professional. An SSEP system configured to implement the method summarized above is also disclosed.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS Priority under 35 U.S.C. § 119 is claimed to U.S. Provisional Patent Application No. 63 / 393,230, filed July 28, 2022, entitled "METHOD AND SYSTEM FOR SSEP (SOMATOSENSORY EVOKED POTENTIALS) WITH MONITORABLE BASELINE WAVEFORM DETERMINATION" (the "'230 Provisional Application"). The entire disclosure of the '230 Provisional Application is incorporated herein by reference.

[0002] The present disclosure relates to an evoked potential (EP) monitoring system that avoids or reduces iatrogenic nerve injury during surgery. [Background technology]

[0003] Short-latency somatosensory evoked potentials (SSEPs) are small sensory potentials (also called responses or waveforms) that are time-related to (typically electrical) stimulation of peripheral or cranial nerves. Because they are small, they are typically identified by averaging multiple fixed-period recordings time-related to multiple stimuli to improve signal-to-noise ratios. These potentials are often recorded in the head or neck region. Monitoring subjects using SSEPs during various surgical procedures has been shown to provide early identification of iatrogenic or other damage to nervous system structures. Early identification of these physiological changes in nervous system function allows intervention to avoid or mitigate potential damage.

[0004] Such monitoring is typically performed by trained technicians under medical supervision using complex multi-channel amplifier and display equipment, but such personnel and equipment are limited in availability, require advance reservations, can be of variable quality, and are costly.

[0005] Intraoperative SSEP monitoring requires the acquisition of a baseline recording and the determination of baseline potentials. The baseline potential is the standard against which subsequently acquired SSEP potentials can be compared throughout the monitoring period. Baseline recordings are typically acquired after induction of anesthesia and before the start of the surgical intervention.

[0006] Many existing SSEP systems rely primarily on medical personnel (e.g., technicians / neurologists) to interpret potentials to determine whether baseline potentials are present and monitorable within the baseline recording and, through comparison, when significant changes in ongoing SSEP potentials (recorded after baseline establishment) have occurred. Some potentials may be initially identifiable during baseline recording evaluation, but may be too small or insufficient to accurately indicate changes due to developing injury. This process is complex, subject to delays, and subject to medical personnel bias, varying expertise, and concentration. Inappropriate classification of baseline potentials as present, absent, or present but insufficient to monitor can result in over- or under-identification of subsequent changes in neural function, potentially failing to alert the user to possible neural injury or suggesting possible injury when none exists. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] International Publication No. 2013 / 166157 [Patent Document 2] U.S. Patent Application Publication No. 2011 / 230785 Summary of the Invention [Problem to be solved by the invention]

[0008] It is an object of the present disclosure to improve upon conventional approaches to address or mitigate some or all of the above-mentioned problems. [Means for solving the problem]

[0009] A method is disclosed that is performed by a somatosensory evoked potential (SSEP) system. The method includes obtaining at least one SSEP recording from a subject and determining a monitorable baseline potential based on the at least one SSEP recording. The method further includes obtaining an ongoing SSEP recording from the subject, comparing the ongoing SSEP potential to the baseline potential, determining a deviation of the ongoing SSEP potential from the baseline potential according to defined criteria, and issuing an alert if a significant change is identified.

[0010] This may prompt the medical professional to take corrective action(s), such as, but not limited to, repositioning the subject or reducing the surgical retractor, with the goal of preventing or reducing damage to the subject's nervous system. While the ultimate decision as to whether any corrective action should be taken rests with the medical professional, the SSEP system can reliably and more accurately inform or prompt such decisions or actions. This improvement is made possible in part by innovations in how the SSEP system identifies whether there are baseline potentials that can be monitored.

[0011] In some embodiments, obtaining at least one SSEP baseline recording includes obtaining a two-component SSEP recording based on two corresponding individual SSEP data sets, and identifying a baseline potential within both recordings and a grand ensemble average of the individual SSEP data sets. Based on characteristics of the two recordings and a record generated from the grand ensemble average, the baseline potential is deemed monitorable or non-monitoring. This allows the SSEP system to reliably identify whether a monitorable baseline potential exists.

[0012] In some forms, calculating or identifying a major peak of the potential response for a baseline SSEP recording includes identifying candidate peaks in the upright-displayed recording and in the inverted-displayed baseline SSEP recording, and identifying which candidate peak is most prominent based on how distinct the candidate peak is by its inherent height and its position relative to other candidate peaks, and the most prominent candidate peak is used to compare the baseline SSEP potential with the ongoing SSEP potential.

[0013] In some forms, identifying a major vertex of the potential for each ongoing SSEP recording includes identifying a candidate vertex in the baseline SSEP recording that has the same polarity as the major vertex. In some forms, this means selecting a major vertex from the candidate vertices based on vertex characteristics and information from the baseline SSEP waveform and previous ongoing SSEP waveforms. In some forms, this may mean selecting a candidate vertex based on either the proximity of the candidate vertex to the major vertex of the identified potential in the previous ongoing recording, or the prominence of the candidate vertex relative to other vertices. This may allow the SSEP system to reliably compare and identify the ongoing SSEP potential with the baseline SSEP potential.

[0014] Also disclosed is a non-transitory computer readable medium having recorded thereon statements and instructions that, when executed by a processor of an SSEP system, configure the processor to perform the methods summarized above.

[0015] Additionally, a baseline optimization state is disclosed, which is an adaptive baseline setting.

[0016] Also disclosed is an SSEP system comprising a stimulating electrode configured to elicit a response from a peripheral or cranial nerve traversing a subject through its nervous system, and a recording electrode configured to sense electrical potentials as they traverse the nervous system, and a nerve injury detection device coupled to the stimulating and recording electrodes and configured to perform the methods summarized above.

[0017] Other aspects and features of the present disclosure will become apparent to those of ordinary skill in the art upon review of the following description of various embodiments of the present disclosure. [Brief explanation of the drawings]

[0018] Embodiments will be described with reference to the accompanying drawings, in which: [Figure 1] FIG. 1 is a block diagram of a SSEP (Somatosensory Evoked Potential) system for subject monitoring. [Figure 2] 1 is a flowchart of a method for monitoring a subject for risk of neurological injury. [Figure 3] 1 is a flowchart of a method for identifying the principal and reference peaks of potentials in a baseline SSEP recording. [Figure 4] 1 is a graph of a baseline SSEP recording displayed upright. [Figure 5] 1 is a graph of an inverted display of a baseline SSEP recording. [Figure 6] 10 is a flowchart of a method for determining whether a baseline recording contains a monitorable baseline potential. [Figure 7] 1 is a graph of a baseline SSEP recording. [Figure 8] 1 is a set of charts for baseline SSEP recordings including monitorable potentials. [Figure 9] 1 is a set of charts for baseline SSEP recordings including non-monitoring potentials. [Figure 10] A set of charts for baseline SSEP recordings containing monitorable potentials, where response quality improves with the collection of more single trials. [Figure 11] 1 is a flowchart of a method for identifying the main and reference peaks of the potential in an ongoing SSEP recording. [Figure 12] 1 is a graph of baseline and ongoing SSEP recordings from a case in which monitorable baseline potentials are shown in a "waterfall" diagram. [Figure 13]10 is a graph of baseline and ongoing SSEP recordings from a case where unmonitored baseline potentials and user override of baseline classification are shown in a "waterfall" diagram. [Figure 14] 1 is a graph of SSEP recordings for a case involving monitorable baseline potentials. [Figure 15] 1 is a graph of SSEP recordings for a case involving unmonitored baseline potentials. [Figure 16] 1 is a graph of SSEP recordings for a case involving unmonitored baseline potentials. DETAILED DESCRIPTION OF THE INVENTION

[0019] Initially, exemplary embodiments of one or more embodiments of the present disclosure are provided below, but it should be understood that the disclosed systems and / or methods may be implemented using any number of technologies. The present disclosure is in no way limited to the examples, drawings, and technologies illustrated below, including the exemplary designs and configurations shown and described herein, but may vary within the full scope of the appended claims and their equivalents.

[0020] Evoked potentials are electrical signals generated by the nervous system in response to sensory stimuli. In other words, evoked potentials are neurophysiological responses to electrical stimulation. Auditory, visual, and somatosensory stimuli are commonly used in clinical evoked potential studies. SSEPs consist of a series of waves reflecting the sequential activation of neural structures along the somatosensory pathway. While SSEPs can also be elicited by mechanical stimulation, clinical studies use electrical stimulation of peripheral nerves, which produces larger and more powerful responses. Selected nerves are typically stimulated with monophasic square pulses of 100 to 300 microseconds in duration. Stimuli are delivered using either constant-voltage or constant-current stimulators.

