Automatic calibration of a 3D orientation sensor in an implant

Abstract

A method (300) for calibrating a three-dimensional sensor (100) implantable into a patient's body (200) comprises obtaining (310) a plurality of three-dimensional datasets (320) of an external field with a predetermined field direction acquired by the sensor (100), selecting (330) a first subset (340) of the plurality of three-dimensional datasets (320) based at least in part on a first statistical signature of the three-dimensional datasets (320), and determining (360) a first orientation (370) of the sensor (100) relative to the predetermined field direction, based at least in part on the first subset (340) of the plurality of three-dimensional datasets (320).

Description

[0001] Applicant: BIOTRONIK SE & Co. KG

[0002] Date: 14.08.2025

[0003] Our Reference: 23.151P-WO

[0004] Automatic calibration of a 3D orientation sensor in an implant

[0005] The present invention relates to methods, apparatuses, and computer programs for calibrating a three-dimensional sensor implantable into a patient’s body.

[0006] Three-dimensional (orientation) sensors are widely used to determine the orientation of patient’s bodies equipped with said sensor. E.g., active electronic implantable medical devices may often comprise such sensor to determine the patient’s body position. However, it is often difficult to align the sensor with the body it is mounted to. Light inaccuracies of aligning the sensor and the body will falsify all orientations determined by said sensor unless one accounts for the orientation of the sensor relative to the body.

[0007] Especially in medical applications, e.g., comprising implanting the sensor into the patient’s body, it may even be impossible to align the sensor with the patient’s body without risking the patient’s health. Thus, the orientation of the sensor in the patient after mounting the sensor to the body, e.g., by implantation, must be known. Typically, this is done via postmounting calibration. This may, e.g., in the example of implantable medical devices be based on X-ray images or information on the exact spatial orientation of the implant in the body after implantation. These, however, are usually not available or can only be obtained at great expense. In the example of X-ray, there is even a health risk associated with said calibration. Even if they are available, the information on the orientation of the implant is usually inaccurate.

[0008] Hence, there is a need to improve the calibration of three-dimensional orientation sensors (hereinafter also referred to as three-dimensional sensors or simply sensors) mounted to bodies.

[0009] The aspects described herein address the above need at least in part. A first aspect of the invention relates to a method for calibrating a three-dimensional sensor implantable into a patient’s body, in particular without human intervention. The method comprises obtaining a plurality of three-dimensional datasets of an external field with a predetermined field direction acquired by the sensor, selecting a first subset of the plurality of three-dimensional datasets based at least in part on a first statistical signature of the three- dimensional datasets, and determining a first orientation of the sensor relative to the predetermined field direction, based at least in part on the first subset of the plurality of three-dimensional datasets. A statistical signature is an identified cluster, an activity score and / or any other information relevant for statistical data or analyses.

[0010] In general, the plurality of three-dimensional datasets of an external field with a predetermined field direction may be acquired by the sensor without any knowledge about the current patient state / orientation. Thus, the method works in a highly comfortable way for both, the patient and the attending health professional(s). E.g., the sensor may be implanted into the patient’s body without the need for a subsequent calibration session executed by the patient and the attending health professional(s) which may be time-consuming and costly.

[0011] The inventors found out that selecting a suitable subset of said plurality of datasets allows to reliably determine a first orientation of the sensor based thereon, e.g. an orientation relative to the earth’s gravitational and / or magnetic field. In detail, they found out that the three- dimensional datasets comprise statistical signatures that allow to draw conclusions on the respective patient orientation associated with various subsets of the plurality of three- dimensional datasets. In detail, it can be assumed that the proportion of time in which the patient is standing or sitting upright makes up the largest part of their day. Therefore, when three-dimensional datasets are acquired by the sensor repeatedly over the day, all the datasets during which the patient is in the upright posture will share an essentially common direction, e.g., when the three-dimensional datasets comprise a vector (e.g., as described herein). These three-dimensional datasets may be seen as the first subset that may be considered to determine the first orientation of the sensor relative to the predetermined field direction, based thereon. All other datasets that do not share said common orientation, e.g., because the patient is lying down may form other subset(s). E.g., the three-dimensional datasets may further differ for the patient standing and the patient sitting as typically one is in a slightly

[0012] 23.151P-WO / 14.08.2025 more reclined position while sitting compared to standing. Thus, sitting and standing may result in different subsets of the three-dimensional datasets, e.g., the first subset described herein may be associated with the patient standing (upright). Thereby, the understanding of the inventors allows to categorize the conventionally acquired three-dimensional datasets such as to select the first (and optionally further) subset(s) live or in retrospective.