[0021] SSEPs are not typically visible in raw data recorded from surface electrodes. Signal averaging is used to extract SSEPs from other electrical signals acquired by recording electrodes. The waveform resulting from nerve stimulation is displayed as a plot of voltage versus time and characterized by measurements of the post-stimulus latency (in milliseconds (ms)) and amplitude (in microvolts (μV)) of specific peaks. The recording is a time series showing the electrical signal measured over time. Several characteristics of SSEPs can be measured, including peak latency, component amplitudes, and waveform morphology. Following common convention, deflections below the baseline are considered positive and deflections above the baseline are considered negative. Waveforms are commonly identified by a letter specifying the direction of the deflection, followed by a number representing the latency of the waveform.

[0022] Peak latency is typically consistent across subjects, whereas amplitude varies widely across subjects. Interpretation of extraoperative diagnostic SSEP studies is primarily based on peak latency and derived measures, such as interpeak interval and laterality. Component amplitudes are more consistent during repeated SSEP recordings in the same subject and may change faster than latency or even be absent if somatosensory pathways are damaged during surgery. Therefore, both peak latency and component amplitudes must be measured and tracked during intraoperative monitoring.

[0023] 1 shows a block diagram of a SSEP (Somatosensory Evoked Potential) system 100 for monitoring a subject 101. The SSEP system 100 includes one or more recording electrodes 102, one or more stimulating electrodes 103, and a NIDD (Nerve Injury Detection Device) 104. In some forms, the SSEP system 100 further includes a platform integration unit 105, an alarm and display unit 106, and a platform 107. The SSEP system 100 may also include other components not shown. Note that the subject 101 is not part of the SSEP system 100.

[0024] In some forms, the subject 101 rests on a platform 107. Recording electrodes 102 are coupled to the subject 101, e.g., their head, neck, arms, legs, upper body, cervical nerve points, and / or torso. Stimulating electrodes 103 are also coupled to the subject 101, e.g., their arms and / or legs. The stimulating electrodes 103 are configured to generate SSEP responses that traverse the subject 101 through the nervous system and are detected by the recording electrodes 102. Iatrogenic damage in the nervous system of the subject 101 can be detected based on the SSEP responses detected by the recording electrodes 102. SSEP baseline recordings, ongoing SSEP recordings, and derived features or information can be displayed on the SSEP system's alert and display 106.

[0025] Determining whether a subject's 101 nervous system structures are at risk is technically challenging in part because SSEP potentials vary widely between subjects and SSEP recordings may contain electrical noise and / or artifacts. To assess whether SSEP potentials indicate a risk to a subject's 101 nervous system structures, for example, as a first step before performing surgery or other procedures on the subject 101, the ongoing SSEP potentials are measured, identified, and compared to baseline potentials that may be classified as monitorable or non-monitoring. Depending on whether the ongoing SSEP potentials match or deviate from the monitorable baseline potentials, an alert may be issued. The comparison of the ongoing SSEP potentials to the baseline SSEP potentials may be based on salient features such as evoked potential amplitude and / or latency. Further detailed examples are provided below.

[0026] Referring now to Figure 2, a flowchart of a method for monitoring risk of nerve injury for a subject 101 is shown. The method can be performed in an SSEP system, such as the SSEP system 100 shown in Figure 1. In certain embodiments, the method is performed by the NIDD 104 of the SSEP system 100. More generally, the method of Figure 2 can be performed in any suitable SSEP system.

[0027] In step 2-1, the SSEP system 100 acquires SSEP recordings. For example, as described above, the stimulating electrodes 103 can generate electrical stimuli that activate peripheral nerves of the subject 101, sending action potentials through the subject's 101 nervous system and across the subject's 101, which are detected by the recording electrodes 102. Responses recorded after individual stimuli or single-trial responses can be ensemble-averaged to improve the signal-to-noise ratio of the evoked potential response. In this manner, the SSEP system 100 acquires one or more SSEP recordings. Independent SSEP data sets are acquired in separate trials, with no single-trial overlap between two independent SSEP data sets.

[0028] In steps 2-2 and 2-3, the SSEP system 100 determines whether a monitorable baseline potential exists based on the SSEP recording obtained in step 2-1. Note that steps 2-1 through 2-3 may be repeated until a monitorable baseline potential is found. The SSEP system 100 can determine whether a monitorable baseline potential exists in various ways. In some embodiments, the SSEP system 100 uses a machine learning algorithm to classify each baseline SSEP recording as having a monitorable or non-monitoring potential. In some embodiments, the machine learning algorithm used to classify the baseline potential is a supervised machine learning algorithm, in which the features used by the algorithm are based on morphology and repeatability.

[0029] While the morphology of evoked potentials can vary between subjects, some common / typical morphologies exist. There is also a range in the expected latency between stimulation and the appearance of the evoked potential in the recording, depending on where the stimulus is applied to the subject 101 (e.g., stimulating the ankle results in a longer latency due to the longer distance traveled). If the deflection in the recording is simultaneously repeated in a subsequent independent data set to the electrical stimulus, there is a high probability that the response contains an actual SSEP potential, rather than artifact, noise, or other physiological or non-physiological sources. Actual SSEP potentials with identifiable dominant and basal peaks and sufficient amplitude above the noise floor will be monitorable. Accurate classification can be achieved by comparing morphological features alone and over time. A more detailed example of how the SSEP system 100 determines whether there is a monitorable baseline potential is provided below.

[0030] In some embodiments, the SSEP system 100 identifies the peaks of the SSEP potentials, as shown in step 2-2. The SSEP system 100 can then use these peak markings to evaluate morphology and repeatability to determine whether there are monitorable baseline potentials. The SSEP system 100 can identify the major and reference peak(s) of the SSEP potentials in various ways. In some embodiments, the SSEP system 100 uses algorithms to identify peaks based on maximum, minimum, derivative, or second derivative. In an exemplary embodiment, the SSEP system 100 identifies the major peaks using the prominence of candidate peaks within a stimulation / recording site-specific analysis period. The prominence of a candidate peak measures how distinctive a candidate peak / valley is by its inherent height and its location relative to other candidate peaks / valleys. An isolated low peak in a signal may be more prominent than a higher, but otherwise unremarkable, group of signal peaks. In some forms, the SSEP system 100 may find the reference vertex by finding a maximum or minimum value of opposite polarity to the primary vertex within a region surrounding the primary vertex. Further examples of how the SSEP system 100 may identify the primary and reference vertices of the baseline SSEP potentials are detailed below.

[0031] If the baseline classification algorithm determines in step 2-3 that there are no monitorable baseline potentials, the SSEP system 100 may, in step 2-4, modify acquisition or processing parameters to improve the quality or monitorability of the SSEP potentials. The purpose of this modification is to enable the system to establish a monitorable baseline. For example, the SSEP system 100 may increase the stimulation current administered by the stimulating electrodes 103, reduce the stimulation rate by the stimulating electrodes 103, increase the number of single trials included in the ensemble average, or reduce the amount of digital filtering applied to the SSEP recordings. The improvement or optimization of these parameters may be driven by output from the baseline classification algorithm or other features calculated from the SSEP recordings, such as signal-to-noise ratio. The improvement of the acquisition and processing parameters may be automated so as to eliminate the need for user intervention. Further examples of how the SSEP system 100 may improve or optimize acquisition and / or signal processing parameters in response to an unmonitoring classification are detailed below.

[0032] If it is determined that there are monitorable baseline potentials, the SSEP system 100 can be used to monitor the subject 101 during ongoing activities, such as during surgery or other medical procedures. Specifically, in step 2-5, the SSEP system 100 acquires an ongoing recording and identifies potential vertices from the recording. The dominant and reference vertices can be identified in a manner similar or identical to step 2-2. In an exemplary embodiment, the SSEP system 100 identifies vertices of interest in the ongoing SSEP potential using parameters established when identifying the dominant and reference vertices of the baseline potential and information from a previous ongoing SSEP recording. The SSEP system 100 then compares the ongoing potential with the baseline potential in step 2-6. For example, the SSEP system 100 can compare amplitude and / or latency based on the identified dominant and reference vertices. In an exemplary embodiment, the amplitude comparison between the baseline and ongoing potentials is the amplitude defined by the vertical distance between the dominant and reference vertices, and the latency comparison is defined by the latency of the dominant vertex. Such comparisons are performed according to some defined criteria to assess whether the ongoing potentials are consistent with a baseline waveform that may indicate a normal state or inconsistent with a baseline waveform that may indicate a threat of nerve damage.

[0033] If the criteria defined in steps 2-7 are met, there is a risk of iatrogenic nerve injury, and the SSEP system 100 issues an alert in steps 2-8. In some forms, the alert and display unit 106 of the SSEP system 100 generates an alert. There can be many types of alerts. For example, the alert can include an audible alert, a visual alert, and / or a tactile alert. The alert is intended to notify a physician and / or other person that the subject 101 may be at risk for iatrogenic nerve injury. The physician and / or other person can then decide to take action, such as repositioning the subject, to prevent excessive pressure / tension on the nerve or reduced blood supply in order to eliminate or reduce the risk of nerve injury to the subject 101.

[0034] Steps 2-5 through 2-8 may be repeated in progress throughout the entire surgical or other medical procedure. Alternatively, the method may end if step 2-9 causes SSEP system 100 to terminate the method. The method may end when input is received to terminate monitoring of subject 101, for example, via alert and display portion 106 of SSEP system 100. Such input may be provided, for example, by a physician and / or other person after the surgical or other medical procedure is completed, or for other reasons.

[0035] The criteria defined above for steps 2-7 may include, for example, a defined decrease in amplitude (i.e., the vertical distance from the apex marker to the onset or offset marker) and / or a defined increase in latency (i.e., the time from stimulation to the apex). During surgery or other procedures, the ongoing recording can be updated in a sliding window manner on the alert and display unit 106. For each updated ongoing recording, the SSEP system 100 can calculate the amplitude and latency of the evoked potential (latency is the timing of the apex). The amplitude and latency of the ongoing potential are compared to the amplitude and latency of the baseline potential. Standard thresholds used in intraoperative neuromonitoring to determine a significant change from baseline are a 50% decrease in the amplitude of the ongoing potential relative to the baseline potential or a 10% increase in the latency of the ongoing potential relative to the baseline (a 10% delay in the apex occurrence time). Other thresholds are possible. In some forms, the SSEP system 100 applies an additional level of logic beyond these thresholds, namely, a "voting" scheme used to reduce false positive alarms, as disclosed in U.S. Patent No. 11,197,640. When a single ongoing potential exceeds a defined threshold, it is considered a vote for an alarm. The number of alarm votes within a sliding window must exceed a defined alarm vote threshold before an alarm is issued by the SSEP system 100 via the alarms and displays 106. For example, an alarm is issued when 80% (24 / 30) of the previous 30 active traces exceed the SSEP system's 100 threshold ("Y" votes).

[0036] In some embodiments, the voting process is asymmetric, and different voting ratios can be used to trigger (onset) and deactivate (offset) a final warning to the user that neurological damage to the subject is possible or imminent. The warning process examines the metadata (alert votes) of overlapping individual epochs. Because actual changes in any data continue and always result in a 100% (or nearly so) Y / N alert vote, the specificity of the warning process can be manipulated based on the votes, independent of the system's sensitivity. For example, a user might change the ratio to be less specific by including only 30% or 50% of the votes, as opposed to a more specific 80% vote. In this way, the number of false alarms can be reduced without actually changing the process's sensitivity to detecting actual, ongoing changes. The voting ratio can reduce the impact of noisy signals. In some embodiments, SSEP recordings with large fluctuations are removed from the voting process, eliminating intermittent noise that escapes the frequency filter. Thus, in some embodiments, the method for dealing with the impact of noise in SSEP recordings is improved. Furthermore, the effect of this noise is neutralized by the voting process itself.

[0037] There may be a variety of SSEP system 100 components. In some forms, the NIDD 104 includes a computer. In some forms, the NIDD 104 is electrically, electronically, and / or mechanically coupled to the platform integration unit 105 and / or the alarm and display unit 106. In some forms, the platform integration unit 105 is mechanically and / or electronically coupled to the platform 107 and / or the NIDD 104. In some forms, the platform integration unit 105 is integrated into the platform 107 and / or the NIDD 104. In some forms, the alarm and display unit 106 is mechanically and / or electrically coupled to the NIDD 104 and / or the platform 107. In some forms, the alarm and display unit 106 is integrated into the platform 107 and / or the NIDD 104 or is displayed on another screen available to the user.

[0038] As previously mentioned, the NIDD 104 of the SSEP system 100 may include a computer. In some forms, the NIDD 104 uses software executed by a processor of the computer to perform the methods described herein. According to embodiments of the present disclosure, a non-transitory computer-readable medium is provided having recorded thereon statements and instructions that, when executed by a processor, perform the methods described herein. Examples of non-transitory computer-readable media include a solid-state drive (SSD), a hard disk drive, a compact disc (CD), a digital video disc (DVD), a Blu-ray disc (BD), a memory stick, or any suitable combination thereof.

[0039] It should be noted that non-software implementations are possible and within the scope of this disclosure. Other implementations include additional or alternative hardware components, such as appropriately configured FPGAs (Field Programmable Gate Arrays) and / or ASICs (Application Specific Integrated Circuits). More generally, the NIDD 104 of the SSEP system 100 may be implemented with any suitable combination of hardware, software, and / or firmware.

[0040] Further detailed examples are provided below of how SSEP system 100 can identify the dominant and reference apexes of baseline SSEP potentials, how SSEP system 100 can determine whether monitorable baseline locations exist, how SSEP system 100 can improve the collection and / or processing of SSEP recordings, and how SSEP system 100 can identify the dominant and reference apexes of ongoing SSEP potentials to enable monitoring of nerve damage. It should be understood that the details of these examples are very specific and other configurations are possible and within the scope of this disclosure.

[0041] Vertex identification in baseline SSEP recordings Referring now to Figure 3, a flowchart of a method for identifying peaks of interest in evoked potentials in a baseline SSEP recording is shown. This method is an exemplary embodiment of step 2-2 of the method described above in connection with Figure 2. Thus, like the method of Figure 2, the method of Figure 3 can be performed by an SSEP system, such as the SSEP system 100 shown in Figure 1. In a particular embodiment, the method is performed by the NIDD 104 of the SSEP system 100. More generally, the method of Figure 3 can be performed by any suitable SSEP system.

[0042] Identifying the vertex of interest in the baseline SSEP recording is an important step in the NIDD 104. In a preferred embodiment, vertex prominence is used to identify the principal vertex of the evoked potential, although other embodiments are possible. The selection of the principal vertex can be performed based on the maximum value, maximum absolute value, slope, or second derivative near the vertex.

[0043] The method of Figure 3 will now be described with reference to Figures 4 and 5, which are graphs of monitorable baseline SSEP potentials in an upright (Figure 4) and an inverted (Figure 5) display. It will be understood that the configuration shown in Figure 3 and the baseline recordings shown in Figures 4 and 5 are very specific for illustrative purposes only, and that other configurations and other baseline recordings are possible and within the scope of this disclosure.

[0044] In step 3-1, the SSEP system 100 identifies candidate vertices in the upright and inverted views. Note that there may be multiple candidate vertices in the baseline recording. For example, six candidate vertices are identified in FIG. 4, and five candidate vertices are identified in FIG. 5. The goal is to identify which of these candidate vertices is the most prominent.

[0045] In step 3-2, for each candidate vertex identified in step 3-1, the SSEP system 100 calculates its prominence. To calculate the prominence, the SSEP system 100 first extends a horizontal line from the candidate vertex to the left or right until it either (a) intersects the signal due to a higher vertex or (b) reaches the left or right edge of the signal. In some forms, the horizontal line extended from the vertex for the prominence calculation is further limited by the analysis range. Next, the SSEP system 100 finds the minimum signal for each of the two intervals defined above. This point is one of the local minima or signal endpoints. Finally, the SSEP system 100 determines the prominence based on the vertical distance between the vertex and the larger of the two interval minima.

[0046] In step 3-3, candidate vertices are filtered based on the analysis range. The analysis range is the range of expected latencies within which evoked potentials are likely to occur. The analysis range can be based on the stimulation site, since the relationship between the distance evoked potentials travel through the nervous system between the stimulation and recording sites is roughly known (the longer the travel distance, the longer the expected latency). Natural anatomical, neurophysiological, and pathophysiological variations between subjects will result in variability in expected latencies. The analysis range for a given stimulation site can be set from values obtained from the literature, expert opinion, or previous subject data. In some forms, the analysis range can correspond to evoked potential responses from other stimulation or recording channels. Note that filtering of candidate vertices by the analysis range can occur before the prominence calculation in step 3-2.