[0013] For example, the method could be executed as follows:

[0014] Obtaining the plurality of three-dimensional datasets may be done throughout the day during the patient’s daily life. No specific tasks are required from the patient. Assuming ca. 6-10 hours of sleep, it may be expected that the patient spends these 6-10 hours in a reclined, e.g., essentially horizontal, posture (e.g., when lying in bed). The other 14-18 hours the patient may be in an essentially vertical posture (e.g., standing / walking / . . .).

[0015] The selecting of the first subset of the plurality of three-dimensional datasets may be based at least in part on the first statistical signature (as described herein) of the three-dimensional datasets, which may allow to distinguish those datasets associated with an essentially vertical posture from those associated with a reclined posture. Thus, the three-dimensional datasets from the first subset associated with an essentially vertical posture, when, e.g., expressed as a plurality of vectors, these vectors may cluster such as they all may point in an essentially similar direction.

[0016] Thus, the first orientation may be determined by calculating the average vector (e.g., by lavg=Si ^i,i / N) of the said plurality of vectors (e.g., expressed by dl i(i = 1, 2, . . ., N)) of the first subset. There may be outliers that may be disregarded in the determining the first orientation, e.g., by conventional statistical methods.

[0017] Based on the first orientationl avg, at any other point in time, the angle a between the orientation at that point in time atmay be determined, e.g., by a =

[0018] Additionally or alternatively, determining the first orientation could be done, for example, by clustering according to space sectors. The cluster with the highest number of

[0019] 23.151P-WO / 14.08.2025 measurements may then correspond to the “upright sector” the average vector of the vectors present in said “upright sector” may then be considered the first orientation.

[0020] Thereby, the method may be executed to calibrate the sensor implantable into the patient’s body without any face-to-face calibration session involving both, the patient and the attending health professional(s). In result, the method described herein, may reduce the required time and costs for calibration, improve patient and health professional satisfaction and comfort.

[0021] Obtaining the plurality of three-dimensional datasets of an external field with a predetermined field direction acquired by the sensor may comprise receiving the plurality of three-dimensional datasets, e.g., provided by a respective sensor and / or acquiring / measuring the plurality of three-dimensional datasets by the sensor.

[0022] A three-dimensional dataset may comprise one or more of three-dimensional data, e.g., a three-dimensional vector with each coordinate corresponding to a spatial coordinate, preferably orthogonal spatial coordinates. Herein, three-dimensional may be understood as exactly three-dimensional or at least three-dimensional. E.g., the datasets may be fourdimensional, wherein the fourth dimension may be time, e.g., expressed through a timestamp for the respective dataset. The three-dimensional dataset may, e.g., comprise values from a single measurement by the sensor and / or values from a plurality of (e.g., consecutive) measurements of the sensor which may reduce the effects of potential non-representative outliers, and / or a mean value thereof.

[0023] The method may, e.g., further comprise selecting a second subset of the plurality of three- dimensional datasets based at least in part on a second statistical signature of the three- dimensional datasets and determining a second orientation of the sensor relative to the predetermined field direction, based at least in part on the second subset of the plurality of three-dimensional datasets.

[0024] By determining the second orientation, two body axes of the patient’s body, e.g., an upward axis and an axis in the direction of the left or right side of the patient’s body may be determined. Thereby, also the third body axis (which may be defined as being orthogonal to

[0025] 23.151P-WO / 14.08.2025 the two aforementioned ones) of the patient’s body may be determined mathematically without the need for any further measurements and / or datasets.

[0026] The first and second statistical signatures may be identical or different.

[0027] In some examples, the first subset may be associated with an essentially vertical patient posture and / or the second subset may be associated with a reclined patient posture. Typically, while standing, patients may be in an upright or slightly forward-bent posture (e.g., by 0° to 30°, wherein the upright posture corresponds to 0°). On the other hand, while sitting, patients may typically be in a reclined posture (e.g., by 10° to 30° relative to the upright posture). When lying, depending on the bed and / or the placement of cushions, patients may typically be in an essentially horizontal position (e.g., 70° to 90° relative to the upright posture). In order to avoid incorrect calibration in pronounced prone sleepers (around 10% of the population), the determining the second orientation in a sitting position may be a better choice than measurements with the patient in the essentially horizontal position. The vectors used may be, e.g., those that deviate from the upright posture by between 5° (avoidance of numerical instability) and 45° (safe exclusion of prone positions), for example. Due to the assumed posture distributions over time, however, the three-dimensional datasets representing reclined postures may occur more frequently and the averaged results may be correct.