[0047] In this way, the SSEP system 100 identifies all positive and negative extremes (also called "peaks and valleys" or "maxima and minima") and calculates their respective elevations. As shown in Figures 4 and 5, for both the "upright recording" and the "inverted recording," the system selects the most elevated extremes within the analysis range and compares their elevation values.

[0048] In step 3-4, the SSEP system 100 selects the most prominent candidate vertex as the dominant vertex. In the illustrated example, the most prominent candidate vertex 501 is identified in FIG. 5 as having a prominence of 1.26. Note that because the recording is displayed inverted in FIG. 5, this vertex is actually a trough, or minimum, in the baseline SSEP potential. The other candidate vertices, represented by vertical lines, are shown with smaller prominences.

[0049] In this way, the marking algorithm identifies all positive and negative extremes (also called "peaks and valleys" or "maxima and minima") within the analysis range and calculates the elevation for each. For positive extrema, the recorded waveform is used to find the peak and calculate the elevation. When calculating the elevation for a valley, the waveform is inverted and the elevation calculation is performed as described above. As shown in Figures 4 and 5, for both the "normal waveform" and the "inverted waveform," the most elevated extrema within the analysis range are selected and their elevation values are compared with each other. Of these two candidate peaks, the most elevated maximum or minimum is selected for marker placement.

[0050] By calculating the elevation of both the positive and negative extremes, major vertices can be identified as either peaks or valleys. If the peak marker is placed at a minimum, or valley (i.e., at a maximum in the inverted record), the peak is said to have negative polarity. If the peak marker is placed at a maximum, the peak has positive polarity.

[0051] In steps 3-5, an onset vertex 401 and an offset vertex 402 (see FIG. 4) are identified on either side of the most prominent vertex 501. These onset vertex 401 and offset vertex 402 have opposite polarity to the most prominent vertex 501. The onset vertex 401 is identified in the time frame leading up to the most prominent vertex 501, and the offset vertex 402 is identified in the time frame following the most prominent vertex 501. During the onset and offset periods, the extrema with the largest absolute voltage difference from the vertex marker are selected as the onset vertex and the offset vertex, respectively. In another embodiment, only the extrema immediately adjacent to the main vertex are considered as the potential reference vertex.

[0052] In steps 3-6, the onset vertex 401 or offset vertex 402 that maximizes amplitude to the most prominent vertex 501 is selected as the reference vertex. In other embodiments, slope, second derivative, or other evoked potential morphology features may be used to select the reference vertex. Additionally, in some embodiments, markers used to determine amplitude and latency may be placed at both the onset vertex 401 and offset vertex 402.

[0053] In some embodiments, new apex markers may be placed on each updated SSEP ensemble average to track evoked potential amplitude and latency throughout the procedure. Apex marking for ongoing waveforms is similar to the method described above, although adjustments may be made to provide consistent marking between baseline and ongoing waveforms. Specifically, for ongoing waveforms, the polarity of the baseline epoch marking and the determination of the onset or offset marking of the baseline epoch are applied to the ongoing waveform. Additionally, the analysis range may be repositioned around the most prominent apex of the baseline waveform. In some embodiments, the analysis range for ongoing waveform apex marking may be asymmetric, since latency is more likely to be extended than shortened.

[0054] Determining whether baseline SSEP waveforms can be monitored Referring now to FIG. 6, a flowchart of a method for determining whether a baseline recording contains monitorable baseline potentials is shown. This method is an exemplary implementation of steps 2-1 through 2-3 of the method described above in connection with FIG. 2. Thus, like the method of FIG. 2, the method of FIG. 6 may be performed by an SSEP system, such as the SSEP system 100 shown in FIG. 1. In a particular implementation, the method is performed by the NIDD 104 of the SSEP system 100. More generally, the method of FIG. 6 may be performed by any suitable SSEP system.

[0055] The method of Figure 6 is now described with reference to Figure 7, which is a graph of a baseline recording. It should be understood that the configuration shown in Figure 6 and the baseline recording shown in Figure 7 are very specific for illustrative purposes only, and that other configurations and other baseline recordings are possible and within the scope of this disclosure. It should be understood that the configuration shown in Figure 6 and the baseline recording shown in Figure 7 are very specific for illustrative purposes only, and that other configurations and other baseline recordings are possible and within the scope of this disclosure.

[0056] In step 6-1, the SSEP system 100 obtains two independent sets of SSEP records. Each set includes a number of non-overlapping SSEP single trials, for example, 280 non-overlapping SSEP single trials, or any other suitable number of non-overlapping SSEP single trials. In some configurations, each set includes the same number of non-overlapping SSEP single trials. Other configurations are possible. The SSEP system 100 averages the SSEP single trials for each set to generate a record representative of that set. This generates two SSEP records.

[0057] In step 6-2, the SSEP system 100 generates two sets of baseline records, for example, by averaging all SSEP single trials or by averaging two baseline SSEP recordings. For example, if each independent set is an average of 280 single trials, the grand ensemble average may be the average of all 560 single trials. As a result, as shown in FIG. 7, the SSEP system 100 generates three records: a first set of first SSEP recordings 701, a second set of second SSEP recordings 702, and a third record 703, which is a baseline SSEP recording.

[0058] In steps 6-3 and 6-4, the SSEP system 100 finds the main and reference vertices of the potentials in the three recordings 701-703 generated in step 6-2. The vertex search in step 6-3 for the baseline SSEP recording 703 is performed as described above and shown in FIGS. 3-5. The amplitude between the vertex of the baseline SSEP potential and the reference marker is retained for comparison with a minimum amplitude threshold in a later step. The vertex search in the two sets 701, 702 is performed using the ongoing vertex search method described below in FIG. 11. This is done to ensure consistent marker placement in the two sets, using the same polarity and onset / offset markings as the baseline SSEP potential.

[0059] In step 6-5, the SSEP system 100 calculates features based on the SSEP recordings and the identified vertices. A variety of features are possible. Examples of features include the amplitude of the potential in the first recording 701, the amplitude of the potential in the second recording 702, the slope between the major vertex of the potential in the baseline recording 703 and the fiducial marker, the absolute value of the difference in the major vertex latency of the potential between the first recording 701 and the second recording 702, a signal-to-noise ratio (SNR) value calculated from the first recording 701 and the second recording 702, and a measure of the vertex amplitude relative to the root mean square (RMS) of each recording 701-703, as shown in FIGS. 6 and 7 . Additional and alternative features are also possible.

[0060] In some embodiments, the features may be derived from the vertex marking algorithm applied in steps 6-3 and 6-4, or from additional processing performed on the recordings 701-703. The first two features in the exemplary feature sets described above and shown in FIGS. 6 and 7 are the individual potential amplitude values for each of the two sets 701, 702. Amplitude is defined as the absolute difference in voltage between the dominant and the reference vertex. The third feature is the absolute value of the slope between the dominant and the reference vertex of the baseline SSEP potential. The fourth feature is the absolute difference in peak latency or peak marker timing from the potentials in each set 701, 702. The fifth feature is a measure of signal-to-noise ratio (SNR), e.g., SNR as defined by Coppola [Coppola R, Tabor R, Buchsbaum MS. Signal to noise ratio and response variability measurements in single trial evoked potentials. Electroencephalograph and Clinical Neurophysiology, 1978, 44:214-222.]

[0061] This feature is calculated by first determining the correlation coefficient (r) between the two pairs of baseline recordings. This correlation is called the "signal" and divided by (1-r), an estimate of noise. This feature can also be thought of as a measure of inter-pair variability. The sixth feature is another SNR measure derived from the ratio of peak amplitude to RMS of the recordings. This is calculated for all three recordings 701-703 and summed. Finally, a transformation can be applied to the feature values to adjust for skewed feature distributions. This can improve classification performance and / or interpretability for certain types of machine learning classification algorithms. Other sets of features are possible. Potential feature candidates include features quantifying the morphology, repeatability, or SNR of the recordings 701-703.

[0062] In some embodiments, the SSEP system 100 implements a machine learning classifier in step 6-6 to determine whether monitorable baseline SSEP potentials exist based on the features calculated in step 6-5. The machine learning classifier is a supervised machine learning model used for binary classification. In an exemplary embodiment, a support vector machine (SVM) is used as the machine learning classifier. In other embodiments, alternative supervised machine learning algorithms such as gradient boosted decision trees, random forests, logistic regression, and neural networks can be used. In some embodiments, for example, when a deep learning neural network framework is applied, it is not necessary to design specific features.

[0063] In some embodiments, a machine learning classifier is trained to classify baseline potentials as either monitorable or unmonitored based on a database of labeled baseline examples and their corresponding features. Recordings in this database can be judged as either monitorable or unmonitored by an expert clinical assessor. In other embodiments, the machine learning algorithm may apply a multi-class classification approach rather than a binary classification approach. Possible examples of classes used for multi-class classification include monitorable, identifiable but unmonitored, and unidentifiable.

[0064] In some embodiments, additional criteria are applied after machine learning classification. For example, if the minimum amplitude threshold is not exceeded in step 6-7, the baseline potential is deemed unmonitored in step 6-8. In such a case, more single trials can be collected, and the SSEP system 100 can again attempt to find a monitorable baseline potential. On the other hand, if the minimum amplitude threshold is exceeded in step 6-7, the baseline potential is deemed monitorable in step 6-9. Various minimum amplitude thresholds are possible. The minimum amplitude threshold may be, for example, 0.2 μV or other suitable value.

[0065] In some embodiments, the algorithm is stimulation- and recording-site independent. All features in the exemplary feature sets described in Figures 6 and 7 are stimulation / recording-site independent, as stimulation / recording-site-specific parameters (e.g., analysis range) can be applied during the vertex-marking algorithm. This allows the machine learning classifier to be trained on labeled data from any stimulation / recording site. In other embodiments, the features or training data can be specific to a particular stimulation and recording site pair. Additionally, the minimum amplitude threshold can vary depending on the stimulation and / or recording site.

[0066] In some forms, establishing a baseline is a static process in which two data sets (e.g., 280 single trials each, for a total of 560 single trials) are collected and processed by a classification algorithm. If the classification algorithm determines that there are no monitorable potentials, the system may enter a monitoring disabled state for that SSEP channel. In some forms, the SSEP system 100 collects a third set of SSEP single trials and attempts to establish a baseline using sets two and three.

[0067] In some forms, the SSEP system 100 can establish an early baseline if the confidence that the baseline can be monitored (probability of a positive classification) exceeds a high threshold (e.g., 98%).

[0068] In some configurations, the user can override the algorithmically determined baseline classification. If the initial baseline waveform is classified as negative and the user has not yet reset or overridden the baseline, a second attempt is made to classify the baseline as positive. In this example, a third ensemble of 280 single trials is acquired, and a new superensemble is generated from the second and third ensembles and reclassified. If a second negative classification occurs, the baseline is considered "unmonitored."

[0069] In some embodiments, the supervised machine learning algorithm is trained using data from an annotated clinical database. Waveforms in this database are judged by expert clinical raters to be either supervised or unsupervised. Each waveform in the database can be combined with a subsequent waveform to simulate two independent data sets used for baseline classification. Features can then be computed for these waveform sets as described above, and if both underlying waveforms are judged to be supervised, a supervised class label is applied to the feature sample.

[0070] In some embodiments, data for the annotated database is obtained from saphenous nerve stimulation sites and tibial stimulation sites. However, because the feature set is montage-independent, the algorithm is montage-independent. All features in the feature set are derived from vertex marking, and montage-specific parameters can be applied at the vertex marking algorithm stage. In some embodiments, a machine learning algorithm is trained on all available data, and the resulting support vectors are saved and stored in the SSEP system 100 or elsewhere for use in classifying data.

[0071] In other configurations, other methods can be used to identify apexes. For example, apex markers are placed at the most prominent extremes (peaks or troughs) within the analysis range. Onset markers are placed at the maximum value of opposite polarity to the apex marker within the window leading up to the apex marker. To calculate features derived from apex markings, first apex and onset markers are placed on the superensemble average that constitutes the baseline waveform. The amplitude of the markers on the superensemble baseline waveform is retained for comparison with a minimum amplitude threshold. The polarity of the baseline apex (i.e., whether the apex (positive polarity) or trough (negative polarity) is the most prominent apex) is retained to contribute to determining the five features for classification. The apex and onset markers are placed in two independent pairs, and the polarity is forced to match the polarity of the baseline waveform. The analysis period for placing apex markers in the two independent pairs is extended if the baseline apex marker is close to the edge of the analysis period. The length of this extension depends on the width of the baseline analysis period and how close the apex is to the edge of that period.

[0072] Automatic optimization of baseline SSEP recording In a preferred form, the SSEP system 100 has a baseline optimization state in which baseline recording collection is adaptively performed. An advantage of machine learning classification algorithms is that they can provide not only a classification, but also a probability of that classification, which indicates how confident the algorithm is in its classification. This value, along with other calculations (e.g., amplitude or SNR), provides the ability to optimize or improve the baseline collection phase of SSEP monitoring in ways that may not be apparent or available to the average neuromonitoring technician.

[0073] A common measure of confidence in the results provided by a binary classification algorithm is the positive classification probability. The positive classification probability represents the likelihood assigned by the trained model that a sample belongs to the positive class. The positive classification probability represents the likelihood assigned by the trained model that a sample belongs to the positive class. The positive classification probability typically ranges from 0 to 1, with 0 indicating that the model has very high confidence that the sample belongs to the negative class and 1 indicating that the model has very high confidence that the sample belongs to the positive class. Some machine learning models output these probabilities naturally (e.g., neural networks). For other types of models, the positive classification probability must be calculated a posteriori (e.g., by applying Platt scaling to the SVM score). As described above and shown schematically in FIG. 6, for a classification task determining whether a baseline SSEP waveform is observable or unobservable, the positive classification probability (or observable class probability) is the probability that the trained model assigns the sample to the observable baseline SSEP potential.

[0074] An advantage of the disclosed approach is that the SSEP system 100 can periodically (e.g., every 10 collected trials) divide all collected single trials into two sets and perform a comparison using a machine learning classifier. By periodically evaluating the monitoredability of the collected data, the SSEP system 100 can establish an early baseline if the data quality is determined to be high, or take steps to improve signal quality if the recorded data is not monitored. This approach provides clinical benefits to the SSEP system 100.

[0075] In a preferred form, the SSEP system 100 assesses baseline SSEP potentials after 280, 420, and 560 single trials are collected. Baseline assessments can occur at other frequencies. If 280 or 420 single trials result in a positive classification probability from the machine learning classifier that exceeds a high threshold (e.g., 0.98, or 98%), the SSEP system 100 can establish an early baseline. If the system does not establish an early baseline, a nominal threshold of 0.5 (i.e., 50%) positive classification probability is applied after 560 single trials are collected.

[0076] Referring now to FIG. 8, a set of charts for monitorable examples with good quality data are shown, illustrating scenarios where SSEP system 100 can establish an early baseline. In the main chart, data points indicate the probability of a positive classification when SSEP system 100 evaluates baseline monitorability every 10 samples. The solid data points indicate the results of the classification algorithm at preferred evaluation frequencies (after collecting 280, 420, and 560 single trials). The upper dashed line indicates a high threshold of 0.98 (i.e., 98%) for the probability of a positive classification. The lower dashed line is the nominal threshold of 0.5 (i.e., 50%). Charts 8A, 8B, and 8C show two independent sets of data evaluations after 240, 420, and 560 single trials, respectively, and the resulting baselines (with apex and fiducial markers).

[0077] This example shows a very high probability of positive classification (greater than 0.98) even though fewer single trials were collected. Because the probability of positive classification is so high, it can be concluded that monitorable baseline potentials exist with a relatively small number of single trials, allowing SSEP data and a baseline to be established more quickly. The baseline (8A) provided for monitoring when set after 280 single trials is substantially indistinguishable from the baseline (8B) set after 560 single trials, providing equivalent monitoring. In a preferred embodiment, the baseline would be set after collecting 280 single trials in this example.

[0078] In some forms, if the results of 280 or 420 single trials yield a low probability of a positive classification from the machine learning classifier, indicating a possible unsupervised classification after 560 single trials, steps may be automatically taken to improve the signal. For example, the SSEP system 100 may increase stimulation intensity, increase stimulation duration, decrease stimulation frequency, increase the number of single trials to an ensemble average, or change signal processing parameters such as filter settings or a mother wavelet for noise reduction.

[0079] Referring now to FIG. 9 , a set of charts for a non-monitoring example are shown illustrating scenarios that may trigger the SSEP system 100 to expand SSEP collection parameters and signal processing. In this example, the positive classification probability after 280 single trials is very low (0.01, or 1%) and shows no improvement over time. If a static baseline approach were followed, monitoring this channel may not provide meaningful clinical information. In some forms, identifying a low positive classification probability early in the baseline collection period (e.g., after 280 or fewer single trials) triggers a change in collection parameters, such as increasing stimulation intensity, extending the stimulation duration, or modifying signal processing (e.g., filtering). In some forms, the system can store preprocessed SSEP recordings and single trials so that signal processing optimizations can be applied retroactively (i.e., to already collected data). In this manner, the SSEP system 100 is more likely to obtain a monitorable baseline than if a static baseline collection approach were followed.