[0028] The inventor found out that a combination of an essentially vertical patient posture and a reclined patient posture is suitable for the calibration of the sensor of the patient for various reasons: Firstly, it can be ensured that between the positions, the patient is rotated about at least one axis. Secondly, in daily life, said rotation may be on average about the so-called right-arm-axis of the patient’s body, i.e., the axis pointing to the right direction from the patient’s body.

[0029] In one example, the second orientation may be determined as follows: E.g., after determining the first orientation as described herein, the second subset may be associated with the patient sitting. A suitable second statistical signature to identify the patient sitting may, e.g., relate to a reduced activity of the patient when sitting, e.g., expressed through a reduced noise pattern of the three-dimensional datasets compared to situations in which the patient is

[0030] 23.151P-WO / 14.08.2025 walking, running, etc. and / or relate to a minimum and / or maximum angle between the first orientation and the three-dimensional datasets of the second subset, e.g., expressed as vectors. Typically, the patient may be in a slightly more reclined posture when sitting compared to standing upright. Thus, the three-dimensional datasets from the second subset associated with the reclined (sitting) posture, when, e.g., expressed as a plurality of vectors, these vectors may cluster such as they all may point in an essentially similar direction different from that described herein for the first subset. The selection may, e.g., be based at least partly on the second statistical signature comprising e.g., a criterion to select only such three-dimensional datasets that are rotated relative to the first orientation by 10° or more, preferably 20° or more, to ensure that the patient is indeed in a reclined position and not, e.g., bending sidewards, and preferably less than 45° or less than 30°. Thus, the second orientation may be determined, e.g., analogously to the first orientation, by calculating the average vector (e.g., by2 avg= Si a2,i / / V) of the said plurality of vectors (e.g., expressed by2>i (i = 1, 2, . . ., N)) of the second subset. There may be outliers that may be disregarded in the determining the first orientation, e.g., by conventional statistical methods.

[0031] In another example, the concept described herein for the sitting posture may, e.g., be performed for an essentially horizontal, e.g. lying, posture. The inventors found that for the lying posture, wherein the patient may typically turn multiple times during the night, may benefit from further measures to improve the reliability of the method, e.g., as described herein.

[0032] The method may, for example, further comprise determining an orthogonal orientation based at least in part on the components of the second subset of the plurality of three-dimensional datasets (orthogonal) to the second orientation.

[0033] Determining said orthogonal orientation (e.g., orthogonal to the first orientation) may ensure that the first and second orientation, which may, e.g., be based on the orthogonal orientation, are orthogonal to one another. The inventors found that using orthogonal orientations may increase the reliability of the method as the projection onto the orthogonal plane may cancel out changes in sleeping height (e.g., depending on how many cushions are used by the patient, e.g., under their back as well). In such cases, the patient may turn not exclusively

[0034] 23.151P-WO / 14.08.2025 around their body axis. However, the described embodiment of the method may work independently form such complications.

[0035] Therein, the orthogonal orientation may, e.g., be determined as follows: The three- dimensional data of the second subset may be projected onto a plane perpendicular to the first orientation. Based on the distribution of the projected vectors, a orthogonal vector may be determined, which then (in the example of an acceleration sensor) points to the patient's back.

[0036] Repeating the determination of the "upright" vector at weekly or monthly intervals would allow the identification of a rotation of the implant since implantation (e.g. fiddler, relevant dislocation) in comparison to the initial upright vector and could also maintain the accuracy of the position determination over time by replacing the old "upright" vector.

[0037] In some examples, the second orientation comprises a cross product of the first orientation with the orthogonal orientation.

[0038] In the above example, a so-called right-arm-vector (or alternatively a so-called left-armvector) can then be calculated from the cross product of the first vector with the orthogonal vector, so that from then on, a distinction may also be made between, for example, supine, right-sided and left-sided positions. If the projected three-dimensional data of one night are not sufficient, the three-dimensional datasets of several nights may also be used together.

[0039] In some examples, the selecting of the first and / or second subset may additionally or alternatively be based at least partly on clustering the plurality of three-dimensional datasets, wherein clustering may preferably be determined at least in part based on a standard deviation and / or a mean value.