[0080] Referring now to FIG. 10, a set of charts for a monitorable example is shown, in which the probability of a positive classification improves as more single trials are collected. In this example, the initial probability of a positive classification is approximately 57%, which improves to approximately 96%. Compared to the example of FIG. 8, more single trials of SSEP data may be required to conclude that a monitorable baseline potential exists. While the baseline in FIG. 10 is monitorable, in some configurations, improvements during the SSEP baseline collection period, including increased stimulation intensity and / or others as described above, may be beneficial.

[0081] In some forms, the user can override the algorithmically determined baseline classification. If the initial baseline potential is classified as negative, the user can override the non-monitoring classification and enable monitoring of that channel. The positive classification probability can contribute to the user's decision to override the baseline classification algorithm's decision. If the user overrides a properly classified non-monitoring baseline, this can result in very poor monitoring and unreliable information for the user.

[0082] In some forms, an attempt to override an unmonitored baseline prompts the SSEP system 100 to display confidence information to the user when attempting to override the baseline (e.g., "Do you want to monitor this baseline? The algorithm is 95% confident that this baseline is unmonitored"). There are many other messages and ways to display this information to the user. This message may only be generated if the positive classification probability indicates that the system is very confident in its classification.

[0083] In some configurations, the positive classification probability calculated by the machine learning classifier when a baseline is established is also used to select stimulation and recording channels to display and to provide warning information to the user. For example, depending on the recording montage, there may be multiple channels of cranial information (e.g., CPz-FPz and CP3-CP4). Often, the signal quality of one channel may be higher than the others, allowing for improved monitoring (increased sensitivity and specificity). The positive classification probability can be used to select one of these for ongoing monitoring or to calculate the contribution of each to leading to a warning state.

[0084] The above refinement and optimization process relies on the positive classification probability provided by the machine learning classifier. Alternatively, the refinement and optimization process can be achieved by other values derived from the SSEP data (e.g., amplitude or SNR) that are independent of or linked to the positive classification probability.

[0085] Vertex identification in ongoing SSEP recordings Referring now to FIG. 11, a flowchart of a method for identifying peaks of interest in evoked potentials in an ongoing SSEP recording is shown. This method is an exemplary implementation of steps 2-5 of the method described above in connection with FIG. 2. Thus, like the method of FIG. 2, the method of FIG. 11 can be performed by an SSEP system, such as the SSEP system 100 shown in FIG. 1. In a particular implementation, the method is performed by the NIDD 104 of the SSEP system 100. More generally, the method of FIG. 11 can be performed by any suitable SSEP system.

[0086] Identifying the dominant and baseline vertices within ongoing SSEP recordings is a critical step in NIDD104 monitoring because it allows the amplitude and latency of the evoked potential response to be determined, compared to those found in the baseline waveform, and triggers an alert if alert criteria are met. To ensure accurate comparisons with baseline potentials, information from baseline vertex marking must be applied when identifying vertices in ongoing SSEP recordings. Furthermore, to ensure the same regions of the evoked potentials are being compared, this method tracks the dominant vertex of the ongoing potentials over time rather than relying solely on vertex prominence.

[0087] In step 11-1, the SSEP system 100 identifies a candidate vertex in the ongoing recording that has the same polarity as the dominant vertex of the potential in the baseline SSEP recording. If the dominant vertex of the baseline SSEP potential is found to be a maximum, then the dominant vertex in the subsequent ongoing SSEP recording will be a maximum. Similarly, if the dominant vertex of the baseline SSEP potential is found to be a minimum, then the dominant vertex in the subsequent ongoing SSEP recording will be a minimum.

[0088] In step 11-2, candidate vertices are filtered based on the analysis range. The analysis range for ongoing vertex marking corresponds to the latency of the primary vertex of the baseline SSEP potential. This analysis range or period can be a percentage of the baseline latency or a fixed time. In a preferred embodiment, the ongoing analysis range is asymmetric because latencies are more likely to lengthen than shorten during monitoring, ranging from 17% before the primary vertex to 30% beyond the primary vertex. In some embodiments, the ongoing vertex search / analysis range can depend on the stimulation or recording site. Other ongoing vertex search / analysis ranges are possible within the scope of this disclosure.

[0089] In step 11-3, the amplitude of the potential in the previous ongoing SSEP recording is compared to a threshold. In some forms, this threshold is 25% of the baseline SSEP potential amplitude. Other thresholds are possible, and the threshold may correspond to the noise floor or SNR.

[0090] If the amplitude exceeds a threshold, the evoked potential is expected to be diminished but still present during the response. In this situation, a vertex tracking technique is preferred. In step 11-4, this tracking technique is applied by selecting as the primary vertex the candidate vertex from the previous ongoing SSEP recording that is closest in latency to the primary vertex of the potential.

[0091] If the amplitude is below the threshold in step 11-3, the evoked potential may no longer exist, or noise may be equal to or greater than the response. In this case, the method is similar to the baseline vertex marking algorithm. In step 11-5, the prominence of each vertex is calculated, and in step 11-6, the most prominent vertex within the ongoing analysis range is selected as the primary vertex.

[0092] In step 11-7, a reference vertex for the potential is identified within the region surrounding the primary vertex. The reference vertex for the ongoing potential must be in the same direction relative to the primary vertex as the reference vertex for the baseline SSEP potential. For example, if the baseline reference vertex is at onset (before the primary vertex), this must also be true for the ongoing potential.

[0093] Finally, in step 11-8, the reference vertex that maximizes the amplitude of the ongoing SSEP potential is selected as the reference. Other methods for selecting the reference vertex are also possible. For example, the reference vertex could be the vertex of opposite polarity closest to the primary vertex in the direction specified by the baseline vertex marking.

[0094] Case studies Referring now to FIG. 12, a graph of baseline and ongoing SSEP recordings is shown for an example of monitorable baseline potentials. This view of the data is referred to as a "waterfall" view, which displays a stack of independent sets (non-overlapping ensemble averages). In this case, each recording is an ensemble average of 280 single trials collected with 3.1 Hz stimulation (approximately 90 seconds of data collection). The time for the case elapses from the start of the case at the top of the figure, and the last recording collected for that case is shown at the bottom of the figure. The data here is viewed via a tool that allows retrospective viewing of the case data.

[0095] The top two recordings 1201 and 1202 are the pair used to form the baseline. The third recording is the baseline recording 1203. In the first example, the evoked potentials are quite clear. There is also a positive classification probability near the right of the baseline, here 0.95 (95%), indicating that the algorithm is able to monitor the baseline with a high degree of confidence. Below the baseline is the ongoing SSEP recording. The amplitude and latency of the potentials in the ongoing SSEP recording are stable and similar to the baseline, with no warning of significant changes.

[0096] Referring now to FIG. 13, a graph of the baseline and ongoing SSEP recordings are shown for a case of an unmonitored baseline potential. In this example, the baseline recording 1303 (the third thick trace with a marker) is classified as unmonitored. Although there appears to be an evoked potential response in the baseline, the baseline classification algorithm is highly confident that baseline recording 1303 does not contain any monitorable potentials, with a 2% probability of positive classification. The next four ongoing SSEP recordings 1304–1307 are grayed out because this channel does not exhibit a monitorable baseline. The user then overrides the baseline classification (approximately half the thick line 1308). The channel quickly transitions to a warning state (see markers 1309A and 1309B), and the markers in the ongoing trace do not actually mark evoked potentials. Monitoring after the override does not provide reliable information to the user. By informing the user of the unlikely baseline waveform given by the algorithm, the user can avoid inappropriately overriding the classification, as shown here.

[0097] Referring now to FIG. 14, a graph of SSEP recordings for a case of monitorable baseline potentials is shown. The SSEP recordings include a first SSEP recording 1401 of a first set, a second SSEP recording 1402 of a second set, and a third recording 1403, which is a baseline SSEP recording. The vertical dashed lines indicate the analysis range of the baseline apex marking. The potentials of SSEP recordings 1401-1403 are consistent with the expected morphology, and the case shows a 99% probability of positive classification.

[0098] 15, a graph of SSEP recordings for a case of unmonitored baseline potentials is shown. The SSEP recordings include a first SSEP recording 1501 of a first set, a second SSEP recording 1502 of a second set, and a third recording 1503, which is a baseline SSEP recording. SSEP recordings 1501-1503 are not very consistent with each other, with the case exhibiting only a 3% probability of positive classification.