[0040] The clustering may, e.g., be assessed as described herein. In some examples, e.g., when the patient is in an upright / standing position, this may result in a plurality of three-dimensional datasets / vectors that cluster by pointing in essentially the same direction and / or having essentially the same amplitude. The statistical signature may quantize said clustering, e.g., by a mean value of the respective subset which is accompanied by a relatively low mean

[0041] 23.151P-WO / 14.08.2025 deviation (e.g. lower than an absolute threshold and / or lower than a threshold related to the mean value). The clustering may in particular pertain to a plurality of datasets that are temporally subsequent to each other. In some examples, a mean value for a pair-wise distance between the endpoints of next neighbors of the plurality of three-dimensional datasets / vectors may be used to identify a clustering, which may allow to identify all or most datasets / vectors of one cluster. For example, for vectors to be determined to be part of the same cluster, the pair-wise distance between any two vectors of the cluster may be required to be below a certain threshold.

[0042] For example, at least some of the plurality of three-dimensional datasets may be associated with a specific time of day. Datasets of the first subset and / or of the second subset may be selected at least in part based on a predetermined time window. For example, the first subset may be selected, based on the first statistical signature, not necessarily from all available datasets. Instead, for example, a preselection may be carried out, and, e.g. the first subset may be selected from datasets associated with a predetermined time window, e.g. a time of the day at which the patient is expected to be active. Preferably, a first time window may be associated with an expected essentially vertical patient posture (and thus, e.g., with the first subset) and / or a second time window may be associated with a reclined, e.g., an essentially horizontal, patient posture (and thus, e.g., with the second subset).

[0043] Assigning such time windows may provide a first pre-selection that, e.g., determines that only datasets from the first time window may be selected for the first subset and / or only datasets from the second time window may be selected for the second subset.

[0044] The calibration can be improved by considering the time of day (and / or the activity level), e.g., as follows: Based on the time of day, the measurements can be assigned to the "activity phase", e.g., from 8 am to 8 pm and the "rest phase" from 0 am to 6 am, for example.

[0045] This may increase the reliability of the method and reduce the number of incorrectly selected three-dimensional datasets for the respective subset. Adjusting the time windows to the individual patient could further increase accuracy.

[0046] 23.151P-WO / 14.08.2025 In some examples, the first and / or second statistical signature may comprise an activity score based on three-dimensional data of the sensor, wherein the activity score may preferably be based at least partly on a noise pattern of the three-dimensional datasets.

[0047] Additionally or alternatively, the activity score may be based on three-dimensional data of a further sensor. E.g., the further sensor may provide further three-dimensional datasets analogously to the three-dimensional datasets of the sensor described herein.

[0048] The invention described herein relies at least partly on the in-depth understanding on how different activities translate into different patient postures, which may be exploited for calibrating the three-dimensional sensor. Therefore, basing the method at least partly on an activity score may further improve the accuracy and reliability of the method, especially of the selecting of the first and / or the second subset described herein.

[0049] The activity score can, e.g., be determined by the sensor and / or by a further, e.g., wearable, sensor. Phases of activity and phases of rest may be identified from this data and the three- dimensional datasets may be assigned accordingly. If multiple statistical signatures (e.g., an identified cluster and an activity score) are used together, a three-dimensional dataset may only then be selected for a respective subset when said three-dimensional dataset fulfills the criteria of the at least two statistical signatures: E.g., those that do not fit into the sequence based on the activity score may not be selected for the respective subset even if they belong to the same cluster. This could, e.g., allow to distinguish two subsets of the three-dimensional dataset, one of which was acquired when the patient is sitting upright (e.g., low activity score) and one when the patient is walking in an upright posture (e.g., high activity score). The same may apply to the time window, as described herein: E.g., those that do not fit into the sequence based on the activity score may not be selected for the respective subset even if they belong to the “correct” time window. This could be the case for an exemplary three- dimensional dataset acquired when the patient wakes up and stands up at night.

[0050] In some examples, at least one, preferably each, three-dimensional dataset may comprise at least one three-dimensional vector.

[0051] 23.151P-WO / 14.08.2025 Thereby, the three-dimensional data(sets) and the orientation(s) may be compatible with one another facilitating the interplay of acquiring and / or obtaining the three-dimensional data and determining the orientation(s), e.g., based thereon.

[0052] In some examples, the three-dimensional sensor may comprise a three-axis acceleration sensor and / or the three-dimensional sensor may comprise a three-axis magnetic field sensor. Preferably, the three axes of the orientation sensor may be orthogonal.

[0053] The external field may, e.g., be the gravitational field of the earth and / or the geomagnetic field. The gravitational field (or gravitational acceleration field) and / or the geomagnetic field (or the Earth’s magnetic field) are vector fields with a predetermined field direction and a predetermined field strength. These fields may advantageously be utilized in the context of the invention as the fields are present and strong enough to provide reliable data.

[0054] The sensor may, e.g., measure the strength of the respective field along three axes and provide the as-acquired measures as three-dimensional data, e.g., in the form of a three- dimensional vector. When these three axes are orthogonal, the measures are linearly independent which is a common choice to facilitate data processing etc.

[0055] Accordingly, the three-dimensional data, e.g., in the form of vectors, may comprise three entries each comprising the measures of the respective field along the three axes of the sensor, e.g., the gravitational acceleration values along three orthogonal axes of an exemplary (gravitational) acceleration sensor as described herein.

[0056] For example, the sensor may be comprised in an implantable medical device, wherein preferably the implantable medical device may comprise a spinal cord stimulator and / or a pacemaker.

[0057] Especially in this case, the method described herein is particularly advantageous, as for implantable devices in general, surgeons typically cannot simply rotate the implantable device, e.g., comprising a sensor, such as to align it with the patient’s coordinate system. That could potentially bear substantial risks for the patient’s health. Therefore, a simple, yet accurate calibration may allow to track both, the patient’s orientation / posture throughout the

[0058] 23.151P-WO / 14.08.2025 day as well as the final orientation of the sensor within the patient’s body. This may, e.g., allow to track if the sensor / the implant is migrating or not.

[0059] In this example, the sensor being mounted to a body thus corresponds to the sensor being implanted into the patient. Thereby, the sensor and the (patient’s) body (at least at and near the implantation site) rotate together but not with respect to one another.

[0060] In some examples, the first orientation and / or the second orientation of the sensor may comprise a statistical average of the first and / or the second subset of the plurality of first three-dimensional datasets.

[0061] The statistical average may comprise, e.g., one or more of the following: Mean: an arithmetic mean (calculated by summing up all the values in a dataset and dividing by the total number of values), an actual median (the middle value in a dataset when the values are arranged in ascending or descending order. If there is an even number of values, the median is the average of the two middle values.), a mode (the value that appears most frequently in a dataset), a range (the difference between the maximum and minimum values in a dataset), and / or a so-called robust average (an average that is less sensitive to outliers or extreme values in the dataset).

[0062] For example, the obtaining the plurality of three-dimensional datasets may comprise repeatedly obtaining a three-dimensional dataset, preferably at a predetermined frequency and / or at predetermined times.

[0063] Thereby, a sufficient data base may be obtained for accurate calibration and further, the timing may be adjusted such that it becomes very likely that all required patient postures are captured by the three-dimensional datasets.

[0064] The three-dimensional sensor may, e.g., be configured to acquire three-dimensional data repeatedly and / or upon receiving an according signal comprising an instruction to acquire said three-dimensional data.

[0065] 23.151P-WO / 14.08.2025 In some examples, the orientations determined as described herein are stored, such that subsequently obtained three-dimensional datasets can be re-cast in the corresponding coordinate system, without further calibration steps.

[0066] A second aspect of the present invention relates to an apparatus for calibrating a three- dimensional sensor implantable into a patient’s body, in particular without human intervention. The apparatus comprises means for obtaining a plurality of three-dimensional datasets of an external field with a predetermined field direction acquired by the sensor, means for selecting a first subset of the plurality of three-dimensional datasets based at least in part on a first statistical signature of the three-dimensional datasets, and means for determining a first orientation of the sensor relative to the predetermined field direction, based at least in part on the first subset of the plurality of three-dimensional datasets.

[0067] The apparatus may be integrated into the sensor and / or an implant as described herein, and / or be a remote apparatus, i.e., remote from the patient and / or the sensor, e.g., a remote server, a mobile device of the patient like a mobile phone, etc. The means of the apparatus may be distributed across various devices, e.g., in different locations.

[0068] A third aspect of the present invention relates to a computer program comprising instructions for performing of a method as described herein, when the instructions are executed.

[0069] Said apparatus and / or computer program may yield the advantages described herein in reference to the method.

[0070] Any step described herein in reference to the method may be implemented as means and / or functionalities of the apparatus and / or as instructions of the respective computer program, and vice versa.