[0099] 16, a graph of SSEP recordings is shown for a case where it is not entirely clear whether baseline potentials are monitorable or not. The SSEP recordings include a first SSEP recording 1601 of a first set, a second SSEP recording 1602 of a second set, and a third recording 1603, which is a baseline SSEP recording. SSEP recordings 1601-1603 are noisy and only somewhat consistent with each other, and the case exhibits a positive classification probability of 43%, which could potentially be a false negative classification.

[0100] Many modifications and variations of the present disclosure are possible in light of the above teachings.

[0101] Embodiment 1: A method performed by an SSEP (Somatosensory Evoked Potential) system, the method comprising: obtaining at least one SSEP recording from a patient; determining or calculating the presence and characteristics of monitorable baseline SSEP potentials based on the at least one SSEP recording; obtaining ongoing SSEP recordings from the patient; comparing the ongoing SSEP potentials to the monitorable baseline potentials; and issuing an alert if the ongoing SSEP potentials deviate from the monitorable baseline potentials according to defined criteria.

[0102] Embodiment 2: The method of embodiment 1, wherein obtaining at least one SSEP recording includes obtaining two SSEP recordings based on two corresponding independent SSEP data sets, and when a baseline SSEP potential is deemed monitorable based on two SSEP potentials identified from the two independent SSEP data sets and potential characteristics identified from the grand ensemble average, a monitorable baseline potential is calculated based on the grand ensemble average of the independent SSEP data sets.

[0103] Embodiment 3: The method of embodiment 2, further comprising calculating features of the two SSEP potentials, the features including: an amplitude of a first potential of the two SSEP recordings; an amplitude of a second potential of the two SSEP recordings; an absolute value of the amplitude difference between the first potential and the second potential; an absolute value of the onset latency difference between the first potential and the second potential; an absolute value of the peak latency difference between the first potential and the second potential; and an SNR (signal-to-noise ratio) of the first and second potentials compared to the entire recording.

[0104] Embodiment 4: The method of any one of embodiments 1 to 3, wherein computing features includes computing peak / valley markers for each baseline SSEP potential by identifying candidate vertices in upright and inverted views of the baseline SSEP recording and identifying which candidate vertices are most prominent based on how distinct the candidate vertices are by their intrinsic height and their position relative to other candidate vertices.

[0105] Embodiment 5: The method of any one of embodiments 0 to 0, comprising determining whether the baseline epoch potential is considered monitorable using a support vector machine (SVM) that classifies the baseline epoch potential as either monitorable or non-monitored based on features of two SSEP potentials.

[0106] Embodiment 6: The method of any one of embodiments 1 to 5, further comprising determining whether the baseline epoch potential is considered monitorable using a wavelet convolutional neural network that classifies the baseline epoch potential as either monitorable or non-monitored based on features of the two SSEP potentials.

[0107] Embodiment 7: The method of embodiment 0 or embodiment 0, further comprising determining a confidence value for whether the baseline epoch potentials are monitorable.

[0108] Embodiment 8: The method of embodiment 0, comprising adapting the size of two corresponding independent SSEP datasets depending on the confidence value.

[0109] Embodiment 9: The method of any one of embodiments 0 to 0, comprising calculating peak / valley markers of the ongoing SSEP potentials for each ongoing SSEP recording by identifying candidate vertices in upright and inverted views of the ongoing SSEP recording and identifying which candidate vertex is most prominent based on how prominent the candidate vertex is by its intrinsic height and its position relative to other candidate vertices, wherein the most prominent candidate vertex is used to compare the ongoing SSEP potentials with a monitorable baseline potential.

[0110] Embodiment 10: The method of embodiment 0, wherein the defined criteria include a defined decrease in amplitude based on reduced prominence and / or a defined increase in latency based on peak delay.

[0111] Embodiment 11: The method of any one of embodiments 0 to 0, wherein the warning comprises an audible warning, a visual warning, and / or a tactile warning.

[0112] Embodiment 12: The method of any one of embodiments 0 to 0, further comprising identifying artifacts in ongoing SSEP recordings and potentials due to the presence of anesthesia in the patient and / or noise from the surrounding environment, and compensating for the artifacts to reduce unnecessary alarms.

[0113] Embodiment 13: A non-transitory computer-readable medium having recorded thereon descriptions and instructions that, when executed by a processor of an SSEP (Somatosensory Evoked Potential) system, configure the processor to perform the method described in any one of embodiments 0 to 0.

[0114] Embodiment 14: A SSEP (Somatosensory Evoked Potential) system comprising a recording electrode configured to generate an electrical recording of a patient's nervous system, a recording electrode configured to detect potentials generated as a stimulus traverses the nervous system, and a nerve injury detection device coupled to the recording electrode and configured to perform the method of any one of embodiments 0-0.

[0115] Embodiment 15: The nerve damage detection device is a SSEP system as described in embodiment 0, comprising a processor and a non-transitory computer-readable medium having descriptions and instructions recorded thereon that, when executed by the processor of the SSEP system, configure the processor to realize the nerve damage detection device.

[0116] Embodiment 16: A method comprising any combination of the step or steps described and / or illustrated herein.

[0117] Embodiment 17: A non-transitory computer-readable medium having recorded thereon statements and instructions that, when executed by a processor of a device, configure the processor to perform a method including the step or any combination of steps described and / or illustrated herein.

[0118] Embodiment 18: A device comprising any combination of the component or components described and / or illustrated herein.

[0119] Embodiment A: A method for determining or calculating the presence, absence, and monitorability of SSEP (somatosensory evoked potential) potentials in one or more SSEP recordings, comprising: obtaining at least one SSEP recording from a subject; analyzing the at least one SSEP recording by evaluating the peak prominence contained in the at least one SSEP recording; determining or calculating the presence and characteristics of monitorable baseline SSEP potentials based on the at least one SSEP recording; obtaining an ongoing SSEP recording from the subject; comparing the ongoing SSEP potentials with the monitorable baseline potentials; and issuing an alert if the ongoing SSEP potentials deviate from the monitorable baseline potentials according to defined criteria.

[0120] Embodiment B: The method of embodiment A, wherein obtaining at least one SSEP recording includes obtaining two SSEP recordings based on two independent SSEP datasets, and if a baseline SSEP potential is deemed monitorable based on the SSEP recordings, two SSEP potentials identified from the two independent SSEP datasets, and characteristics of the potentials identified from the grand ensemble average, a monitorable baseline SSEP potential is calculated based on the grand ensemble average of the independent SSEP datasets.

[0121] Embodiment C: The method of embodiment B, further comprising calculating features of the two SSEP potentials, the features including the amplitude of a first potential of the two SSEP recordings, the amplitude of a second potential of the two SSEP recordings, the absolute value of the slope between the dominant peak and the reference peak in the baseline potential, the absolute value of the peak latency difference of the potentials in the first and second SSEP recordings, the SNR (signal to noise ratio) of the first and second potentials compared to the entire recording, and the ratio of the peak amplitude of the potentials to the RMS (root mean square) of the entire recording for the two SSEP recordings and the grand ensemble baseline recording.

[0122] Embodiment D: The method of embodiment C, wherein computing features includes computing peak / valley markers for each baseline SSEP potential by identifying candidate vertices in the upright and inverted views of the baseline SSEP recording and identifying which candidate vertices are most prominent based on how distinct the candidate vertices are by their intrinsic height and their position relative to other candidate vertices.

[0123] Embodiment E: The method of any one of embodiments B-D, comprising determining whether the baseline SSEP potentials are considered monitorable using a machine learning classification algorithm that classifies the baseline SSEP potentials as either monitorable or non-monitored based on two SSEP potential features.

[0124] Embodiment F: The method of embodiment B, further comprising determining whether the baseline SSEP potentials are considered monitorable using a wavelet convolutional neural network that classifies the baseline potentials as either monitorable or non-monitoable without designing two SSEP potential features.

[0125] Embodiment G: The method of embodiment E or F, further comprising determining a confidence value for whether the baseline potential can be monitored.

[0126] Embodiment H: The method of embodiment G, comprising adapting the sizes of the two corresponding independent SSEP datasets depending on the confidence values.

[0127] Embodiment I: The method of any one of embodiments A-H, comprising calculating peak / valley markers of the ongoing SSEP potential for each ongoing SSEP recording by: identifying candidate vertices in an upright or inverted display of the ongoing SSEP recording depending on the polarity of the baseline SSEP potential; comparing the potential in the previous ongoing SSEP recording to a threshold; and selecting the candidate vertex based on either the vertex having the closest latency to the latency of the potential in the previous ongoing SSEP potential, or selecting the most prominent candidate vertex based on how distinctive the candidate vertex is by its inherent height and its location relative to other candidate vertices depending on an amplitude comparison to the previous ongoing SSEP potential; wherein the selected vertex is used to compare the ongoing SSEP potential with the monitorable baseline potential.