[0071] Fig. 1 shows an exemplary orientation sensor.

[0072] Fig. 2a shows a patient’s body with an implanted sensor with the patient’s body in a first orientation.

[0073] 23.151P-WO / 14.08.2025 Fig. 2b shows a patient’s body with an implanted sensor with the patient’s body in a second orientation rotated by an angle a around the z-axis relative to the first orientation shown in Fig. 2a.

[0074] Fig. 3a shows a flow chart for an exemplary method for calibrating a three-dimensional sensor implantable into a patient’s body.

[0075] Fig. 3b shows a schematic representation of a patient’s body with an implanted three- dimensional sensor in an upright posture.

[0076] Fig. 3c shows three schematic representations of a patient’s body with an implanted three-dimensional sensor in an essentially horizontal posture.

[0077] Fig. 3d shows an exemplary plurality of three-dimensional datasets represented as vectors throughout a time period wherein the patient was standing, sitting, and lying down.

[0078] Fig. 4 shows an 3d shows an exemplary plurality of three-dimensional datasets represented as vector components along the three axes and of the sensor over time throughout a time period wherein the patient was standing, sitting, and lying down.

[0079] Fig. 1 shows an exemplary orientation sensor 100 with three axes u, v, w. In the exemplary three-dimensional sensor 100 of Fig. 1, the three axes u, v, w are orthogonal such that each axes acquires the field strength of the external field along the direction of the respective axis. In the example of Fig. 1, the schematically shown springs allow the sensor 100 to sense the acceleration separately along the three axes u, v, vv and / or to provide the corresponding three-dimensional data to a further apparatus, e.g., a remote server (not shown).

[0080] The sensor 100 may, e.g., be an acceleration sensor measuring the gravitational field strength as described herein: Schematically, the sensor may act as if each of said springs are anchored to a common mount at one and to a mass m on the other end. According to Hooke’s law, the spring force F, the mass m is held at an equilibrium position against the gravitational force

[0081] 23.151P-WO / 14.08.2025 or the gravitational acceleration g. Thus, the changes in the length of the spring may be proportional to the projections of the gravitational force on the respective axes u, v, vv of the sensor 100: Ass= — * uT* —g. In total, this may yield:

[0082] = — k * (Ass* u + As * v + As^ * w)

[0083] The change in length of the springs As^, As^, As^ may be measured by the sensor 100 and the three-dimensional data may comprise the values a^, a , a^, e.g.: * Ass= uT*

[0084] —a = — cos(cr) * |a|, wherein a may be the angle between a and u. The same may apply analogously to v and w.

[0085] Figs. 2a and 2b show exemplary embodiments, wherein the sensor 100 is an implantable sensor 100 to be implanted into a patient’s body 200. However, the concept described in reference to said Figures may apply to any other body, e.g., as described herein, analogously.

[0086] Fig. 2a shows a patient’s body 200 with an implanted orientation sensor 100 with the patient’s body 200 in a first orientation. Said first orientation corresponds to the patient standing upright in the example of Fig. 2a. The sensor 100 may, e.g., be a sensor 100 according to the exemplary embodiment of Fig. 1. The axes u, v, w of the sensor 100 may be rotated relative to the patient axes x,y, z, which, e.g., correspond to the direction of the head x, the patent’s right arm y, and the patient’s nose z. In the example of Fig. 2a, the sensor 100 (with its orientation expressed through the axes u, v, w) is not aligned with the body (with its orientation expressed through the axes x,y, z). It may, however, be aligned with them in other examples.

[0087] Fig. 2b shows a patient’s body with an implanted orientation sensor 100 with the patient’s body 200 in a second orientation rotated by an angle a around the z-axis relative to the first orientation shown in Fig. 2a. In comparison of Figs. 2a and 2b, one can see that obviously a rotation of the patient’s body 200 brings along an equal rotation of the sensor 100, as they are fixedly connected to one another. Therefore, a change of orientation of the sensor 100 allows to be translated into a change of orientation of the body 200.

[0088] 23.151P-WO / 14.08.2025 Fig. 3a shows a flow chart for an exemplary method 300 for calibrating a three-dimensional sensor implantable into a patient’s body, in particular without human intervention. The exemplary method comprises the following steps:

[0089] - obtaining 310 a plurality of three-dimensional datasets 320 of an external field with a predetermined field direction acquired by the sensor,

[0090] - selecting 330 a first subset 340 of the plurality of three-dimensional datasets based at least in part based on a first statistical signature of the three-dimensional datasets and optionally a second subset 350 of the plurality of three-dimensional datasets based at least in part on a second statistical signature of the three-dimensional datasets,

[0091] - determining 360 a first orientation 370 of the sensor relative to the predetermined field direction, based at least in part on the first subset 340, and

[0092] - optionally determining 380 a second orientation 390 of the sensor relative to the predetermined field direction, based at least in part on the second subset 350.

[0093] Fig. 3b shows a schematic representation of a patient’s body with an implanted three- dimensional sensor in an upright posture.

[0094] Fig. 3c shows three schematic representations of a patient’s body with an implanted three- dimensional sensor in an essentially horizontal posture. In detail, the left panel shows the patient lying on their left side, the central panel shows the patient lying on their back and the right panel shows the patient lying on their right side. This illustrates a typical range of motion of a patient during a night.

[0095] Fig. 3d shows an exemplary plurality of three-dimensional datasets 320 represented as vectors throughout a time period wherein the patient was standing and lying down.

[0096] The vectors of the first subset 340, corresponding to the patient standing, all point in essentially the same first direction. The second subset 350 represent the three-dimensional data acquired when the patient was lying down, e.g., sleeping and turning from either side to the other. Therefore, the respective vectors accumulate on a half-circle. For example, the first subset 340 can be identified based on the exemplary statistical signatures as described herein, e.g. low standard deviation around their mean value and / or low next-neighbor

[0097] 23.151P-WO / 14.08.2025 distance. The second subset 350 may for example be identified by another statistical signature, e.g. a relatively large standard deviation around their mean value and / or large next-neighbor distance and or by selecting datasets which are approximately located in a plane normal to the first orientation.

[0098] Fig. 4 shows an exemplary plurality of three-dimensional datasets represented as vector components along the three axes and of the sensor over time throughout a time period wherein the patient was standing, sitting, and lying down.

[0099] The first panel shows the acceleration aualong the u-axis, the second panel shows the acceleration avalong the v-axis, and the third panel shows the acceleration awalong the w- axis. The fourth panel shows the total acceleration at which averages at around the value of 9.81 m / s2(earth’s gravitational field strength).

[0100] In a first time period 401 the patient is sitting on a chair which results in the accelerations along the three axes of the sensor staying at constant levels 401u, 401v, 401w. This period is characterized by low noise and low standard deviation of the datasets.

[0101] In a second time period 402 the patient stands up from their chair which results in increased noise of the data and peaks, especially in the w-direction indicating the posture change.

[0102] A third time period 403 splits into three sub-periods, a first sub-period 403a in which the patient is walking, a second sub-period 403b, in which the patient is running, and a third subperiod 403c, in which the patient is walking again. The three sub-periods 403a, 403b, 403c vary in noise (running creates more noise than just walking) but share the same average acceleration 403u, 403 v, 403 w, respectively throughout the three sub-periods 403a, 403b, 403c because the patient’s body was essentially in the same posture for walking and running. From the raw data, for example, short term averages may be calculated to remove the noise, and then, datasets with a relatively low standard deviation around a mean value are obtained, e.g. similar to the datasets 340 shown in Fig. 3d (those from period 401 may also be similar, albeit slightly inclined due to the slight inclination of the bode when sitting).

[0103] 23.151P-WO / 14.08.2025 In a fourth time period 404 the patient lies down which results in increased noise of the data and peaks, especially in the u- and w-direction indicating the posture change.

[0104] A fifth time period 405 splits into three sub-periods, a first sub-period 405a in which the patient lies on their back, a second sub-period 405b, in which the patient turns from one side to the other, and a third sub-period 405c, in which the patient is lying on their back again. The three sub-periods 405a, 405b, 405c do not substantially vary in noise (lying on either side creates the same noise pattern) and the sub-phases 405a, 405c wherein the patient lies on their back share the same average acceleration 405u, 405v, 405w, respectively throughout the two sub-periods 405a, 405c. In the sub-phase 405b, however, substantial variations in the accelerations along the three axes are recorded, especially along the u- and w-axes. The v-axis in this example is almost aligned with the axis around which the patient turned, therefore it records subtle variations upon the patient turning.

[0105] These data illustrate that, in some examples, the statistical signature, the selecting the first and / or the second subset and / or the determining of the method as described herein, may e.g., be further based on detecting peaks in accelerations, e.g., on a second-timescale. These may, e.g., as shown in Fig. 4, indicate posture changes.