[0128] Embodiment J: The method of embodiment I, wherein the defined criteria comprises a defined decrease in amplitude based on reduced prominence and / or a defined increase in latency based on peak delay.

[0129] Embodiment K: The method of any one of embodiments A-J, wherein the alert comprises an audible alert, a visual alert, and / or a tactile alert.

[0130] Embodiment L: A method according to any one of embodiments A to K, further comprising identifying artifacts in ongoing SSEP recordings and potentials due to the presence of anesthesia in the subject and / or noise from the surrounding environment, and compensating for the artifacts to reduce unnecessary alarms.

[0131] Embodiment M: A non-transitory computer-readable medium having recorded thereon statements and instructions that, when executed by a processor of an SSEP (Somatosensory Evoked Potential) system, configure the processor to perform the method of any one of embodiments A to L.

[0132] Embodiment N: A SSEP (Somatosensory Evoked Potential) system comprising a stimulating electrode configured to generate an electrical response from the subject's nervous system, a recording electrode configured to detect the potential generated as the stimulus traverses the nervous system, and a nerve damage detection device coupled to the recording electrode and configured to perform the method of any one of embodiments A-L.

[0133] Embodiment O: The nerve damage detection device is a SSEP system described in embodiment N, comprising a processor and a non-transitory computer-readable medium having descriptions and instructions recorded thereon that, when executed by the processor of the SSEP system, configure the processor to implement the nerve damage detection device.

[0134] Embodiment P: A method comprising any combination of the step or steps described and / or illustrated herein.

[0135] Embodiment Q: A non-transitory computer-readable medium having recorded thereon statements and instructions that, when executed by a processor of a device, configure the processor to perform a method including the step or any combination of steps described and / or illustrated herein.

[0136] Embodiment R: A device comprising any combination of the component or components described and / or illustrated herein.

[0137] Many modifications and variations of the present disclosure are possible in light of the above teachings. It should be understood that the embodiments described herein are directed to SSEPs, but are also applicable to other methods of generating evoked potentials, such as visual evoked potentials or auditory brainstem evoked potentials. Within the scope of the appended claims, the present disclosure may be practiced otherwise than as specifically described herein. [Explanation of symbols]

[0138] 100 SSEP System 101 subjects 102 Recording electrode 103 Stimulation electrode 104 Nerve Injury Detection Device (NIDD) 105 Unit integration part 106 Warning and display unit 107 units

Claims

1. 1. A method for determining the presence, absence, and / or monitorability of evoked potentials in one or more SSEP recordings, comprising: obtaining at least one SSEP recording from a subject; determining the presence and characteristics of the evoked potentials in the at least one SSEP recording to determine the presence of a monitorable baseline potential; obtaining an ongoing SSEP recording from the subject to determine ongoing evoked potentials; comparing the ongoing evoked potentials with the monitorable baseline potentials; issuing an alert when the ongoing evoked potential deviates from the monitorable baseline potential according to defined criteria; A method comprising:

2. Obtaining at least one SSEP record includes: obtaining two independent SSEP recordings, the SSEP recordings including a first and a second SSEP recording, and identifying two SSEP evoked potentials, the first evoked potential from the first SSEP recording and the second evoked potential from the second SSEP recording; determining a baseline record based on a grand ensemble average of the first and second SSEP records of the two independent SSEP records; determining whether the baseline recording is monitorable based on characteristics of the two independent SSEP recordings, two evoked potentials identified from the two independent SSEP recording sets, and / or potentials identified from the grand ensemble average; 2. The method of claim 1, comprising:

3. calculating the features of the first evoked potential and the second evoked potential; The features are: the amplitude of the first potential of the two SSEP recordings; the amplitude of the second potential of the two SSEP recordings; and the absolute value of the slope between the dominant vertex and the reference vertex at the baseline potential; the absolute value of the peak latency difference of the potentials in the first and second SSEP recordings; the SNR (signal to noise ratio) of the first and second potentials compared to each entire recording; the ratio of the peak amplitude of the potential to the root mean square (RMS) of the entire recording for the two SSEP recordings and the grand ensemble baseline recording; Including, 3. The method of claim 2.

4. Calculating the features comprises: Identifying candidate vertices in upright and inverted views of the baseline SSEP recording; Identifying which candidate vertex is most prominent based on how prominent the candidate vertex is by its intrinsic height and its position relative to other candidate vertices, and assigning the most prominent candidate vertex as the dominant vertex; calculating peak / valley markers for each baseline SSEP potential by 4. The method of claim 3.

5. identifying onset and offset vertices surrounding the reference vertex; selecting an onset or offset vertex at which the amplitude of the potential is maximum as the reference vertex of the baseline potential; further comprising:

5. The method of claim 4.

6. determining whether the baseline recording is considered monitorable using a machine learning classification algorithm that classifies the baseline SSEP potentials as either monitorable or non-monitoable based on the characteristics of the two SSEP-evoked potentials.

3. The method of claim 2.

7. determining whether the baseline SSEP potentials are considered monitorable using a wavelet convolutional neural network that classifies the baseline potentials as either monitorable or non-monitoable without designing features of the two SSEP potentials.

3. The method of claim 2.

8. determining a confidence value for whether the baseline potential is monitorable; 7. The method of claim 6.

9. adapting the sizes of the two corresponding independent SSEP datasets in response to the confidence values.

9. The method of claim 8.

10. identifying candidate vertices in an upright or inverted display of the ongoing SSEP recording according to the polarity of the baseline SSEP potentials; comparing the amplitude of said potential in a previous ongoing SSEP recording to a threshold; selecting the candidate vertex based on either the vertex having the closest latency to the latency of the potential in the previous ongoing SSEP potential, or selecting the candidate vertex that is most prominent based on how distinctive the candidate vertex is by its intrinsic height and its location relative to other candidate vertices in response to a comparison of the amplitude to the previous ongoing SSEP potential; calculating for each ongoing SSEP recording peak / valley markers of the ongoing SSEP recording by the selected vertex is used in the comparison of the ongoing SSEP potential with the monitorable baseline potential.

2. The method of claim 1 .

11. The defined criteria include a defined decrease in amplitude based on reduced elevation and / or a defined increase in latency based on peak delay.

11. The method of claim 10.

12. The warning may include an audible warning, a visual warning, and / or a tactile warning.

2. The method of claim 1 .

13. Identifying artifacts in the ongoing SSEP recordings and potentials due to the presence of anesthesia in the subject and / or noise from the surrounding environment; Compensating for said artifacts to reduce unnecessary alarms; further comprising:

2. The method of claim 1 .

14. 10. A non-transitory computer readable medium having statements and instructions recorded thereon that, when executed by a processor of a SSEP (Somatosensory Evoked Potential) system, configures the processor to perform the method of claim 1.

15. a stimulating electrode configured to generate an electrical response from the subject's nervous system; a recording electrode configured to detect electrical potentials generated when a stimulus traverses the nervous system; the recording electrode; and a nerve injury detection device coupled to the recording electrode and configured to perform the method of claim 1.

1. An SSEP (Somatosensory Evoked Potential) system comprising:

16. The nerve damage detection device includes: a processor; a non-transitory computer-readable medium having statements and instructions recorded thereon that, when executed by the processor of the SSEP system, configure the processor to implement the nerve injury detection device; 16. The SSEP system of claim 15, comprising:

17. determining whether the baseline recording is deemed monitorable by identifying the dominant and basal peaks of the two SSEP potentials; 3. The method of claim 2.

18. 1. A method for identifying one or more peaks of interest of evoked potentials in an ongoing SSEP recording, comprising: Identifying, in an ongoing SSEP recording, one or more candidate vertices of the same polarity as the dominant vertex of the evoked potential in a predetermined baseline SSEP recording; filtering the one or more candidate vertices based on an analysis scope; comparing the amplitude of the evoked potential in a previous ongoing SSEP recording to a threshold; applying apex tracking by selecting as the dominant apex the candidate apex closest in latency to the dominant apex of the potential from the previous ongoing SSEP recording when the amplitude exceeds the threshold; selecting the most elevated vertex within the ongoing analysis range as the dominant vertex; identifying a potential reference vertex within a region around the primary vertex; selecting a criterion that maximizes the amplitude of the ongoing SSEP-evoked potential; A method comprising:

Images