[0106] 23.151P-WO / 14.08.2025

Claims

Claims1. A method (300) for calibrating a three-dimensional orientation sensor (100) implantable into a patient’s body (200), the method (300) comprising: obtaining (310) a plurality of three-dimensional datasets (320) of an external field with a predetermined field direction acquired by the sensor (100); selecting (330) a first subset (340) of the plurality of three-dimensional datasets (320) based at least in part on a first statistical signature of the three-dimensional datasets (320); and determining (360) a first orientation (370) of the sensor (100) relative to the predetermined field direction, based at least in part on the first subset (340) of the plurality of three-dimensional datasets (320).

2. The method (300) of claim 1, further comprising: selecting (330) a second subset (350) of the plurality of three-dimensional datasets (320) based at least in part on a second statistical signature of the three-dimensional datasets (320); and determining (380) a second orientation (390) of the sensor (100) relative to the predetermined field direction, based at least in part on the second subset (350) of the plurality of three-dimensional datasets (320).

3. The method (300) of claim 2, wherein the first subset (340) is associated with an essentially vertical patient posture and / or the second subset (350) is associated with a reclined patient posture.

4. The method (300) of claim 2 or 3, further comprising determining an orthogonal orientation based at least in part on the components of the second subset (350) of the plurality of three-dimensional datasets (320) to the second orientation (390).

5. The method (300) of claim 4, wherein the second orientation (390) comprises a cross product of the first orientation (370) with the orthogonal orientation.23.151P-WO / 14.08.20256. The method (300) of any of claims 1 to 5, wherein the selecting (330) of the first and / or the second subset (340, 350) is based at least partly on clustering the plurality of three- dimensional datasets (320); wherein clustering is preferably determined at least in part based on a standard deviation and / or a mean value.

7. The method (300) of any of claims 1 to 6, wherein at least some of the plurality of three-dimensional data sets are associated with a specific time of day; and datasets of the first subset (340) and / or the second subset (350) are selected at least in part based on a predetermined time window; wherein preferably a first time window is associated with an essentially vertical patient body (200) posture and / or a second time window is associated with an essentially horizontal patient body (200) posture.

8. The method (300) of any of claims 1 to 7, wherein the first and / or second statistical signature comprises an activity score determined based on the three-dimensional datasets (320), wherein the activity score preferably is based at least partly on a noise pattern of the three-dimensional datasets (320).

9. The method (300) of any of claims 1 to 8, wherein at least one of the three-dimensional datasets, preferably each of the three-dimensional datasets, comprises at least one three-dimensional vector.

10. The method (300) of any of claims 1 to 9, wherein the sensor (100) comprises a three- axis acceleration sensor (100) and / or wherein the sensor (100) comprises a three-axis magnetic field sensor (100); and wherein preferably the three axes of the sensor (100) are orthogonal.

11. The method (300) of any of claims 1 to 10, wherein the sensor (100) is comprised in an implantable medical device, wherein preferably the implantable medical device comprises a spinal cord stimulator and / or a pacemaker.

12. The method (300) of any of claims 1 to 11, wherein determining (360, 380) the first and / or the second orientation (370, 390) of the sensor (100) comprises determining a23.151P-WO / 14.08.2025statistical average of the first and / or the second subset (340, 350) of the plurality of first three-dimensional datasets (320).

13. The method (300) of any of claims 1 to 12, wherein the obtaining (310) the plurality of three-dimensional datasets (320) comprises repeatedly obtaining (310) a three- dimensional dataset, preferably at a predetermined frequency and / or at predetermined times.

14. An apparatus for calibrating a three-dimensional orientation sensor (100), the three- dimensional sensor (100) implantable into a patient’s body (200), the apparatus comprising: means for obtaining (310) a plurality of three-dimensional datasets (320) of an external field with a predetermined field direction acquired by the sensor (100); means for selecting (330) a first subset (340) of the plurality of three-dimensional datasets (320) based at least in part on a first statistical signature of the three- dimensional datasets (320); and means for determining (360) a first orientation (370) of the sensor (100) relative to the predetermined field direction, based at least in part on the first subset (340) of the plurality of three-dimensional datasets (320).

15. A computer program comprising instructions for performing of a method (300) according to any of claims 1 to 13, when the instructions are executed on the apparatus of claim 14.23.151P-WO / 14.08.2025

Images